# OneLessTool — full content (data version 1c3df44d0d3f)
Evidence statuses: verified, editorial, estimate, illustrative, ai_suggested, unverified. Structured access: https://onelesstool.com/api/v1 and MCP at https://onelesstool.com/api/mcp.
---
# Workflows
# Build and host a small business website with AI coding tools instead of Webflow, Squarespace or Wix
> For a simple 3–8 page marketing site: write a site brief, have Claude Code or Cursor (or a chat assistant, for non-coders) build plain HTML/CSS, and host it on Cloudflare's free plan with Workers static assets. Includes the agent prompt, deployment commands, a form option, redirect handling and an honest list of what you give up.
- Outcome: I want a fast, professional website for my business without paying a monthly website-builder subscription.
- Replaces: Webflow (full), Squarespace (full), Wix (full)
- Uses: Claude — Claude Code (included in Claude's paid plans) builds and edits the site files from the brief; Claude chat works for non-coders who copy files by hand.; Cursor (optional) — Alternative AI code editor if you already pay for it.; ChatGPT (optional) — Alternative chat assistant for generating pages and copy.; Cloudflare Workers & Pages — Free hosting for the static files (Workers static assets), custom domain, HTTPS and redirects.; Google Workspace (optional) — Google Forms + Sheets for the simplest contact-form option.
- Extra monthly cost: $0 — Cloudflare's Free plan serves static asset requests free and without a request limit; Worker code (only needed for a custom form handler) gets 100,000 requests a day on Free. Assumes you already pay for an AI coding tool. Your domain renewal is unchanged — but if your domain was bundled with the website builder, you will need to pay for it at a registrar after moving.
- Difficulty: moderate; skill: technical; setup: 1–2 days for a 5-page site including content migration, review and DNS changes
- Evidence: editorial; illustrative example: true; last reviewed 2026-10-13
- Canonical: https://onelesstool.com/workflows/business-website-with-ai-coding
## Good fit for
- Service businesses, consultants, clinics and trades with a brochure-style site that changes a few times a month
- Founders comfortable running a few terminal commands, or with a technical friend for the first deploy
- Sites where speed, low cost and full ownership of the files matter more than a visual editor
## Not a fit for
- Teams where non-technical marketers publish pages or blog posts weekly and expect a visual editor
- Ecommerce stores, booking systems, member areas or anything with payments and logins
- Sites with large blogs or CMS collections that non-developers update
- Owners who need phone or chat support from their website provider
## Before
Pay a monthly or annual website-builder plan. Edit pages in the visual editor, use its built-in contact form, and rely on its hosting, SSL, templates and support.
## After
The site is a folder of HTML/CSS files you own. An AI coding tool builds and edits it from a written brief; `npx wrangler deploy` publishes it to Cloudflare's free plan on your own domain. Forms go to a Google Form or a small Worker. Edits are made by asking the AI and redeploying.
## Steps
### 1. Inventory the current site before touching anything
List every page, its URL, page title and meta description, every form and where its submissions go, and every integration (booking widget, newsletter signup, chat, analytics, store).
Copy all text into a Google Doc and download all images you own the rights to. Check your builder's own export options; exports are often partial, so a manual copy is the safe baseline.
Note who controls the **domain** (registrar login) and the **DNS**. If the domain was bought through the website builder, plan to transfer it or point it elsewhere before cancelling.
### 2. Write the site brief
Fill in the **site brief template** (artifact below) and save it as `SITE_BRIEF.md` in a new, empty project folder. Include the old URL for each page so redirects can be created.
The brief is what keeps the AI from inventing content. Anything you leave blank becomes a placeholder you'll fill later.
### 3. Have the AI build the site
**Coders / comfortable with a terminal:** open the folder in Claude Code (included in Claude's paid plans) or Cursor and paste the **Agent build prompt**. Review the home page in the browser before letting it build the rest. Run it locally with `npx wrangler dev` or by opening the HTML files.
**Non-coders:** use Claude or ChatGPT chat with the **Single-page generator prompt**, one page at a time. Save each answer as a file with the exact name the AI gives (e.g. `index.html`, `styles.css`) in a `public` folder. Ask for a fixed list of files up front so nothing is missed.
Keep the site **plain HTML + CSS** unless you have a reason not to; it's easiest for any AI or developer to edit later.
### 4. Review it like a visitor
Use the **QA prompt** to have the AI check for broken links, missing alt text and titles. Then check yourself:
- On your phone: menu works, text readable, buttons tappable, no sideways scrolling.
- Every claim, price and testimonial is real and approved.
- Contact details match your Google Business Profile and other listings exactly.
### 5. Deploy to Cloudflare's free plan
Cloudflare's docs now say to start new projects on **Workers** (Pages still works). For a static site:
1. Create a free Cloudflare account.
2. In the project folder, make sure `wrangler.jsonc` contains your project name, a `compatibility_date` and `"assets": { "directory": "./public" }`.
3. Run `npx wrangler deploy`. The first run asks you to log in. You get a `*.workers.dev` URL to test.
4. **Custom domain:** a Workers Custom Domain needs the domain added to Cloudflare as an active zone (free plan is fine), which usually means switching your domain's nameservers to Cloudflare. Recreate any email DNS records (MX, SPF, DKIM) **before** switching, or your email will break.
5. Add the domain under the Worker's **Settings > Domains & Routes**.
Free plan limits that matter (checked today): requests for static assets are free and unlimited; up to 20,000 files per version and 25 MiB per file; Worker code (only if you add a form handler) has 100,000 requests a day.
### 6. Handle forms and redirects
**Forms, simplest option:** create a Google Form (responses go to a Sheet, with email notifications) and link to it or embed it on the contact page. Looks less integrated but needs no code.
**Forms, integrated option:** ask the AI to add a small Worker route that receives the form POST, checks a Cloudflare Turnstile token to block spam, and forwards the message to you (for example to a Google Sheet via Apps Script, or by email). This uses Worker requests, which count toward the 100,000/day free limit.
**Redirects:** a `_redirects` file in `public/` maps old builder URLs to new ones (e.g. `/about-us /about 301`). Cloudflare allows up to 2,000 static and 100 dynamic redirects. This protects your search rankings.
### 7. Run both sites in parallel, then cancel
Keep the old builder subscription until the new site has been live on your domain for at least a week and the verification checklist passes. Submit the new `sitemap.xml` in Google Search Console and watch for crawl errors.
Save the project folder somewhere backed up (a private Git repo is ideal). That folder *is* your website now.
## Prompts
### Agent build prompt (Claude Code / Cursor)
Have an AI coding agent build the whole site from the brief, with checkpoints.
```text
You are building a static marketing website for {{business_name}}. Read SITE_BRIEF.md in this folder first; it is the source of truth.
Requirements
- Pages: {{page_list}}. One HTML file per page in ./public (e.g. public/index.html, public/services/index.html) so URLs have no .html extension.
- Plain HTML + one shared CSS file (public/styles.css). No framework and no build step unless I ask. Minimal JavaScript (mobile menu only).
- Shared header/nav and footer must be identical on every page. Footer: {{footer_contents}}.
- Design: colours {{brand_colors}}, fonts {{fonts}}, tone {{design_tone}}. Mobile-first, readable at 375px wide, no horizontal scroll.
- Accessibility: semantic headings (one h1 per page), alt text on every image, visible focus states, colour contrast at least WCAG AA.
- SEO: unique and meta description per page (use the ones in the brief), Open Graph tags, a sitemap.xml and robots.txt in public/, canonical URLs on {{domain}}.
- Contact form: {{form_approach}}.
- Redirects: create public/_redirects with 301s from each old URL in the brief to its new path.
- Images: put them in public/images, sized for the web (max ~1600px wide, compressed), with width/height attributes.
- Hosting: Cloudflare Workers static assets. Create wrangler.jsonc with name "{{worker_name}}", today's compatibility_date and "assets": {"directory": "./public"}. No Worker script unless the form approach needs one.
Process
1. Propose the file structure and a one-paragraph design direction. Wait for my OK.
2. Build the home page first. Stop so I can review it in the browser.
3. Build the remaining pages, then run a self-check: broken internal links, missing alt text, missing titles, pages wider than the viewport at 375px. Report what you found and fixed.
Do not invent testimonials, client names, prices, awards or statistics. Use clearly marked placeholders where content is missing.
```
### Single-page generator prompt (chat, for non-coders)
Generate one page at a time in a chat assistant and save it as a file.
```text
I'm building a simple static website for {{business_name}} with these pages: {{page_list}}. All pages share one stylesheet called styles.css.
Here is my site brief:
{{site_brief}}
Now write the complete file for the "{{page_name}}" page, saved as {{file_path}}. Use plain HTML, link to /styles.css, include the shared header and footer exactly as in previous pages: {{shared_header_footer_or_none}}.
Use only content from the brief; mark anything missing as [PLACEHOLDER: ...]. Give me the whole file in one code block and nothing else.
```
### QA prompt
Have the AI audit the finished site before launch.
```text
Review every file in ./public for launch readiness on {{domain}}. Check and fix: broken internal links; images without alt text or width/height; pages missing a unique title or meta description; more than one h1; layout overflow at 375px; contrast below WCAG AA for {{brand_colors}}; any invented testimonials, numbers or client names; _redirects entries for these old URLs: {{old_urls}}.
Report a table of issues found, what you changed, and anything that needs my decision.
```
### Content edit prompt (after launch)
Make routine changes safely and redeploy.
```text
Change request for the {{business_name}} site: {{change_description}}.
Only edit the files needed. Keep header/footer identical across pages. Show me a summary of the diff, then run `npx wrangler deploy` once I confirm.
```
## Templates and scripts
### Site brief template (template)
```markdown
# Site brief – {{business_name}}
_Fill this in before you prompt the AI. The more specific, the less rework._
## 1. Basics
- Business name: {{business_name}}
- One-sentence description: {{one_liner}}
- Primary audience: {{audience}}
- Main action we want visitors to take: {{primary_cta}} (e.g. book a call, request a quote, call us)
- Domain: {{domain}} — registrar: ______ — DNS currently managed by: ______
- Current site URL (to migrate from): {{current_site_url}}
## 2. Pages (3–8)
| Page | URL path | Purpose | Key content | Old URL to redirect from |
|---|---|---|---|---|
| Home | / | Explain what we do, prove it, CTA | Hero, 3 services, testimonials, CTA | / |
| Services | /services | Detail each service | 3 sections + FAQ | /what-we-do |
| About | /about | Trust | Story, team photos, credentials | /about-us |
| Contact | /contact | Convert | Form, phone, address, map link, hours | /contact |
| (optional) Work / Case studies | /work | Proof | 3 short case studies | |
| Privacy | /privacy | Legal | Your privacy policy text | |
## 3. Brand
- Colours (hex): primary ____ accent ____ text ____ background ____
- Fonts: ____ (or "system fonts")
- Logo files: SVG/PNG paths
- Sites we like the look of (and why): ____
- Tone of voice: ____
## 4. Content inventory
- Copy: where it lives now (export from old site, Google Doc link)
- Images: folder link; all images we have rights to use? yes/no
- Testimonials (with permission): ____
- Legal: privacy policy, cookie notice needed? (depends on analytics/forms used and your location)
## 5. Forms and integrations
- Contact form fields: name, email, phone (optional), message, how did you hear
- Where submissions go: Google Form + Sheet / email via a small Worker / other
- Booking link (e.g. existing scheduler): ____
- Analytics: none / privacy-friendly analytics / Google Analytics
- Anything else from the old builder we rely on: blog, store, member area, newsletter signup, chat widget
## 6. SEO essentials
- Page titles and meta descriptions: keep the current ones? (copy them from the old site)
- Business address and phone shown consistently on every page footer
- Old URLs that get traffic (from your analytics) → redirect targets
## 7. Who maintains it
- Who will edit text after launch: ____
- How often: ____
- Their comfort level: edits in a code editor / asks the AI / needs a visual editor
```
### Minimal wrangler.jsonc for a static site (config)
```jsonc
{
"$schema": "./node_modules/wrangler/config-schema.json",
"name": "{{worker_name}}",
// Set to the date you create the project
"compatibility_date": "{{compatibility_date}}",
"assets": {
"directory": "./public"
}
}
```
## Limitations
- No visual drag-and-drop editor: changes go through an AI tool or code, which suits founders more than marketing teams.
- No built-in CMS or blog management for non-technical editors.
- No built-in forms, ecommerce, bookings, memberships or email marketing; each needs a separate tool or custom code.
- No vendor support desk; when something breaks you rely on the AI tool, docs, or a developer.
- Domain and DNS changes are on you; mistakes can take a site or email offline for hours.
- Builder templates, apps and some content may not export cleanly; expect manual copying.
## Risks
- **migration_complexity** (medium): Moving DNS can break email if MX/SPF/DKIM records aren't recreated first. Check every record before switching nameservers.
- **functionality_loss** (medium): Builder-native features (forms, store, scheduling, newsletter popups) disappear on cancel; each must be inventoried and replaced.
- **security** (low): Static HTML has a small attack surface, but a custom form handler must validate input and use spam protection such as Turnstile.
- **reliability** (low): AI-generated code can contain subtle layout or accessibility bugs; the QA step and a mobile check are essential.
- **vendor_lock_in** (low): Plain HTML files are portable to any host, which reduces lock-in compared with a builder.
## Verification checklist (run before cancelling)
- [ ] Every page from your inventory exists on the new site and old URLs return a 301 to the right new page (test each in a private window).
- [ ] The `*.workers.dev` preview works on a phone at 375px wide with no horizontal scrolling and a working menu.
- [ ] A test contact-form submission arrives where you expect (Sheet, inbox) within minutes, and a blank or spammy submission is rejected.
- [ ] Email still sends and receives on your domain after DNS changes (send a test both ways).
- [ ] Page titles and meta descriptions match or improve on the old site's; sitemap.xml loads and is submitted in Search Console.
- [ ] You (or whoever maintains the site) made a text change and redeployed it unaided in under 15 minutes.
- [ ] The project folder is backed up outside your laptop.
- [ ] The old builder stays active until the new site has run on your domain for at least a week without issues.
## Savings methodology
Savings = your website-builder plan (monthly, or annual price ÷ 12) − incremental cost ($0 hosting on Cloudflare's Free plan; $0 for the AI tool if you already pay for it). If the builder bundled your domain, subtract the domain's annual registrar price ÷ 12. If you'd need to buy an AI coding subscription only for this, subtract that too. Setup time (1–2 days) and later editing time are not monetised.
Example (illustrative): A 5-page consultancy site on a website-builder plan billed annually, owner already pays for Claude Pro and has the domain at a separate registrar. $25/mo → $0/mo. Assumptions: Illustrative: $25/month is a round placeholder for a mid-tier builder plan. Webflow, Squarespace and Wix pricing pages showed non-US currencies or JavaScript-rendered prices to our fetcher on 2026-10-13, so no builder price was verified. Cloudflare Free plan limits (static asset requests free and unlimited; 20,000 files/version; 25 MiB per file) were verified on developers.cloudflare.com on 2026-10-13. Claude Pro is already paid for other work; Claude's pricing page lists Claude Code as included in all paid plans. The domain renewal is unchanged and not counted.
## Sources
- Cloudflare Workers static assets: billing and limitations: https://developers.cloudflare.com/workers/static-assets/billing-and-limitations/ (accessed 2026-10-13)
- Cloudflare Workers limits: https://developers.cloudflare.com/workers/platform/limits/ (accessed 2026-10-13)
- Cloudflare Pages limits: https://developers.cloudflare.com/pages/platform/limits/ (accessed 2026-10-13)
- Cloudflare Pages docs: https://developers.cloudflare.com/pages/ (accessed 2026-10-13)
- Cloudflare Workers static assets: https://developers.cloudflare.com/workers/static-assets/ (accessed 2026-10-13)
- Cloudflare Workers static assets: get started: https://developers.cloudflare.com/workers/static-assets/get-started/ (accessed 2026-10-13)
- Cloudflare Workers static assets: redirects: https://developers.cloudflare.com/workers/static-assets/redirects/ (accessed 2026-10-13)
- Cloudflare Workers Custom Domains: https://developers.cloudflare.com/workers/configuration/routing/custom-domains/ (accessed 2026-10-13)
- Claude pricing: https://claude.com/pricing (accessed 2026-10-13)
- Squarespace pricing: https://www.squarespace.com/pricing (accessed 2026-10-13)
- Wix plans: https://www.wix.com/plans (accessed 2026-10-13)
# Write client proposals with AI and Google Docs instead of PandaDoc
> Turn discovery-call notes into a scoped, priced proposal with a four-step AI prompt sequence, lay it out in a reusable Google Docs template, and collect signatures with Google Docs eSignature (if your Workspace plan includes it) or a DocuSign account you already have. Replaces PandaDoc's writing and template work; not its analytics, CRM sync or payments.
- Outcome: I want to send professional, accurate proposals quickly and get them signed without paying for PandaDoc.
- Replaces: PandaDoc (partial)
- Uses: Claude — Turn discovery notes into scope, pricing table and risks; review the draft like a sceptical client.; ChatGPT (optional) — Alternative assistant for the same prompt sequence.; Google Workspace — Google Docs proposal template, Drive folder per client, and (on eligible plans, when the admin enables it) Docs eSignature.; DocuSign eSignature (optional) — Signature option if your Google Workspace plan lacks eSignature and you already pay for DocuSign.
- Extra monthly cost: $0 — Assumes you already pay for an AI assistant and a Google Workspace plan. Google Docs eSignature is listed for Workspace Individual, Business Standard and Plus, and several Enterprise/Education editions when the admin turns it on; it is not listed for Business Starter. If you don't have it and don't already pay for DocuSign or similar, you'll need a separate e-signature tool, which is an extra cost.
- Difficulty: easy; skill: basic; setup: 2–3 hours to build the template, pricing rate card and saved prompts
- Evidence: editorial; illustrative example: false; last reviewed 2026-10-13
- Canonical: https://onelesstool.com/workflows/client-proposals-with-ai-docs
## Good fit for
- Freelancers, consultancies and agencies sending a handful of custom proposals a month
- Teams already on Google Workspace Business Standard or higher (eSignature included when enabled)
- Service businesses whose proposals are mostly custom scope rather than catalogue quotes
## Not a fit for
- Sales teams that rely on PandaDoc's CRM sync to create documents from deals and update deal stages automatically
- Businesses that collect payment inside the proposal at signing
- Teams that need view tracking (who opened what, for how long) to time follow-ups
- Organisations with formal approval chains and a shared, governed content library
- Regulated agreements where your legal team requires a specific e-signature provider
## Before
Create a PandaDoc document from a template or CRM deal, drag in content-library blocks, build the pricing table, route it for internal approval, send it, watch view analytics, and collect signature and payment in PandaDoc.
## After
Paste discovery notes into Claude and run four prompts (brief → scope → pricing → risks), paste the result into a copy of your Google Docs proposal template, check every number against your rate card, export or share, and request signatures with Docs eSignature (or DocuSign). Follow-ups are scheduled in your calendar instead of triggered by view alerts.
## Steps
### 1. Check what you'll lose and what you'll use instead
List the PandaDoc features you use weekly. PandaDoc's own plan comparison puts **CRM integrations, content library, document analytics and approval workflows on Business** (not Starter), and **payment collection** on both Starter and Business. If any of those drive your sales process, this workflow is only a partial replacement.
Then confirm your signature route:
- **Google Docs eSignature**: listed for Workspace Individual, Business Standard/Plus and some Enterprise/Education editions, **only when your admin enables it**. Up to 10 signers and 200 fields per request; an audit-trail page is added to the signed PDF. Not listed for Business Starter.
- **DocuSign** or another e-signature tool you already pay for.
- If you have neither, budget for one before cancelling PandaDoc.
### 2. Build your Google Docs template and rate card
Paste the **proposal template** (artifact below) into a new Google Doc: turn on Markdown (Tools > Preferences), then right-click > **Paste from Markdown**. Add your logo, set heading styles and brand colours, and save it in a Drive folder named `Templates`.
Keep a **rate card** in a Google Sheet: service, unit, price, minimum, notes. The AI will only use prices from this sheet. This is the most important control in the workflow.
Move your reusable blocks (case studies, team bios, terms) into one `Proposal library` Doc. It's a manual content library: copy, don't link.
### 3. Run the prompt sequence on your discovery notes
In one conversation (or a Claude Project / ChatGPT Project holding your rate card, library and template):
1. **Discovery → brief**: structure the messy notes and list unknowns.
2. **Brief → scope**: phases, deliverables, acceptance criteria, out of scope.
3. **Scope → pricing table**: line items priced *only* from your rate card, with the arithmetic shown.
4. **Risks and assumptions**, then the **executive summary** last.
Answer the unknowns the AI lists before moving on. Don't paste confidential client data your agreement or your AI plan's data settings don't allow.
### 4. Assemble the document and check every number
Make a copy of the template in the client's Drive folder (`Proposals/<Client>/<YYYY-MM-DD> <Project>`). Paste each section in using Paste from Markdown.
Then verify by hand:
- Each line item's price matches the rate card; recalculate totals in the Sheet, not in your head.
- Dates are achievable and dependencies are stated.
- Names, company details and the validity date are correct.
Finally, run the **Sceptical client review prompt** and fix what's fair.
### 5. Send and collect signatures
**Google Docs eSignature** (if enabled): in the Doc, go to **Tools > eSignature**, choose each signer under "Insert fields for", drag in Signature and Date signed fields, then click **Request signature**, add signer emails and optionally turn on automatic reminders. When everyone has signed, the final PDF (with its audit trail page) is saved in the same Drive folder as the original Doc, and linked in the completion email.
**DocuSign**: download the Doc as PDF (**File > Download > PDF**), upload it to DocuSign, place fields and send.
Either way, also email the client a short cover note with the summary and a deadline for questions.
### 6. Replace the missing tracking and payment steps
- **Follow-ups**: without view analytics, add a calendar reminder 2 and 5 business days after sending. Use the **Follow-up email prompt** with the proposal summary.
- **CRM**: paste the proposal link and value into the deal record by hand (or with an automation tool you already use).
- **Deposits**: send the deposit invoice from your accounting/invoicing tool when the signed PDF arrives.
## Prompts
### 1. Discovery notes → structured brief
Turn raw call notes into a clean brief and surface what you still need to ask.
```text
You are helping me write a proposal for {{client_company}}. My company: {{your_company}}, which does {{your_services}}.
Here are my raw discovery notes:
{{discovery_notes}}
Produce: 1) the client's situation in 3 sentences; 2) their goals, each with a measurable success indicator; 3) constraints (budget hints, deadlines, decision-makers, approvals); 4) a numbered list of open questions I must answer before scoping. Don't fill gaps with assumptions; list them as questions.
```
### 2. Brief → scope of work
Define phases, deliverables, acceptance criteria and exclusions.
```text
Using the brief above and my answers to the open questions ({{answers_to_open_questions}}), write the scope of work:
- {{phase_count}} phases, each with duration estimate, deliverables (with format), and an acceptance criterion per deliverable.
- An "Out of scope" list of at least 5 items the client might reasonably assume are included.
- Dependencies on the client (inputs, approvals, access) with deadlines.
Use plain language. Match how we usually deliver: {{delivery_approach}}.
```
### 3. Scope → pricing table (rate card only)
Price the scope strictly from your own rate card, with arithmetic shown.
```text
Price the scope above using ONLY this rate card. Never invent a price or discount:
{{rate_card}}
Output a Markdown table: item | qty | unit | unit price | line total. Then subtotal, tax at {{tax_rate}}, and total in {{currency}}.
Show your quantity reasoning for each line in one short sentence underneath. If something in scope has no rate-card entry, put "PRICE NEEDED" instead of a number.
Add a separate "Optional add-ons" table for: {{optional_items}}.
Payment terms to state: {{payment_terms}}.
```
### 4. Risks, assumptions and executive summary
Protect the project with explicit assumptions, then write the summary last.
```text
For this proposal:
1) List 5–8 assumptions that, if wrong, would change cost or timeline (inputs, feedback rounds, access, third-party tools).
2) A risk table: risk | likelihood | impact | mitigation. Include at least one risk on the client side and one on ours.
3) Then write a 4–5 sentence executive summary for {{decision_maker_role}}: their situation, outcome, our approach, total investment of {{total_price}}, and the next step ({{next_step}}). No superlatives, no claims we can't prove.
```
### Sceptical client review prompt
Stress-test the draft before sending.
```text
Act as {{decision_maker_role}} at {{client_company}} reading this proposal for the first time, with a budget of about {{client_budget_hint}}. Be sceptical.
List: anything unclear; anything that looks padded; questions you'd ask before signing; places where scope and price don't match; and the single biggest reason you might say no. Then suggest the smallest edits that fix each point.
Proposal:
{{proposal_text}}
```
### Follow-up email prompt
Replace view-triggered follow-ups with a scheduled, helpful nudge.
```text
Write a short follow-up email (under 90 words) to {{client_name}} about the {{project_name}} proposal I sent on {{sent_date}}. Mention one specific benefit from the summary: {{key_benefit}}. Offer a 15-minute call to answer questions and note the proposal is valid until {{valid_until}}. Friendly, no pressure, no "just checking in".
```
## Templates and scripts
### Proposal template (Markdown — paste into Google Docs with Paste from Markdown) (template)
```markdown
# Proposal: {{project_name}}
**Prepared for:** {{client_name}}, {{client_company}}
**Prepared by:** {{your_name}}, {{your_company}}
**Date:** {{date}} · **Valid until:** {{valid_until}} · **Reference:** {{proposal_id}}
---
## 1. Summary
_Three to five sentences: their situation, the outcome they want, what you'll do, the investment, and the next step._
## 2. Your goals (as we understood them)
- Goal 1 — how we'll know it's achieved: ____
- Goal 2 — ____
- Constraints we heard: budget, deadline, internal capacity, approvals
## 3. Scope of work
### Phase 1 – {{phase_1_name}} ({{phase_1_duration}})
- Deliverable: ____ — format: ____ — acceptance: ____
### Phase 2 – {{phase_2_name}} ({{phase_2_duration}})
- Deliverable: ____
### Out of scope
- ____ (anything the client might reasonably assume is included)
## 4. Timeline
| Milestone | Target date | Depends on |
|---|---|---|
| Kick-off | ____ | Signed proposal + deposit |
| Phase 1 complete | ____ | Client feedback within 3 business days |
| Final delivery | ____ | ____ |
## 5. Investment
| Item | Qty | Unit price | Total |
|---|---|---|---|
| ____ | 1 | ____ | ____ |
| **Subtotal** | | | ____ |
| Tax ({{tax_rate}}) | | | ____ |
| **Total** | | | **____** |
**Optional add-ons** (not included in total)
| Option | Price | Notes |
|---|---|---|
| ____ | ____ | ____ |
**Payment terms:** {{payment_terms}} (e.g. 50% on signature, 50% on delivery; net 14). Invoices sent from {{invoicing_tool}}.
## 6. Assumptions
- Client provides ____ by ____.
- One round of consolidated feedback per deliverable; extra rounds billed at {{hourly_rate}}.
## 7. Risks and how we'll handle them
| Risk | Likelihood | Mitigation |
|---|---|---|
| ____ | Low/Med/High | ____ |
## 8. Why us
- Relevant result 1 (real, approved for reference)
- Relevant result 2
## 9. Terms
{{terms_reference}} (link to or attach your standard terms; have a lawyer review them once.)
## 10. Acceptance
By signing, {{client_company}} accepts this proposal and the terms referenced above.
| | {{client_company}} | {{your_company}} |
|---|---|---|
| Name | | |
| Title | | |
| Signature | | |
| Date | | |
```
## Limitations
- No document view analytics (opens, time per page); follow-ups are scheduled by hand.
- No automatic CRM sync of documents, values or deal stages.
- No payment collection at signing; deposits are invoiced separately.
- No governed content library or approval workflow; reusable blocks live in a Doc and approvals are manual.
- Pricing tables in Docs are static; there are no client-selectable optional items or automatic totals.
- Google Docs eSignature is only available on certain Workspace editions and only when your admin enables it; it allows up to 10 signers and 200 fields per request.
## Risks
- **reliability** (high): AI can miscalculate totals or invent prices. Price only from your rate card, mark gaps 'PRICE NEEDED', and recalculate totals in a spreadsheet.
- **sensitive_data** (medium): Discovery notes may include confidential client information. Check your client agreements and your AI plan's data settings before pasting them in.
- **compliance** (medium): Some contracts or industries require a particular e-signature provider or audit standard. Confirm Google Docs eSignature meets your legal requirements before relying on it.
- **functionality_loss** (medium): Losing CRM sync and view tracking can slow follow-ups for sales-led teams.
- **data_portability** (low): Export signed PandaDoc documents and your templates' text before cancelling; they won't be accessible afterwards.
## Verification checklist (run before cancelling)
- [ ] You rebuilt your last real proposal with this workflow and every price, total and date matches the original (or is knowingly updated).
- [ ] You confirmed in Google Docs that the eSignature option appears on your account (or you have an existing DocuSign/e-signature account), and completed a test signature with a colleague.
- [ ] The signed test PDF includes the audit trail and is saved in the right Drive folder.
- [ ] A full proposal took you under 60 minutes from notes to send.
- [ ] Your team agrees which lost features (view tracking, CRM sync, payments, approvals) you can live without, and has a manual replacement for each.
- [ ] All active and signed PandaDoc documents and template content are exported before cancelling.
## Savings methodology
Savings = PandaDoc seats × price per seat (monthly, or annual ÷ 12) − incremental cost. Incremental cost is $0 when you already pay for an AI assistant and a Google Workspace plan with eSignature, or for an existing DocuSign account. If you must buy a separate e-signature tool, subtract its monthly cost. Setup and per-proposal time are not monetised.
Example: A two-person consultancy on PandaDoc Business (billed annually) that sends 6–8 proposals a month, uses Google Workspace Business Standard and already pays for Claude. $98/mo → $0/mo. Assumptions: PandaDoc Business is $49 per seat per month billed annually ($65 billed monthly) per PandaDoc's plan comparison page, checked 2026-10-13: 2 seats × $49 = $98. Google Workspace Business Standard lists eSignature in Docs and the admin enables it; Workspace is already paid for and not counted. Claude is already paid for other work and not counted. The consultancy doesn't rely on PandaDoc's CRM sync, analytics or in-document payments.
## Sources
- PandaDoc Help: Compare subscription plan features: https://support.pandadoc.com/en/articles/9715033-compare-subscription-plan-features (accessed 2026-10-13)
- Google Docs Help: Request eSignatures: https://support.google.com/docs/answer/12315692 (accessed 2026-10-13)
- Google Workspace pricing: https://workspace.google.com/pricing (accessed 2026-10-13)
- Google Docs Help: Use Markdown in Docs: https://support.google.com/docs/answer/12014036 (accessed 2026-10-13)
# Run your company wiki in Google Workspace or Microsoft 365 instead of paying for Notion or Confluence
> For teams of up to about 15 who already pay for Google Workspace or Microsoft 365: move the wiki into a shared drive of Google Docs with a Google Sites front page (or SharePoint/OneNote on Microsoft 365). Includes an information architecture template, a Notion export-and-import checklist, AI prompts for cleanup, and the honest list of what you lose — mainly databases, Notion AI and slick page linking.
- Outcome: I want one place where my team finds policies, how-tos and onboarding, without a separate Notion or Confluence subscription.
- Replaces: Notion (full), Confluence (full)
- Uses: Google Workspace — Shared drive for wiki docs, Google Docs for pages, Google Sites (listed on all Business plans) as the internal front page, Drive search.; Microsoft 365 (optional) — Alternative stack: a SharePoint team or communication site with pages and a document library, plus a team OneNote notebook for informal notes.; Claude (optional) — Clean up exported pages, rename files, draft the IA and rewrite stale docs during migration.; ChatGPT (optional) — Alternative assistant for the same cleanup prompts.
- Extra monthly cost: $0 — Uses Google Workspace or Microsoft 365 you already pay for. Google says shared drives became available on Business Starter from September 2024 (with fewer admin controls than higher plans), and Google Sites is listed on all Business plans. On Microsoft 365, SharePoint is included in plans that license it (check your admin centre).
- Difficulty: moderate; skill: basic; setup: 1–3 days for a typical small-team wiki (export, restructure, import, relink, front page)
- Evidence: editorial; illustrative example: false; last reviewed 2026-10-13
- Canonical: https://onelesstool.com/workflows/company-wiki-with-existing-suite
## Good fit for
- Teams of up to ~15 using Notion or Confluence mainly as a wiki: policies, how-tos, onboarding, meeting notes
- Companies already on Google Workspace (Business Starter and up) or a Microsoft 365 plan with SharePoint
- Teams that already write most documents in Google Docs or Word and keep copying them into the wiki
## Not a fit for
- Teams that rely on Notion databases as an app — relations, rollups, formulas, boards and views updated daily (CRM, project tracker, content calendar)
- Engineering teams that depend on Confluence's Jira integration and page trees tied to tickets
- Organisations that heavily use Notion AI or Notion Agent over their workspace
- Large orgs needing granular page-level permissions, page analytics and verified-page workflows
## Before
Pay per member for Notion (or Confluence). Pages, nested sub-pages and databases live there; people search Notion, @-link pages, and use Notion AI to summarise. Many documents also exist in Google Docs or Word, so content is split across two places.
## After
All wiki content lives as Google Docs in one shared drive with a simple folder tree and owners. An internal Google Site acts as the front page with search, a 'Start here' link and the top docs. People find things with Drive or Sites search and link docs with @-mentions. The paid wiki subscription is cancelled.
## Steps
### 1. Decide what moves and what doesn't
List every Notion teamspace/top-level page (or Confluence space) and every **database**. Sort databases into two groups:
- **Document lists** (meeting notes, policies, how-tos): these become folders of Docs.
- **Apps** (relations, rollups, formulas, kanban boards updated daily): these don't fit a wiki. Move them to Google Sheets, a tool you already have, or keep a smaller Notion plan for that team. If most of your Notion is apps, stop here; this workflow isn't for you.
Also note integrations and embeds pointing at Notion.
### 2. Design the structure before importing
Use the **information architecture template** (artifact below): one shared drive called `Wiki`, numbered top folders, max two levels deep, an owner and review date on every doc.
Run the **IA draft prompt** with your Notion sidebar (copy the page titles) to get a proposed mapping from old pages to new folders. Edit it with one or two colleagues; agreement here saves weeks of "where is it?" later.
On Google Workspace **Business Starter**, shared drives work but with fewer admin and security controls than higher plans; check your admin console settings.
### 3. Export from Notion (or Confluence)
In Notion, a workspace owner exports the **entire workspace** as **Markdown & CSV** (desktop or web only), including subpages. Export an **HTML** copy too as an archive.
Things Notion's help centre warns about:
- Download links **expire after 7 days**; big exports can take **up to 30 hours**.
- Other members' **private pages are not included**; ask people to export their own.
- Full-page databases export as a **CSV plus one Markdown file per row/page**; only the current or default view is exported.
- Callout blocks come out as HTML.
For Confluence, use its space export options and expect similar cleanup.
### 4. Clean up and import into Google Docs
Work through the **migration checklist** (artifact below). In short:
1. Rename files (Notion adds long IDs) and arrange them in the new folder tree locally.
2. Use the **Page cleanup prompt** on messy pages: fix headings, convert HTML leftovers, flag outdated content.
3. Import: in Google Docs, **File > Open > Upload** a `.md` file to convert it into a Google Doc, or upload to Drive and **Open with > Google Docs**. Google started rolling out native `.md` viewing and editing in Docs in October 2026; convert the docs people edit often so comments and smart chips work as usual.
4. Open each database CSV in Google Sheets.
### 5. Relink and build the front page
Exported internal links point at local `.md` files, not at the new Docs. Fix the **top 20 most-used docs first**, replacing links with links or @-mention chips to the new Google Docs. Fix the rest as people report them.
Create an internal **Google Site** (restricted to your organisation) as the front page: search box, a "Start here" button, the top 8 docs, team sections and an "ask here" line. On Microsoft 365, use a SharePoint communication site home page with quick links and a document library web part instead.
### 6. Microsoft 365 variant
If you're on Microsoft 365 instead of Google:
- Create a **SharePoint team site** called Wiki. Use **pages** for front-door content (Start here, policies people read) and a **document library** with the same folder tree for Word docs.
- Use the site's **team OneNote notebook** for informal notes and meeting notes.
- Import Notion Markdown by converting to Word (ask your AI assistant to convert each `.md` to clean `.docx`-ready text, or open the HTML export in Word) and uploading to the library.
- Microsoft Copilot Chat (included with eligible business plans at no extra cost) can help draft and summarise; grounding answers on your organisation's files depends on licensing.
### 7. Cut over, verify, then cancel
Set a cut-over date. Put a "Moved to …" banner on the Notion home page and make it read-only. Ask three colleagues who weren't involved to find 10 common answers using only the new wiki (see verification checklist).
Cancel or downgrade Notion/Confluence at the **renewal date**, keeping the export zip in a restricted archive shared drive.
## Prompts
### IA draft prompt
Map your existing Notion/Confluence pages onto the new folder structure.
```text
We're a {{team_size}}-person {{company_type}} moving our wiki from {{old_tool}} to {{new_suite}}.
Here is our current page tree (indentation = nesting):
{{page_tree}}
Using this target structure: {{target_folder_tree}}
Produce a table: old page | new folder | new doc title (phrased as people would search) | action (migrate / merge with X / archive / delete) | suggested owner role.
Flag pages that look like databases used as apps, duplicates, and anything that looks out of date.
```
### Page cleanup prompt
Tidy an exported Markdown page before importing it into Google Docs.
```text
Clean up this page exported from {{old_tool}} so it imports well into Google Docs. Rules:
- Keep all facts. Don't add new information.
- One H1 title: "{{new_doc_title}}". Logical H2/H3 headings.
- Convert any HTML (e.g. callouts) into Markdown: callouts become a bold "Note:" paragraph.
- Replace internal links to .md files with [[LINK: page name]] placeholders so I can relink them.
- Add this header table at the top: Owner {{owner}} | Last reviewed {{review_date}} | Review every {{review_interval}} | Status Current.
- At the end, list anything that looks outdated (old dates, people who may have left, retired tools).
Page:
{{page_markdown}}
```
### Stale content review prompt
Run the quarterly review faster.
```text
Here is a list of wiki docs with owner and last-modified date: {{doc_list}}
Today is {{today}}. Flag every doc not modified in {{stale_months}} months, group by owner, and draft a short message to each owner asking them to confirm, update or archive their docs by {{deadline}}.
```
### Find-it test prompt
Generate realistic questions to test whether people can find things in the new wiki.
```text
We're a {{team_size}}-person {{company_type}}. Based on this wiki folder tree: {{folder_tree}}
Write 15 realistic questions employees ask in their first 3 months (e.g. "How do I claim expenses?"). For each, name the doc that should answer it. I'll ask colleagues to find each answer and time them.
```
## Templates and scripts
### Wiki information architecture template (template)
```markdown
# Company wiki – information architecture template ({{company_name}}, ≤15 people)
## Where things live
- **Front page:** Google Site "{{company_name}} Wiki" (internal-only), or a SharePoint communication site / team site home page on Microsoft 365.
- **Content:** one shared drive (Google) or one document library (SharePoint) called `Wiki`. Docs, not pages, hold the content.
- **Rule:** every wiki doc lives in `Wiki`, never in someone's My Drive / OneDrive. Personal drafts move in when they're ready.
## Folder tree (max 2 levels deep)
Wiki/
├── 00 Start here/ → Welcome, How this wiki works, Who to ask for what, Glossary
├── 01 Company/ → Mission & values, Org chart, Holidays & time off, Benefits, Expenses
├── 02 Policies/ → Security & passwords, Data handling, Remote work, Code of conduct
├── 03 How we work/ → Meetings & rituals, Tools we use (and who owns each), Writing guide
├── 04 Teams/
│ ├── Sales/ → Playbook, Pricing rules, Proposal templates
│ ├── Delivery/ → Project kickoff checklist, QA checklist, Client handover
│ └── Ops & Finance/ → Month-end close, Invoicing, Vendor list
├── 05 Onboarding/ → First day, First week, Role checklists
├── 06 Templates/ → Meeting notes, Project brief, Decision record, Post-mortem
├── 07 Decisions/ → YYYY-MM-DD decision records (one doc each)
└── 99 Archive/ → Outdated docs (moved, never deleted, prefix "ARCHIVED – ")
## Naming conventions
- Doc titles are questions or nouns people search for: "How do I submit expenses?", "Expense policy".
- No dates in titles except decision records and meeting notes: `2026-10-13 – Decision – Switch wiki to Google Drive`.
- No "final", "v2", "new" in titles. Version history handles that.
## Every doc starts with this header block
| Owner | Last reviewed | Review every | Status |
|---|---|---|---|
| @name | YYYY-MM-DD | 6 months | Current / Draft / Archived |
## Front page (Google Sites / SharePoint) sections
1. Search box (Sites search, or SharePoint search)
2. "Start here" button → 00 Start here
3. Top 8 most-used docs (embed or link) — check the list quarterly
4. Team sections → link to each 04 Teams subfolder
5. "Recently changed" — a manually updated list or an embedded Drive folder view
6. "Can't find it? Ask in {{chat_channel}}" with the wiki owner's name
## Permissions
- Shared drive / library members: everyone in the company as Contributor (Google) or Member (SharePoint); 2 people as Managers/Owners.
- Sensitive areas (HR cases, salaries) do NOT go in the wiki. Use a separate restricted shared drive or library.
- Guests/contractors: share individual folders, not the whole wiki.
## Ownership & upkeep
- Wiki owner: {{wiki_owner}} — runs a 30-minute review each quarter.
- Each folder has an owner listed in "00 Start here/Who to ask for what".
- Docs not reviewed in 12 months get flagged; owner updates or archives.
```
### Notion → Google Workspace migration checklist (checklist)
```markdown
# Notion → Google Workspace migration checklist
## A. Before exporting (week 1)
- [ ] List every Notion top-level page and teamspace; mark each: migrate / archive only / delete.
- [ ] List every Notion **database**. For each: is it a document list (fine as a folder) or an app (relations, rollups, formulas, boards people update daily)? Apps need a separate plan (Google Sheets, AppSheet, or keep Notion for that team).
- [ ] List integrations, automations, embeds and forms that point at Notion.
- [ ] Note who has private pages they want to keep (exports skip other people's private pages).
- [ ] Freeze big restructures in Notion until the move is done.
## B. Export
- [ ] Workspace owner: Settings → export **entire workspace** (desktop or web only) as **Markdown & CSV**, with subpages. Also export as **HTML** as an archival copy that keeps the look.
- [ ] Download promptly: Notion's export links expire after 7 days, and a large export can take up to 30 hours to arrive.
- [ ] Each member exports their own private pages they want kept.
- [ ] Unzip into a local folder; keep the original zip in a restricted "Notion archive" shared drive.
## C. Prepare files
- [ ] Rename files to drop the long ID suffix Notion adds (ask an AI assistant or use a bulk-rename tool).
- [ ] Map Notion's folder tree to the new IA (see IA template). Move files locally first.
- [ ] Databases: open each CSV in Google Sheets; delete columns nobody uses; keep the CSV as the record.
- [ ] Images and attachments: check they're in the export folder next to each page.
## D. Import into Google Docs
- [ ] Small batches: Google Docs → File → Open → Upload the `.md` file (it converts to a Google Doc). Or upload to Drive, right-click → Open with → Google Docs.
- [ ] Large batches: upload `.md` files to the shared drive and convert the most-used ones first; leave rarely used pages as Markdown files (Drive can preview and Docs can now open .md files without converting, rolling out from October 2026).
- [ ] Spot-check 10 converted docs: headings, tables, code blocks, images, callouts (Notion exports callouts as HTML).
- [ ] Re-insert images that didn't come across.
## E. Fix links
- [ ] Internal links in the export point at local files, not Google Docs. Search each doc for links ending in `.md` and replace them with links (or @-mention smart chips) to the new Google Docs. Start with "Start here" and the top 20 docs.
- [ ] Update links in chat tools, email signatures, bookmarks and other docs that pointed at Notion URLs.
## F. Build the front page
- [ ] Create the internal Google Site; add the sections from the IA template; restrict it to your organisation.
- [ ] Add the shared drive folders and top docs.
## G. Cut over
- [ ] Announce the date; for 2 weeks, add a banner at the top of the Notion home page: "Moved to <link>. Read-only from <date>."
- [ ] Run the verification checklist with 3 colleagues who weren't involved.
- [ ] Downgrade or cancel Notion at the renewal date; keep the export archive.
```
## Limitations
- No Notion-style databases: relations, rollups, formulas and multiple views don't exist in Docs; Sheets covers only simple tables.
- No Notion AI or Notion Agent over the whole wiki. Gemini in Docs is listed on Workspace Business Standard and Plus, not Starter; Copilot's work-data grounding depends on Microsoft licensing.
- Page linking is clunkier: no backlinks or nested sub-pages inside a page; you rely on folders, links and @-mention chips.
- Exported internal links break and must be relinked by hand.
- Google Sites is a simple front page; it isn't a full wiki engine with page trees and inline editing of every doc.
- On Business Starter, shared drives lack some admin and security controls available on higher plans.
## Risks
- **migration_complexity** (medium): Large or deeply nested workspaces take longer than expected; broken links and missing images are common. Migrate the top-used content first.
- **data_portability** (medium): Exports skip other members' private pages and only export one database view. Have each person export what they need and keep the original zip archived.
- **functionality_loss** (high): Teams using Notion databases as apps lose them. Identify these before cancelling and give them a new home.
- **collaboration** (low): People may keep docs in personal drives. The 'everything lives in the Wiki shared drive' rule and a named owner prevent sprawl.
- **security** (low): Check sharing settings on the shared drive and Site so internal content isn't shared publicly or with the wrong guests.
## Verification checklist (run before cancelling)
- [ ] Every Notion database has been classified, and each one used as an app has a working new home (or a deliberate decision to keep a small Notion plan).
- [ ] The full workspace export (Markdown & CSV plus HTML archive) is downloaded and stored in a restricted archive location.
- [ ] The top 20 most-used docs are in the shared drive with working internal links and images.
- [ ] Three colleagues not involved in the migration found answers to 10 'find-it test' questions in under 2 minutes each using Drive or Sites search.
- [ ] The Google Site (or SharePoint home page) is restricted to your organisation, and guests can only see folders shared with them.
- [ ] Every top-level folder has a named owner and a review date.
- [ ] Notion has been read-only for at least two weeks with no requests to switch back before you cancel at renewal.
## Savings methodology
Savings = paid wiki seats × price per seat per month (annual ÷ 12 if billed annually) − incremental cost ($0, since Google Workspace or Microsoft 365 is already paid). If you keep a smaller Notion plan for a team using databases as apps, subtract those seats. Migration time (1–3 days) and quarterly upkeep are not monetised.
Example: A 10-person agency on Notion Plus (billed annually) used purely as a wiki, already paying for Google Workspace Business Standard. $100/mo → $0/mo. Assumptions: Notion Plus is $10 per member per month on yearly billing, per notion.com/pricing on 2026-10-13 (the page's default billing toggle is yearly; monthly billing costs more). 10 members × $10 = $100. Google Workspace is already paid for; shared drives and Google Sites are available on its Business plans. No Notion databases are used as apps, so no Notion seats are kept.
## Sources
- Notion pricing: https://www.notion.com/pricing (accessed 2026-10-13)
- Notion Help: Export your content: https://www.notion.com/help/export-your-content (accessed 2026-10-13)
- Google Workspace Updates: Shared drives for Business Starter: https://workspaceupdates.googleblog.com/2024/08/shared-drive-access-business-starter.html (accessed 2026-10-13)
- Google Workspace Updates: Markdown files natively in Drive and Docs: https://workspaceupdates.googleblog.com/2026/10/preview-edit-and-collaborate-on-Markdown-files-natively-across-Drive-and-Docs.html (accessed 2026-10-13)
- Google Docs Help: Use Markdown in Docs: https://support.google.com/docs/answer/12014036 (accessed 2026-10-13)
- Google Workspace pricing: https://workspace.google.com/pricing (accessed 2026-10-13)
- Microsoft Learn: SharePoint service description: https://learn.microsoft.com/en-us/office365/servicedescriptions/sharepoint-online-service-description/sharepoint-online-service-description (accessed 2026-10-13)
- Microsoft Learn: Manage Microsoft Copilot Chat: https://learn.microsoft.com/en-us/copilot/manage (accessed 2026-10-13)
# Consolidate overlapping AI assistant subscriptions (ChatGPT, Claude, Gemini, Perplexity, Grok, Copilot)
> If you pay for two to four AI assistants, run a one-week, side-by-side test on your real tasks with a provided prompt set and scoring sheet, check which assistant your Google Workspace or Microsoft 365 plan already includes, and cancel the rest at their renewal dates. Some heavy users will rightly keep two; the test tells you which.
- Outcome: I pay for several AI chat subscriptions and want to keep only the one(s) I actually need.
- Replaces: ChatGPT (full), Claude (full), Google Gemini (full), Perplexity (full), Grok (full), Microsoft 365 Copilot (full)
- Uses: Google Workspace (optional) — Google Sheets for the scoring sheet. If your organisation uses Workspace, the Gemini app and Gemini in Gmail are listed on Business plans (Gemini in Docs on Standard and Plus).; Microsoft 365 (optional) — Excel for the scoring sheet. Eligible Microsoft 365 business and enterprise plans include Microsoft Copilot Chat at no extra cost when you sign in with a work account.; Claude (optional) — One of the candidates you test and may keep.; ChatGPT (optional) — One of the candidates you test and may keep.
- Extra monthly cost: $0 — No new tools are bought. You keep one (sometimes two) of the subscriptions you already have, plus whatever assistant your existing Google Workspace or Microsoft 365 plan includes.
- Difficulty: easy; skill: none; setup: 1 hour to set up, plus about 20–30 minutes a day of testing for one week
- Evidence: editorial; illustrative example: false; last reviewed 2026-10-13
- Canonical: https://onelesstool.com/workflows/consolidate-ai-assistants
## Good fit for
- Individuals paying for two or more AI chat subscriptions out of habit or FOMO
- Small teams where people expense different assistants with no shared standard
- Companies on Google Workspace or Microsoft 365 that also pay for separate AI chat seats
## Not a fit for
- Developers who rely on a specific coding agent tied to one subscription and use another assistant for a distinct job — keeping two can be rational
- Teams with a compliance requirement to use one approved vendor (the decision is made by policy, not a taste test)
- People who use one assistant lightly; just keep the free tier of the others
## Before
Pay for several assistants (for example ChatGPT Plus, Claude Pro and Perplexity Pro), switch between them by habit, and forget that your work suite already includes Gemini or Copilot Chat.
## After
One primary paid assistant chosen by evidence from your own tasks, the assistant bundled with your work suite used for email and documents, free tiers for occasional second opinions, and cancelled subscriptions timed to their renewal dates.
## Steps
### 1. List every AI subscription, its price and renewal date
Check card statements, App Store / Google Play subscriptions and company expense reports. For each assistant record: **plan, monthly cost, monthly or annual billing, renewal date, who uses it**.
Annual plans usually can't be refunded mid-term, so the renewal date decides *when* you save, not *whether*.
### 2. Check what your work suite already includes
- **Google Workspace**: Google's plan page lists Gemini in Gmail and the Gemini app on Business Starter, Standard and Plus (basic access on Starter, expanded on Standard and Plus), with **Gemini in Docs** on Standard and Plus. Ask your admin which apps are switched on.
- **Microsoft 365**: Microsoft Copilot Chat (recently renamed from Microsoft 365 Copilot Chat) is available **at no extra cost** to users signing in with a **work or school (Entra) account** on qualifying plans, including Microsoft 365 Business Basic, Standard and Premium and E3/E5. The separately licensed Copilot add-on adds Copilot inside the Office apps and work-data grounding; Copilot Chat is the baseline.
Include the bundled assistant in your test as a candidate. It may cover email and document work well enough on its own.
### 3. Log your real tasks for 3–5 days
Every time you use an AI assistant, note one line: **task, which assistant, why that one, was the result usable**. Typical categories: writing/editing, summarising documents, web research with sources, data analysis, coding, images, voice.
This list prevents the test from being decided by benchmark hype instead of your work. Weight tasks by how often you do them and how much they matter.
### 4. Run the one-week side-by-side test
Use the **evaluation set** (artifact below): pick at least 6 tasks and replace the placeholders with your real material. Run each task in **every** assistant in a **fresh chat**, with the same files.
Paste outputs into a doc labelled A/B/C rather than by product so you score them blind (ask someone else to label them if you can). Fill in the **scoring sheet** CSV. Use the **Blind judge prompt** for a second opinion, but your own scores decide.
Also note features you can't score with a single prompt but genuinely rely on: projects/memory, file limits, voice mode, deep research, image generation, coding agents, integrations.
### 5. Check data and privacy settings for the finalists
For the assistant(s) you might keep, check data controls: whether chats are used for model training on your plan and how to opt out, retention, and whether business plans give contractual protections your work needs. If you handle client or personal data, a business/team plan or the bundled work assistant may be the only acceptable option.
### 6. Decide: one primary, maybe one specialist
Rules of thumb:
- Keep the assistant with the **highest weighted score on your top-3 tasks**.
- Keep a **second** only if it wins clearly on a task you do **weekly** that the first can't do acceptably (e.g. a coding agent, or cited research), and the time saved is worth its price.
- Use the **bundled** Gemini or Copilot Chat for email and document chores, and **free tiers** for occasional second opinions.
### 7. Export, then cancel at renewal
Before cancelling each subscription:
1. Export your data (most assistants offer a data export in settings) and copy any custom instructions, project instructions, custom GPTs/Gems or saved prompts into a doc.
2. Move important ongoing projects to the assistant you keep and test them there.
3. Cancel auto-renew now so the plan ends at the paid-through date. Put a reminder a week before the date to confirm.
## Prompts
### Task inventory prompt
Turn your rough task log into a weighted test plan.
```text
Here is my log of AI tasks from the last {{days_logged}} days:
{{task_log}}
Group these into 5–8 task types. For each give: how often I do it per week, how important it is (1–3), and one representative test task I could run in several assistants with the same inputs. Output a table I can paste into a spreadsheet.
```
### Blind judge prompt
Get a structured second opinion on anonymised outputs (run it in an assistant you're not testing, or in each and compare).
```text
You are judging answers from anonymous assistants A, B and C to the same task. Don't guess which product wrote which.
Task given to each: {{task_prompt}}
What a great answer must include: {{success_criteria}}
A: {{output_a}}
B: {{output_b}}
C: {{output_c}}
For each, score 1–5 on accuracy, completeness, and usefulness for {{audience}}, and list factual errors or unsupported claims. Then rank them and give the deciding reason in one sentence.
```
### Citation check prompt
Audit the sources a research answer gave you.
```text
Below is an answer with citations, and the text I copied from each cited page.
Answer: {{research_answer}}
Source texts: {{source_texts}}
For each claim, say whether the cited source supports it (yes / partly / no / source missing). Count the unsupported claims.
```
### Decision memo prompt
Summarise the test into a decision you (or your manager) can sign off.
```text
Here is my scoring sheet: {{scoring_sheet_csv}}
Subscriptions and costs: {{subscriptions_with_prices_and_renewal_dates}}
Bundled assistants available through work: {{bundled_assistants}}
Write a short decision memo: which assistant to keep as primary and why; whether a second is justified (only if it wins clearly on a weekly task); which to cancel and on what date; the monthly saving; and the risks (lost features, data to export).
```
## Templates and scripts
### One-week evaluation prompt set (template)
```markdown
# AI assistant evaluation set – one week
Run each task in EVERY assistant you're testing, in a fresh chat, with the same inputs.
Replace {{placeholders}} with your real material. Use at least 6 of the 10; add your own from your task log.
Paste outputs into a doc labelled A/B/C (not by product) so you score them blind.
## T1 – Summarise a real document (reading & accuracy)
Attach: {{a_real_10_to_30_page_pdf}}
"Summarise this for {{audience}} in 8 bullets. Then list 3 things that are easy to miss, each with the page number."
Check: are page numbers and facts correct? Did it miss anything you know matters?
## T2 – Write in your voice (writing)
"Here are 3 emails I wrote: {{three_sample_emails}}. Write a reply to this message in my style: {{incoming_message}}."
Check: would you send it with fewer than 3 edits?
## T3 – Current-events research with sources (web research)
"What changed in {{topic_in_your_field}} in the last 3 months? Give 5 points, each with a source link and date."
Check: open every link. Count dead, wrong or unsupported citations.
## T4 – Spreadsheet analysis (data)
Attach: {{a_csv_or_xlsx_you_know_well}}
"Which 3 {{rows_entity}} changed most between {{period_a}} and {{period_b}}? Show the calculation and a chart."
Check: verify one number by hand.
## T5 – Your hardest recurring task (domain)
"{{your_hardest_recurring_task_prompt}}"
Check: score against what an expert colleague would produce.
## T6 – Code or formula (technical)
"{{a_formula_script_or_code_task_you_actually_need}}"
Check: does it run first time? How many follow-ups to make it work?
## T7 – Long conversation memory (context)
Paste a 3,000+ word brief, then ask 5 follow-up questions over 15 minutes, including one that depends on a detail from the start.
Check: did it keep the early detail right?
## T8 – Image or visual (only if you need it)
"{{an_image_or_diagram_you_actually_need}}"
Check: usable without another tool?
## T9 – Work-data question (only for bundled work assistants)
In Gemini (Workspace) or Copilot Chat (Microsoft 365): "{{question_about_your_own_files_or_email}}"
Check: does it find the right file/email? (Depends on your plan and admin settings.)
## T10 – Refusal and honesty
"{{a_question_where_the_honest_answer_is_i_dont_know}}"
Check: does it admit uncertainty instead of inventing an answer?
```
### Scoring sheet (CSV, import into Google Sheets or Excel) (template)
```csv
task_id,task,weight(1-3),assistant_label,quality(1-5),accuracy(1-5),edits_needed(0=none;5=rewrite),minutes_to_usable,would_pay_for_this_alone(y/n),notes
T1,Summarise document,3,A,,,,,,
T1,Summarise document,3,B,,,,,,
T1,Summarise document,3,C,,,,,,
T2,Write in my voice,2,A,,,,,,
T2,Write in my voice,2,B,,,,,,
T2,Write in my voice,2,C,,,,,,
T3,Research with sources,2,A,,,,,,
T3,Research with sources,2,B,,,,,,
T3,Research with sources,2,C,,,,,,
T4,Spreadsheet analysis,1,A,,,,,,
T4,Spreadsheet analysis,1,B,,,,,,
T4,Spreadsheet analysis,1,C,,,,,,
T5,Hardest recurring task,3,A,,,,,,
T5,Hardest recurring task,3,B,,,,,,
T5,Hardest recurring task,3,C,,,,,,
,,,,,,,,,
# Score per assistant = SUM(weight * (quality + accuracy + (5 - edits_needed))) / SUM(weight)
# In Google Sheets / Excel, put this in a summary tab:
# =SUMPRODUCT((D2:D16="A")*C2:C16*(E2:E16+F2:F16+(5-G2:G16)))/SUMPRODUCT((D2:D16="A")*C2:C16)
# Then list, per assistant: monthly cost, billing type (monthly/annual), renewal date, unique features you used this week.
```
## Limitations
- A one-week test reflects today's models; vendors update models and plans often, so re-run it every 6–12 months.
- Blind scoring is imperfect: answer styles can give away the product.
- Bundled work assistants (Gemini in Workspace, Copilot Chat) vary by plan and admin settings, and may not include features of the consumer paid plans.
- Some unique features (a specific coding agent, deep-research mode, image generation, voice) can't be replaced by another assistant one-for-one.
- Chat history, custom GPTs/Gems and project setups don't transfer between vendors automatically.
## Risks
- **functionality_loss** (medium): Cancelling can remove a feature you only notice later (for example a coding agent or deep research). Keep the free tier of cancelled assistants for occasional use.
- **data_portability** (medium): Chat histories, custom instructions and project files stay with the vendor. Export them before cancelling.
- **sensitive_data** (medium): Testing with real documents sends them to every assistant. Use non-confidential material or business plans with suitable data terms.
- **compliance** (low): Employers may have an approved-tools policy; consolidating onto a bundled work assistant can simplify compliance.
## Verification checklist (run before cancelling)
- [ ] Your list shows every AI subscription with price, billing type and renewal date, and you've checked your Workspace or Microsoft 365 plan for an included assistant.
- [ ] You ran at least 6 tasks from your own log in every candidate, in fresh chats, with the same inputs.
- [ ] The scoring sheet has a weighted score per assistant, and the winner also wins (or ties) on your single most important task.
- [ ] For one full working day you used only the assistant you plan to keep (plus the bundled one) and didn't hit a blocker.
- [ ] Data exports and copies of custom instructions/projects exist for each assistant you will cancel.
- [ ] Auto-renew is switched off for the cancelled plans, with a calendar reminder before each end date.
## Savings methodology
Savings = sum of monthly costs of the subscriptions you cancel (annual prices ÷ 12). Incremental cost is $0 because you keep a subscription you already pay for and any assistant bundled in your existing suite. Savings start at each plan's renewal date; annual plans are not assumed to refund. Testing time is not monetised.
Example: A consultant on Google Workspace pays for Claude Pro monthly and Perplexity Pro annually. After the test, Claude wins on writing and document work, Perplexity's cited research is only slightly better on a weekly task, and Gemini in Gmail handles email chores. $37/mo → $20/mo. Assumptions: Claude Pro: $20/month billed monthly, per claude.com/pricing on 2026-10-13. Perplexity Pro: $17/month billed annually, per perplexity.ai/hub/pricing on 2026-10-13. The saving starts at the annual renewal date. Google Workspace is already paid for; Gemini in Gmail and the Gemini app are listed on its Business plans, so no new cost. If the consultant also paid for ChatGPT Plus, cancelling it would add to the saving; its price wasn't verified today because OpenAI's pages blocked our fetcher.
## Sources
- Claude pricing: https://claude.com/pricing (accessed 2026-10-13)
- Perplexity pricing: https://www.perplexity.ai/hub/pricing (accessed 2026-10-13)
- Google Workspace pricing: https://workspace.google.com/pricing (accessed 2026-10-13)
- Microsoft Learn: Manage Microsoft Copilot Chat: https://learn.microsoft.com/en-us/copilot/manage (accessed 2026-10-13)
- ChatGPT pricing: https://openai.com/chatgpt/pricing/ (accessed 2026-10-13)
# Replace Tableau Creator seats with Data Studio (Looker Studio) dashboards on Google Sheets
> For SMB reporting on spreadsheet-sized data: prep your Sheets properly, build the dashboard in Google's free Data Studio (renamed from Looker Studio in 2026), use Claude or ChatGPT for analysis and formulas, and schedule PDF email delivery. Microsoft 365 shops can use free Power BI Desktop, but sharing needs Power BI Pro.
- Outcome: A refreshable, shareable dashboard of the business's key numbers, emailed on a schedule, without paying for Tableau Creator licences.
- Replaces: Tableau (partial)
- Uses: Looker Studio — Free dashboarding tool (now called Data Studio) with a native Google Sheets connector, calculated fields, sharing and scheduled email delivery.; Google Workspace — Google Sheets as the data source and Drive for sharing. A free Google account also works with lower email-delivery quotas.; Claude (optional) — Data cleaning formulas, calculated-field syntax, SQL if you move to BigQuery, and narrative analysis of exported data.; ChatGPT (optional) — Alternative for formula help and analysing CSV exports.; Microsoft 365 (optional) — Alternative path: Excel PivotTables/charts and the free Power BI Desktop app. Sharing Power BI reports requires Power BI Pro (included in Microsoft 365 E5) or Premium Per User.
- Extra monthly cost: $0 — Data Studio's standard tier and Google Sheets cost nothing extra. Data Studio Pro (team workspaces, organisation-owned content, more schedules) is a paid upgrade you only need for governance. On the Microsoft path, Power BI Desktop is free but sharing needs Power BI Pro ($14/user/month paid yearly per Microsoft's pricing page) unless you're on E5.
- Difficulty: moderate; skill: basic; setup: 1–2 days for the first dashboard (data prep is most of it); a few hours for each after
- Evidence: editorial; illustrative example: true; last reviewed 2026-10-13
- Canonical: https://onelesstool.com/workflows/dashboards-from-spreadsheets
## Good fit for
- Small and mid-sized businesses whose data already lives in Google Sheets, CSV exports or Google products
- A handful of standard dashboards (sales, marketing, operations) rather than heavy ad-hoc exploration
- Teams paying for one or two Tableau Creator seats mainly to publish a few recurring reports
- Anyone comfortable with spreadsheets who can learn calculated fields
## Not a fit for
- Row-level security (each viewer sees only their own rows), certified data sources and formal governance — stay on a BI platform or use Data Studio Pro plus BigQuery with care
- Large data (millions of rows) living in spreadsheets — Sheets and the Sheets connector aren't built for it; use a warehouse
- Analysts who rely on Tableau's advanced visual analytics (LOD expressions, complex table calcs, custom geocoding, extensive drag-and-drop exploration)
- Regulated reporting that needs audit trails on metric definitions and data lineage
## Before
One or two people hold Tableau Creator licences to connect to spreadsheets/CSVs, build workbooks and publish dashboards; others view them with Viewer/Explorer licences. Most of the dashboards are a few scorecards, trend lines and a table refreshed from spreadsheets.
## After
Clean, append-only Google Sheets tabs feed a Data Studio report built from a written dashboard spec. Calculated fields define metrics once (win rate, net revenue). Viewers open a link; owners get a scheduled PDF every Monday. When someone asks 'why did revenue drop?', you export the table and ask Claude or ChatGPT for an analysis — and verify its numbers against the sheet.
## Steps
### 1. List what your Tableau dashboards actually do
For each published workbook, note: viewers, how often it's opened, data sources, row counts, and any feature you'd miss — **row-level security, LOD expressions, extracts from databases, subscriptions**. If a dashboard relies on row-level security or millions of rows, keep it on Tableau (or plan a warehouse) and migrate the rest. Many SMBs find most dashboards are scorecards, trends and tables — those move easily.
### 2. Prepare the data in Google Sheets
Work through the **Data-prep checklist** artifact. The two rules that prevent most problems:
1. **One header row, one record per row, one type per column.**
2. **Long, not wide** — a `Month` column, not twelve month columns.
Use an AI assistant on a **50-row sample** (not your whole sensitive dataset) with the **Find data-quality problems** prompt to get exact cleaning formulas, e.g. `=DATEVALUE()`, `=TRIM()`, `=IFERROR(VALUE(...))`. Paste the fixed values back.
Note: the Sheets connector **includes hidden and filtered cells by default** — delete junk rows rather than hiding them.
### 3. Write the dashboard spec before opening the tool
Fill in the **Dashboard spec** template: audience, the decision it supports, at most six metrics with plain-English definitions, default date range, filters, layout and acceptance numbers.
This replaces the most valuable thing a Tableau specialist brought — agreed metric definitions. Get the person who reads the dashboard to sign off on the metric definitions now, not after building.
### 4. Build in Data Studio (formerly Looker Studio)
1. Open Data Studio (formerly Looker Studio) → **Blank report** → **Google Sheets** connector → pick spreadsheet and tab, keep **Use first row as headers** on.
2. **Resource → Manage data sources → Edit** and add the calculated fields from the **Calculated-field cookbook** (e.g. `Is won = CASE WHEN Stage = "Won" THEN 1 ELSE 0 END`, `Win rate = SUM(Is won) / SUM(Is closed)`).
3. Build row by row from the spec: scorecards with **comparison date range**, a time series, a bar chart breakdown, a table with conditional formatting.
4. Add **Date range** and **Drop-down list** controls for the filters in the spec.
5. **Theme and layout**: set your brand colours once in the theme so every page matches.
6. Data credentials: *Owner's credentials* lets viewers see data without access to the sheet — decide deliberately.
If you add or rename columns in the sheet later, **refresh fields** in the data source.
### 5. Share and schedule email delivery
- **Share** with named people or a Google Group (avoid 'anyone with the link' for business data).
- **Schedule delivery** (Share menu → *Schedule delivery*): the report is emailed as a **PDF** with a link; choose pages and recipients (up to 50 addresses), and a repeat frequency. Without Pro, the most frequent option is **daily** and each report can have **one schedule**.
- Quotas: individual accounts can send 200 scheduled emails/day and 800/month; Google Workspace 500/day and 3,000/month, per Google's docs. Scheduled PDFs use US number/date formats.
**Microsoft 365 alternative:** Excel PivotTables and charts cover simple needs. **Power BI Desktop** is a free download, but sharing reports requires Power BI **Pro** ($14/user/month paid yearly) or Premium Per User; Pro is included in Microsoft 365 E5 / Office 365 E5.
### 6. Use AI for analysis — and check its arithmetic
Export a chart's table (**⋮ → Export → CSV/Google Sheets**) and use the **Explain a change in a metric** prompt. AI assistants are good at spotting patterns and drafting the commentary for a monthly report, but they can mis-add or misread columns. Require the assistant to show its working table, and check two numbers against the dashboard before sending anything.
Then run both systems for one reporting cycle and compare the acceptance numbers from the spec before cancelling Tableau seats.
## Prompts
### Find data-quality problems
Get exact cleaning formulas from a small sample before building the dashboard.
```text
Here are the header row and a 50-row sample from my Google Sheets tab "{{tab_name}}" (CSV):
{{sample_csv}}
One row should represent one {{row_grain}}. Identify:
1. Columns with mixed types or text where a number/date is expected.
2. Inconsistent category spellings (list the variants you see).
3. Rows that look like subtotals, blanks or duplicates.
4. Whether the table is 'wide' and should be reshaped long.
For each problem, give the exact Google Sheets formula for a helper column that fixes it (assume data starts in row 2). Don't guess about rows you haven't seen; tell me what to check in the full sheet.
```
### Write a calculated field
Get correct Data Studio / Looker Studio formula syntax for a metric definition.
```text
Write a calculated field for Google Data Studio (formerly Looker Studio). My data source fields (name: type) are:
{{field_list}}
Metric I want, in words: {{metric_definition}}
Give: (1) the formula, (2) whether it must be created in the data source or the chart, (3) the field type/format to set, (4) one way to sanity-check the result against the source sheet. Use only functions that exist in Data Studio; if what I want needs a blend or can't be done, say so. If my previous attempt failed, here is the formula and error: {{previous_error}}
```
### Explain a change in a metric
Draft monthly commentary from exported dashboard data, with visible working.
```text
I'm the {{role}} at a {{business_type}}. Below is a table exported from our dashboard (CSV). {{metric}} changed from {{previous_value}} in {{previous_period}} to {{current_value}} in {{current_period}}.
{{exported_csv}}
1. Show a working table that reproduces both values from the data. If you can't reproduce them, stop and tell me why.
2. Break the change down by {{breakdown_dimension}}: which segments drove it (absolute and % contribution).
3. Note anything that looks like a data problem rather than a business change.
4. Write a 4-sentence summary for the owners in plain English, no jargon, no speculation beyond the data.
```
## Templates and scripts
### Data-prep checklist for Google Sheets sources (checklist)
```
DATA-PREP CHECKLIST — do this in Google Sheets before connecting Looker Studio / Data Studio
Shape
[ ] One table per tab, starting at A1, ONE header row (Data Studio uses the first row as field names)
[ ] One row = one event/record (one order, one deal, one ticket) — no subtotal rows, no blank spacer rows, no merged cells
[ ] Long, not wide: a "Month" column + a "Value" column, NOT one column per month
[ ] Every column has a single data type all the way down (no "N/A" text in a number column — leave blank)
[ ] Stable, unique ID column (Order ID, Deal ID) for COUNT_DISTINCT and joins (blends)
Types and formats
[ ] Dates are real dates (right-aligned in Sheets), one format, no text like "Q3"
[ ] Currency columns are plain numbers; currency code in its own column if you have several
[ ] Categories spelled consistently (use data-validation dropdowns at the source; =UNIQUE(C2:C) to spot variants)
[ ] Percentages stored as numbers (0.25), not text ("25%")
Hygiene
[ ] Raw data tabs are append-only; calculations live in Data Studio calculated fields or a separate "model" tab
[ ] Hidden/filtered rows: Data Studio INCLUDES hidden and filtered cells by default — delete junk rows instead of hiding them
[ ] Freeze the header row and protect it (Data → Protect sheets and ranges) so nobody renames a column
(renaming breaks fields; after any column change, Resource → Manage data sources → Edit → Refresh fields)
[ ] Store the sheet in a normal Drive folder (the connector docs say shared/Team Drive files aren't supported)
[ ] Volume check: if a tab is approaching hundreds of thousands of rows or reports feel slow, move the data to BigQuery
AI help (paste a 50-row sample, never the full sensitive dataset)
[ ] Ask the assistant to list type problems, inconsistent categories and outliers in the sample
[ ] Ask for the exact Sheets formulas to fix them in a helper column, then paste values
```
### Dashboard spec template (template)
```
DASHBOARD SPEC — fill this in before building anything (one page per dashboard)
1. Purpose
Audience: ______ (e.g. owners + 2 account managers)
Decision it supports: ______ (e.g. "which clients need attention this week")
Review cadence: ______ (e.g. Monday 9:00, emailed PDF)
2. Data sources
| Source (sheet/tab) | Grain (one row = …) | Refresh (manual/form/import) | Owner | Rows today |
|--------------------|---------------------|------------------------------|-------|-----------|
3. Metrics (max 6 on page 1)
| Metric | Definition in words | Data Studio formula | Target / good-bad | Format |
| Revenue | Sum of invoice amount, excl. credit notes | SUM(Amount) with filter Type = Invoice | ≥ 40k/mo | Currency, 0 dp |
| Win rate | Won deals ÷ closed deals | SUM(CASE WHEN Stage = "Won" THEN 1 ELSE 0 END) / SUM(CASE WHEN Stage IN ("Won","Lost") THEN 1 ELSE 0 END) | ≥ 30% | Percent |
| Avg deal size | Won value ÷ won deals | SUM(Won value) / SUM(Is won) | | Currency |
4. Dimensions & filters
Date field: ______ Default range: ______ (e.g. last 12 months, compare to previous period)
Filter controls: ______ (e.g. Owner, Client segment)
Breakdowns: ______ (e.g. by month, by service line)
5. Layout (top → bottom)
Row 1: scorecards (metric + comparison)
Row 2: trend line (metric by month)
Row 3: breakdown bar chart + table with conditional formatting
Page 2+: detail tables for drill-down
6. Access & delivery
Viewers: ______ Editors: ______ Data credentials: owner's / viewer's
Scheduled email: recipients ______, frequency ______, pages ______
7. Acceptance test
Three numbers checked against the source sheet by hand: ______ / ______ / ______
```
### Calculated-field cookbook (template)
```
CALCULATED-FIELD COOKBOOK (Data Studio, formerly Looker Studio) — add via Resource → Manage data sources → Edit → Add a field
Field names below assume headers: Date, Amount, Stage, Client, Owner, Type, Deal ID.
Flag won deals (number, so it can be summed)
Is won = CASE WHEN Stage = "Won" THEN 1 ELSE 0 END
Closed deals
Is closed = CASE WHEN Stage IN ("Won", "Lost") THEN 1 ELSE 0 END
Win rate (aggregate field — set type to Percent)
Win rate = SUM(Is won) / SUM(Is closed)
Won value
Won value = CASE WHEN Stage = "Won" THEN Amount ELSE 0 END
Deal count (unique)
Deals = COUNT_DISTINCT(Deal ID)
Revenue excluding credit notes
Net revenue = CASE WHEN Type = "Credit note" THEN -Amount ELSE Amount END
Simple bucketing
Deal size band = CASE WHEN Amount >= 10000 THEN "10k+" WHEN Amount >= 2500 THEN "2.5k–10k" ELSE "<2.5k" END
Cleaning inconsistent text
Client clean = TRIM(UPPER(Client))
Age in days
Days open = DATE_DIFF(TODAY(), Date)
Tips
- Monthly trend: change the Date field's type to "Year Month" in the chart, rather than a formula.
- Put row-level CASE fields in the data source; ratios like Win rate must aggregate (SUM(...)/SUM(...)), never average a per-row ratio.
- If a formula errors, paste the error and your field list into the "Write a calculated field" prompt.
```
## Limitations
- No row-level security in the standard tier when using owner's credentials — every viewer of a report sees all the data it shows.
- Governance is light: content belongs to individual users unless you pay for Data Studio Pro (organisation-owned content, team workspaces).
- Google Sheets is not a database; large or fast-growing data makes reports slow and should move to BigQuery or another warehouse.
- Fewer advanced analytics and visual options than Tableau (no equivalent of LOD expressions; limited custom visuals).
- Scheduled delivery without Pro: at most daily, one schedule per report, PDF in US number/date formats, email quotas per account.
- AI-generated analysis can contain arithmetic or interpretation errors; it needs checking before it's shared.
## Risks
- **security** (medium): Reports using owner's credentials expose the underlying data to anyone the report is shared with, even if they can't open the sheet. Share with named people/groups only.
- **functionality_loss** (medium): Losing row-level security, certified data sources and advanced calculations can matter for multi-team or client-facing reporting.
- **reliability** (medium): Renamed or retyped columns in the source sheet break fields silently until someone refreshes the data source. Protect header rows.
- **collaboration** (low): If the report owner leaves, ownership must be transferred; Data Studio Pro's organisation ownership avoids this.
- **sensitive_data** (low): Pasting full customer or financial datasets into an AI assistant may conflict with your data policies; use samples or aggregated exports and a business plan where possible.
## Verification checklist (run before cancelling)
- [ ] For every migrated dashboard, three acceptance numbers from the spec match the Tableau version (and the source sheet) for the same date range.
- [ ] Changing a value in the source sheet appears in the report after refresh, and adding a new row for today shows up in the current period.
- [ ] A test viewer without access to the sheet can see exactly what you intend (and nothing more) using the chosen data-credential setting.
- [ ] The scheduled email arrives on time with the right pages as a PDF for at least two cycles, and recipients can open the link.
- [ ] Renaming a test column breaks a test field as expected, and your documented fix (refresh fields) restores it — so you know the recovery path.
- [ ] Every person who views a Tableau dashboard today has confirmed the replacement meets their need before seats are cancelled.
## Savings methodology
Savings = Tableau licences you can cancel (Creator, and any Explorer/Viewer seats only used for these dashboards) × price per month − incremental cost (normally $0 for Data Studio on Google Sheets; Data Studio Pro or Power BI Pro if you need them). Tableau is typically billed annually, so savings start at renewal. Build time is not counted in dollars.
Example (illustrative): A 25-person services firm with 2 Tableau Creator seats used to publish 4 recurring dashboards from Google Sheets, moving to Data Studio. $150/mo → $0/mo. Assumptions: ILLUSTRATIVE: 2 Creator seats at $75/user/month billed annually, as reported by third-party pricing guides. Tableau's own pricing pages returned HTTP 403 to our fetch today, so this price is not verified — use your contract. No Explorer/Viewer seats are included; viewers move to free Data Studio sharing. Standard (free) Data Studio is sufficient; no Data Studio Pro. Savings realised at the next Tableau renewal.
## Sources
- Data Studio — Schedule automatic report delivery: https://docs.cloud.google.com/looker/docs/studio/schedule-automatic-report-delivery (accessed 2026-10-13)
- Data Studio — Connect to Google Sheets: https://docs.cloud.google.com/looker/docs/studio/connect-to-google-sheets (accessed 2026-10-13)
- About Data Studio Pro: https://docs.cloud.google.com/looker/docs/studio/about-looker-studio-pro (accessed 2026-10-13)
- Power BI pricing: https://www.microsoft.com/en-us/power-platform/products/power-bi/pricing (accessed 2026-10-13)
- Tableau pricing: https://www.tableau.com/pricing/teams-orgs (accessed 2026-10-13)
# Drop Grammarly when you already pay for ChatGPT, Claude, Gemini or Copilot
> If you already pay for an AI assistant, prompts handle rewrites, tone and careful proofreading, and built-in checkers (Google Docs, Microsoft Editor, Apple Writing Tools) cover everyday spelling and grammar. What you genuinely lose is Grammarly's real-time underlines everywhere you type, plus plagiarism checks, AI detection and team style guides on paid plans.
- Outcome: Write clean, on-tone emails and documents without paying for a separate writing-assistant subscription.
- Replaces: Grammarly (full)
- Uses: ChatGPT (optional) — Proofreading, rewriting and tone prompts in a saved Project.; Claude (optional) — Proofreading, rewriting and tone prompts in a saved Project.; Google Gemini (optional) — Same prompts in a Gem, plus Gemini in Gmail/Docs on eligible Workspace plans.; Microsoft 365 Copilot (optional) — Same prompts inside Word and Outlook for Microsoft 365 users with Copilot.; Google Workspace (optional) — Google Docs' built-in spelling and grammar suggestions.; Microsoft 365 (optional) — Microsoft Editor in Word and Outlook (advanced refinements with a Microsoft 365 subscription).
- Extra monthly cost: $0 — $0 extra: this uses an AI assistant you already pay for plus spell/grammar checkers built into Google Docs, Microsoft 365/Edge and Apple devices with Apple Intelligence.
- Difficulty: easy; skill: none; setup: 30–45 minutes (style guide, saved Project, snippet/shortcut)
- Evidence: editorial; illustrative example: false; last reviewed 2026-10-13
- Canonical: https://onelesstool.com/workflows/drop-grammarly-with-existing-ai
## Good fit for
- Individuals and small teams paying for Grammarly Pro on top of an AI assistant subscription
- People who mostly use Grammarly for occasional rewrites, tone changes and a final proofread
- Writers who prefer to see a list of changes and decide, rather than accept underlines one by one
- Teams that write mainly in Google Docs, Word or Outlook, where built-in checkers already exist
## Not a fit for
- People who depend on real-time suggestions in every text box (CRM, support desk, Slack, browser forms) — copy-paste to an AI is slower
- Teams that need an enforced shared style guide or brand tones across many members, with admin controls
- Anyone who needs plagiarism checks or AI-text detection (both included in Grammarly Pro per its plans page)
- Non-native writers who learn from inline explanations as they type, or users who rely on Grammarly's accessibility benefits
## Before
Grammarly runs as a browser extension, desktop app and inside Office/Google Docs, underlining issues as you type everywhere and offering one-click fixes, tone suggestions, AI rewrites, and on Pro, plagiarism and AI detection plus a style guide and brand tone.
## After
Spelling and grammar are caught by the checker built into the app you're in (Google Docs, Word/Outlook's Editor, Edge, or Apple Writing Tools). For anything that matters — client emails, proposals, posts — you paste the text into a pinned 'Proofread' Project in ChatGPT/Claude/Gemini/Copilot that holds your house style, and get a list of every change with a reason, then the corrected text. A text-replacement snippet or Quick Action makes it a few keystrokes.
## Steps
### 1. Be clear about what you'll lose
Grammarly's plans page today lists (Free → Pro at **$12/month** → Enterprise):
- **Real-time suggestions everywhere you type** — browser extension (Chrome, Safari, Firefox, Edge), desktop apps (Windows, Mac), Office, Google Docs, mobile.
- **Plagiarism detection** and an **AI detector** — Pro and Enterprise only.
- **Style guide** and **brand tones** — 1 each on Pro, unlimited on Enterprise.
- **AI prompts** — 100/month Free, 2,000/member/month Pro.
(Grammarly is now part of Superhuman, per its site.) If plagiarism checks, enforced team style or underlines in every web form are essential to you, keep Grammarly — or drop to Grammarly **Free** for basic underlines and use your AI assistant for the rest.
### 2. Turn on the checkers you already have
- **Google Docs:** red underlines = spelling, blue = grammar; press **Tab** to accept. Run **Tools → Spelling and grammar** before sending. Grammar suggestions support English, Spanish, French, German, Portuguese and Italian.
- **Microsoft Word / Outlook:** **Microsoft Editor** (Home tab in Word for the web and Word for Microsoft 365 on Windows; Outlook on the web/Outlook.com). Basic spelling and grammar is free with a Microsoft account; clarity, conciseness and formality refinements need Microsoft 365.
- **Browsers:** Microsoft Edge has built-in spelling/grammar in text boxes; the **Microsoft Editor extension** works in Chrome and Edge.
- **Apple devices with Apple Intelligence:** select text → Control-click → **Writing Tools** → *Proofread* or *Rewrite* (Friendly, Professional, Concise). Works in most apps and websites, subject to language and region availability.
### 3. Write a one-page house style guide
Copy the **House style guide** artifact and fill in the brackets: US or UK spelling, Oxford comma, date format, product names, banned words, tone presets.
If you don't have one, paste 5–10 pieces of writing you're proud of into the **Build a style guide from examples** prompt and edit what it returns. Keep it **under one page** — models follow short, concrete rules more consistently.
This replaces Grammarly's style guide feature for an individual or a small team (share the doc), though nothing *enforces* it.
### 4. Create a pinned 'Proofread' workspace
In whichever assistant you pay for, create a **Project** (Claude or ChatGPT), a **Gem** (Gemini) or a **notebook/saved prompt** (Copilot) called **Proofread**. Paste the style guide plus the **Proofread with a change list** prompt as its instructions.
The key design choice: the output is a **numbered list of changes** (`original → corrected (reason)`) *before* the corrected text. You review changes instead of trusting a silent rewrite — the main way AI 'proofreading' goes wrong is quietly changing meaning, numbers or names.
For tone, use the **Rewrite for tone** prompt with one of your presets.
### 5. Make it a few keystrokes
Follow the **Keyboard-shortcut workflow** artifact:
1. A **text replacement** (macOS: System Settings → Keyboard → Text Replacements) so typing `;pf` expands into the proofreading instruction.
2. **Copy → switch to the pinned Proofread project → paste → Enter → copy the result back.** About 20–30 seconds.
3. Optional macOS **Quick Action** built in the Shortcuts app that wraps selected text in the prompt, copies it and opens your AI app, bound to a keyboard shortcut.
Use built-in underlines for everyday typing and the AI pass for anything external or important.
### 6. Test for two weeks before cancelling
Keep Grammarly on but ignore its suggestions for two weeks. On 10 real pieces of writing, run the AI proofread first, then look at what Grammarly would still flag. If Grammarly consistently catches real errors your new routine misses (not just style preferences), keep it or drop to Free. Otherwise cancel at the end of the billing period.
## Prompts
### Proofread with a change list
Careful proofreading that shows every change so nothing is altered silently.
```text
Proofread the text below as a careful copy editor.
Style: {{style_guide}}
Audience: {{audience}}
English variety: {{english_variety}}
Rules:
- Fix only spelling, grammar, punctuation, clear typos, and violations of the style guide.
- Do NOT change meaning, facts, numbers, names, links, quotes or formatting. Don't 'improve' wording that is already correct.
- If something might be an error but you're not sure (a name, a figure, a technical term), don't change it — list it under "Check".
Output exactly:
1. **Changes** — a numbered list: "original" → "corrected" (reason, max 8 words). Write "No changes needed" if none.
2. **Check** — anything ambiguous for me to decide.
3. **Corrected text** — in one block, ready to paste.
Text:
{{text}}
```
### Rewrite for tone
Replace Grammarly's tone suggestions with an explicit tone rewrite.
```text
Rewrite the text below in the {{tone_preset}} tone from my style guide, for {{audience}}.
Style guide:
{{style_guide}}
Constraints: keep every fact, number, date, name and commitment exactly; keep it within {{max_words}} words; keep any links. Give two versions (A: closest to the original wording, B: more rewritten), then one line on what you changed in each.
Text:
{{text}}
```
### Build a style guide from examples
Draft a one-page house style guide from writing you already like.
```text
Here are {{number_of_samples}} examples of writing from {{company_or_person}} that represent the voice we want:
{{writing_samples}}
Draft a one-page house style guide using ONLY patterns you can see in these samples. Sections: Language & spelling, Voice (5 bullet rules), Terms (always/never), Don'ts, three Tone presets (SUPPORT, SALES, INTERNAL) with one example sentence each. Where the samples are inconsistent, show both options and ask me to choose. Keep it under 350 words.
```
## Templates and scripts
### House style guide template (template)
```
HOUSE STYLE GUIDE — paste into a Claude Project / ChatGPT Project / Gemini Gem / Copilot notebook as standing instructions.
Keep it under one page; the AI follows short, concrete rules better than long essays.
Language & spelling
- Variety: [US English | UK English]. Spell-check against it; flag the other variety's spellings.
- Oxford comma: [yes | no]. Numbers: words for one–nine, digits for 10+; always digits with units (5 GB, 3 days).
- Dates: [13 October 2026 | October 13, 2026]. Times: [9:00 | 9am]. Currency: $1,200 (no ".00" unless cents matter).
Voice
- Plain, direct, friendly. Short sentences (aim < 20 words). Active voice.
- We say "you" to the reader and "we" for the company. No "utilize", "leverage", "synergy", "best-in-class".
- Headlines and buttons: [Sentence case | Title Case].
Terms (always / never)
- [Product name] — always capitalised like this; never "[wrong variant]".
- [customer | client] — use "[chosen]" only.
- [sign up] (verb) vs [sign-up] (noun).
Don'ts
- No exclamation marks in customer emails. No emojis in [contracts, invoices, support replies].
- Never promise dates, refunds or legal outcomes in support replies.
Tone presets (ask for one by name)
- SUPPORT: warm, apologise once, give the next step and a timeframe.
- SALES: confident, specific benefit, one question at the end.
- INTERNAL: brief, bullet points, decision first.
Editing rules for the AI
- Preserve meaning, facts, names, numbers, links and formatting. Never add claims.
- When proofreading, change only what's wrong or violates this guide; list every change.
```
### Keyboard-shortcut workflow (checklist)
```
KEYBOARD-SHORTCUT WORKFLOW — get "select text → fixed text" down to a few keystrokes
1. One saved place for the instructions (all platforms)
Create a Claude Project, ChatGPT Project, Gemini Gem or Copilot notebook called "Proofread" and paste the
house style guide + the proofreading prompt as its instructions. Pin it. Each check is then just: paste text → Enter.
2. Snippet expansion for the prompt (no extra software)
macOS: System Settings → Keyboard → Text Replacements → "+"
Replace: ;pf With: Proofread the text below using my house style. List every change as
"original → corrected (reason)", then give the corrected text in one block. Change nothing else.
Windows/other: use your existing text-expander, or keep the prompt as the Project instructions above.
3. Copy → ask → paste back (any app, any OS)
Select text → Cmd/Ctrl+C → switch to the Proofread project (Cmd/Ctrl+Tab or the AI app's own global shortcut,
if your app offers one — check its settings) → Cmd/Ctrl+V → Enter → copy the corrected block → switch back →
Cmd/Ctrl+V.
4. Built-in, no AI subscription needed
- Apple devices with Apple Intelligence: select text → Control-click → Writing Tools → Proofread / Rewrite
(Friendly, Professional, Concise) — works in most apps and websites.
- Microsoft Edge: built-in spelling/grammar in text boxes; Microsoft Editor extension for Chrome/Edge
(basic checks free with a Microsoft account; advanced refinements with Microsoft 365).
- Google Docs: Tab accepts a suggestion; Tools → Spelling and grammar for a full pass.
- Word: Editor pane on the Home tab (Word for the web / Microsoft 365 on Windows).
5. Optional macOS Quick Action (copies text + prompt, opens your AI app)
Shortcuts app → New Shortcut → Details: "Use as Quick Action", receive Text from Services/Quick Actions →
actions: Text ("Proofread using my house style…\n\n[Shortcut Input]") → Copy to Clipboard → Open App (your
AI app). Assign a key in the shortcut's details (Add Keyboard Shortcut). Then: select text → shortcut → Cmd+V → Enter.
```
## Limitations
- No real-time underlines in every app and web form from your AI assistant; you must copy text into it (built-in checkers cover only their own apps).
- No plagiarism checker or AI-text detector (Grammarly Pro includes both).
- Style guides aren't enforced across a team; each person must use the shared Project/instructions.
- AI assistants occasionally 'over-edit' — changing correct wording or, rarely, a fact — which is why the prompt requires a change list.
- Apple Writing Tools need Apple Intelligence on a supported device, language and region; Google Docs grammar covers six languages; Microsoft Editor's advanced refinements need Microsoft 365.
- Usage limits of your AI plan apply; heavy proofreading days can hit them.
## Risks
- **functionality_loss** (medium): Losing inline suggestions everywhere and plagiarism/AI detection can lead to more small errors in quick messages and no originality check on published content.
- **sensitive_data** (medium): Pasting contracts, HR or customer messages into an AI assistant sends them to that provider. Use a business plan whose data terms you've checked, or the on-device/built-in checkers for sensitive text. (Grammarly also processes what you type, so this is a change of processor, not a new category of risk.)
- **reliability** (low): A silent AI rewrite can change meaning. The change-list format and a quick read of the diff before sending mitigates this.
- **collaboration** (low): Team-wide consistency depends on people using the shared style guide; there are no admin controls or analytics.
## Verification checklist (run before cancelling)
- [ ] On 10 real pieces of your writing, the Proofread prompt caught every genuine error Grammarly flagged (style preferences excluded) and changed no facts, numbers or names.
- [ ] The prompt reliably outputs a numbered change list before the corrected text in your chosen assistant (test 5 times).
- [ ] Built-in checking is confirmed working where you write most: Google Docs suggestions on, Microsoft Editor visible in Word/Outlook, or Writing Tools available on your Apple device.
- [ ] The `;pf` snippet (or Quick Action/pinned Project) gets you from selected text to corrected text in under 30 seconds.
- [ ] You've confirmed nobody on the team needs Grammarly's plagiarism checker, AI detector or enforced team style guide — or you've kept seats only for those people.
- [ ] You've noted Grammarly's renewal date and cancel before it, keeping Grammarly Free if you still want basic underlines.
## Savings methodology
Savings = Grammarly paid seats × price per seat per month − incremental cost ($0, because the AI assistant and built-in checkers are already paid for or free). Dropping to Grammarly Free instead of cancelling keeps basic underlines at $0. If your Grammarly plan is billed annually, savings start at renewal.
Example: A freelance consultant paying for Grammarly Pro and ChatGPT Plus cancels Grammarly Pro and moves proofreading into a ChatGPT Project plus Google Docs/Word built-in checks. $12/mo → $0/mo. Assumptions: Grammarly Pro listed at $12/month on grammarly.com/plans on 2026-10-13 (the page did not state whether this figure requires annual billing; check your invoice). The ChatGPT subscription is already paid for other work, so it adds nothing. No need for plagiarism detection or an AI detector. Optionally keeps Grammarly Free ($0) for basic inline suggestions.
## Sources
- Grammarly plans: https://www.grammarly.com/plans (accessed 2026-10-13)
- Grammarly homepage: https://www.grammarly.com/ (accessed 2026-10-13)
- Google Docs — Check spelling and grammar: https://support.google.com/docs/answer/57859 (accessed 2026-10-13)
- Microsoft Editor: https://support.microsoft.com/en-us/office/microsoft-editor-checks-grammar-and-more-in-documents-mail-and-the-web-91ecbe1b-d021-4e9e-a82e-abc4cd7163d7 (accessed 2026-10-13)
- Apple — Use Writing Tools on Mac: https://support.apple.com/guide/mac-help/use-writing-tools-mchldcd6c260/mac (accessed 2026-10-13)
# Extract invoice data with AI into Google Sheets, with validation and human approval
> For low invoice volumes: use Gemini in Google Workspace or ChatGPT file upload with a strict extraction prompt (no-code), or an n8n flow from Gmail to an LLM to Sheets (automated). Either way, totals and duplicates are checked by rules, and a human approves every invoice before payment.
- Outcome: Get vendor, invoice number, dates, line items and totals from PDF invoices into a spreadsheet log, accurately enough to approve and pay — without a dedicated AP-automation subscription.
- Replaces: Zapier (task_specific)
- Uses: Google Workspace — Gmail (incoming invoices), Drive (PDF archive), Sheets (invoice log, validation formulas, approval column).; Google Gemini (optional) — Path A: Gemini in Sheets/Drive side panel on eligible Workspace plans, or the Gemini app, to read invoice PDFs.; ChatGPT (optional) — Path A alternative: upload PDFs and run the extraction prompt.; Claude (optional) — Path A alternative: upload PDFs and run the extraction prompt.; n8n (optional) — Path B: Gmail trigger → PDF text → Information Extractor (LLM, JSON schema) → validation → Sheets.
- Extra monthly cost: $5 — Path A (no-code, using an AI assistant or Gemini plan you already pay for) adds $0. Path B adds LLM API usage on top of an n8n instance you already run. $5/month is a conservative ILLUSTRATIVE allowance for ~100 invoices — we did not verify current per-token API prices today; check your provider's pricing page and the token count of a typical invoice.
- Difficulty: moderate; skill: basic; setup: Path A: 1–2 hours. Path B: 1 day including the parallel-run test.
- Evidence: editorial; illustrative example: true; last reviewed 2026-10-13
- Canonical: https://onelesstool.com/workflows/invoice-processing-with-ai
## Good fit for
- Small businesses receiving tens to roughly a hundred supplier invoices a month
- Bookkeepers who key invoices into a spreadsheet or accounting tool by hand today
- Teams already on Google Workspace with Gmail as the invoice inbox
- Anyone using a general-purpose automation tool plus a parser add-on just for invoices
## Not a fit for
- Hundreds or thousands of invoices a month, multi-entity AP, or three-way PO matching — dedicated AP automation is built for this
- Businesses that need audit-ready approval workflows, segregation of duties or e-invoicing compliance (e.g. mandated structured e-invoice formats) — a sheet doesn't provide these
- Organisations whose data policies don't allow financial documents to be sent to an external LLM provider
- Anyone hoping to remove human review — extraction errors must be caught before payment
## Before
Invoices arrive by email. Someone opens each PDF and types vendor, number, dates and amounts into a spreadsheet or accounting system — or pays for an AP-capture tool or a parser add-on chained into an automation platform to do it.
## After
Path A: once or twice a week, drag 5–10 invoice PDFs into ChatGPT, Claude or Gemini with a saved extraction prompt; paste the JSON/TSV into the Invoice log where formulas flag line-sum, total and duplicate problems. Path B: an n8n workflow watches a Gmail label, extracts PDF text, asks an LLM for schema-valid JSON, validates totals and duplicates in a Code node, appends to the same sheet as 'Awaiting approval', and emails the approver. In both paths nothing gets paid until a person has compared the row with the PDF and marked it Approved.
## Steps
### 1. Set up the Invoice log and a Gmail label
1. Create a Google Sheet **Invoice log** with the columns listed in the *Validation rules* artifact (A–P). Add the **O2** (normalised key) and **P2** (flags) formulas and fill down.
2. Add data validation on **K (Status)**: `Awaiting approval, Approved, Rejected, Blocked – duplicate, Paid`.
3. Conditional formatting: `=$P2<>""` → red fill.
4. In Gmail, create a filter that applies the label **AP/Inbox** to mail from your suppliers or sent to your invoices@ address. Save PDFs to a Drive folder **AP/2026** so the log can link to originals (column N).
### 2. Path A (no-code): extract with your AI assistant
Use whichever you already pay for:
- **ChatGPT or Claude:** start a Project (so the instructions persist), paste the *Invoice JSON schema* and the **Extract invoice to JSON** prompt as project instructions, then upload **5–10 PDFs per message**.
- **Gemini in Google Sheets** (on eligible Workspace editions — Business Starter does not include Gemini in Sheets per Google's pricing page): open the side panel in the Invoice log, reference the Drive PDFs and ask for a table using the same field list. Check that it read the right files.
Ask for the result as **JSON and as TSV**; paste the TSV into the log. The formulas then flag any line-sum, total or duplicate issues.
Scanned or photographed invoices: the assistant may misread digits. Mark them `HUMAN_CHECK` and compare every number.
### 3. Path B (automated): build the n8n workflow
Nodes, in order:
1. **Gmail Trigger** — poll every 15–60 min, filter by label `AP/Inbox`, unread only; enable downloading attachments.
2. **Filter** — keep only items with a PDF attachment.
3. **Extract From File** → *Extract From PDF* on the attachment's binary field. (Text-based PDFs only; n8n's docs don't mention OCR, so route scanned PDFs to a model that accepts images/PDFs directly, or to manual entry.)
4. **Information Extractor** with *Schema Type → Define using JSON Schema*, paste the schema artifact, connect your chat model (OpenAI, Anthropic, Gemini). Put the extraction rules from the prompt in the system message.
5. **Google Sheets → Get rows** from the Invoice log, then a **Code** node ("Run Once for All Items") that builds an `existingKeys` array — or simply rely on the sheet's DUPLICATE formula and keep this step simple.
6. **Code** node *Validate invoice* (artifact) — adds `flags`, `line_sum`, `norm_key`, `status`.
7. **Google Drive → Upload** the PDF to `AP/2026`; **Google Sheets → Append row** with the link and status `Awaiting approval` or `Blocked – duplicate`.
8. **Gmail → Send** the approver a summary with the flags and the sheet link.
9. Attach an **Error Workflow** so failures alert you instead of silently dropping invoices.
### 4. Validate every invoice with rules, not vibes
LLMs are good at reading invoices and bad at noticing their own mistakes. The *Validation rules* artifact catches the errors that cost money:
- **Line items sum to subtotal**, and **subtotal − discount + shipping + tax = total** (±0.01).
- **Duplicate invoice number** from the same vendor after normalising (`INV-00123` = `inv 123`) → blocked.
- **Near-duplicates** (same vendor and amount within 7 days, different number) → flagged; double-billing often looks like this.
- **Dates** in the future or more than a year old → flagged.
- **Bank details present on a new vendor or changed from last time** → verify by phone using a number you already have. Invoice fraud usually arrives as a legitimate-looking invoice with new bank details.
Anything flagged gets a human check against the PDF; a clean row still needs approval.
### 5. Human approval before any payment
Set a simple rule and write it at the top of the sheet: **nobody pays from an unapproved row.**
- The approver opens the PDF link (column N), compares vendor, number, total and due date, resolves flags with a note, then sets **Status = Approved** with their name and date.
- Payment is made in your bank or accounting software by someone other than the approver if you have two people (basic segregation of duties).
- After payment, set **Status = Paid**. Once a month, reconcile Paid rows against bank statements.
Use **Data → Protect sheets and ranges** so only approvers can edit columns K–M.
### 6. Parallel-run for one month, then decide
For one month, keep your current process and run the new one alongside it. Track per invoice: fields wrong, flags raised, minutes spent. Move to the new process only if **zero errors reached the 'Approved' state** and the time saved is real. Re-check the extraction prompt whenever a new supplier's layout causes repeated `HUMAN_CHECK` flags — add a one-line hint per problem supplier to the prompt.
## Prompts
### Extract invoice to JSON
Turn one or more invoice PDFs into schema-valid JSON without inventing values.
```text
You are an accounts-payable data-entry assistant. Extract data from the attached document(s) for {{company_name}}.
Return ONLY JSON that validates against this schema (one object per document; if several documents are attached, return a JSON array in the order given):
{{json_schema}}
Rules:
1. Copy values exactly as printed. Do not calculate, infer or 'fix' anything. If a value is not printed, use null (or 0 for tax_total only when the document clearly shows no tax).
2. Dates → YYYY-MM-DD. If the date format is ambiguous (e.g. 03/04/2026), assume {{date_convention}} and add the field to uncertain_fields.
3. Numbers → plain decimals without currency symbols or thousands separators.
4. invoice_number: keep prefixes, dashes and leading zeros exactly.
5. If it is not an invoice (statement, quote, receipt, marketing email), set document_type accordingly and still fill what you can.
6. Set bank_details_present to true if bank account / IBAN / routing details appear anywhere.
7. confidence = low if the scan is blurry, handwritten, cut off or totals don't appear.
No commentary outside the JSON.
```
### Convert extracted JSON to sheet rows
Produce paste-ready TSV in the Invoice log's column order (Path A).
```text
Convert this extracted invoice JSON into TSV rows for my Invoice log. Columns, in order: Received ({{received_date}}), Vendor, Invoice no, Invoice date, Due date, Currency, Subtotal, Tax, Total, Line sum (sum of line_items.amount, rounded to 2 decimals), Status (always "Awaiting approval"), Approver (blank), Approved on (blank), File link ({{file_link_placeholder}}).
{{extracted_json}}
Output only one TSV code block with no header row. Do not change any extracted value; if a field is null, leave the cell empty. After the code block, list any invoice where Line sum ≠ Subtotal or Subtotal + Tax ≠ Total.
```
### Second-look check on a flagged invoice
Re-read a flagged invoice field by field before a human decides.
```text
This invoice was flagged {{flags}} after extraction. Re-read the attached PDF carefully and compare it against the extracted JSON below, field by field.
{{extracted_json}}
Return a table: field | extracted value | value printed on the document (quote it exactly) | match (yes/no). Then say which flag is explained by an extraction error and which reflects a real problem on the invoice itself (e.g. the vendor's own arithmetic is wrong, or it's a duplicate). Do not recommend paying or not paying; just report what the document says.
```
## Templates and scripts
### Invoice JSON schema (schema)
```json
{
"$schema": "https://json-schema.org/draft/2020-12/schema",
"title": "ExtractedInvoice",
"type": "object",
"additionalProperties": false,
"required": ["document_type", "vendor_name", "invoice_number", "invoice_date", "currency",
"line_items", "subtotal", "tax_total", "total", "confidence", "uncertain_fields"],
"properties": {
"document_type": { "enum": ["invoice", "credit_note", "receipt", "statement", "quote", "not_an_invoice"] },
"vendor_name": { "type": "string", "description": "Legal name as printed on the document" },
"vendor_tax_id": { "type": ["string", "null"], "description": "VAT/GST/EIN etc. exactly as printed, else null" },
"vendor_address": { "type": ["string", "null"] },
"bill_to_name": { "type": ["string", "null"] },
"invoice_number": { "type": "string", "description": "Exactly as printed, including prefixes and leading zeros" },
"po_number": { "type": ["string", "null"] },
"invoice_date": { "type": "string", "format": "date", "description": "YYYY-MM-DD" },
"due_date": { "type": ["string", "null"], "format": "date" },
"payment_terms": { "type": ["string", "null"], "description": "e.g. 'Net 30' as printed" },
"currency": { "type": "string", "pattern": "^[A-Z]{3}$", "description": "ISO 4217, e.g. USD" },
"line_items": {
"type": "array",
"items": {
"type": "object",
"additionalProperties": false,
"required": ["description", "quantity", "unit_price", "amount"],
"properties": {
"description": { "type": "string" },
"quantity": { "type": ["number", "null"] },
"unit_price": { "type": ["number", "null"] },
"amount": { "type": "number", "description": "Line total before tax, as printed" },
"tax_rate": { "type": ["number", "null"], "description": "Percent, e.g. 20 for 20%" }
}
}
},
"subtotal": { "type": "number", "description": "Before tax, after line-level discounts" },
"discount_total": { "type": ["number", "null"], "description": "Document-level discount as a positive number" },
"shipping_total": { "type": ["number", "null"] },
"tax_total": { "type": "number", "description": "0 if no tax shown" },
"total": { "type": "number", "description": "Amount due as printed" },
"amount_paid": { "type": ["number", "null"] },
"bank_details_present": { "type": "boolean", "description": "true if the document shows bank/payment account details" },
"confidence": { "enum": ["high", "medium", "low"] },
"uncertain_fields": { "type": "array", "items": { "type": "string" },
"description": "Names of fields that were hard to read, inferred or missing" }
}
}
Conventions: numbers are plain decimals (1234.5, not "1,234.50"); credit notes use positive numbers with
document_type "credit_note"; never compute a value that isn't printed — put the field in uncertain_fields instead.
```
### Validation rules + Google Sheets formulas (checklist)
```
VALIDATION RULES — apply to every extracted invoice before it can be approved
Tolerance: 0.01 per amount (0.02 if more than 10 line items, to allow per-line rounding).
R1 Line sum | SUM(line_items.amount) = subtotal → else "LINE_SUM_MISMATCH"
R2 Total | subtotal − discount_total + shipping_total + tax_total = total → else "TOTAL_MISMATCH"
R3 Tax sanity | if line tax_rate present: tax_total ≈ Σ(amount × rate/100) → else "TAX_CHECK" (warning)
R4 Duplicate | same normalised vendor + same normalised invoice_number already in the log
| (normalise: upper-case, keep only letters/digits, drop zeros that lead a number) → "DUPLICATE" (block)
R5 Near-duplicate | same vendor + same total + invoice_date within 7 days, different number → "POSSIBLE_DUPLICATE"
R6 Dates | invoice_date not in the future and not > 365 days old; due_date ≥ invoice_date → "DATE_CHECK"
R7 Required | vendor_name, invoice_number, invoice_date, total present → else "MISSING_FIELD"
R8 Doc type | document_type = "invoice" (credit notes routed separately) → else "NOT_INVOICE"
R9 Confidence | confidence = "low" or uncertain_fields not empty → "HUMAN_CHECK"
R10 Bank details | bank_details_present = true AND vendor is new, or bank details differ from the vendor master
| → "VERIFY_BANK_DETAILS_BY_PHONE" (classic invoice-fraud pattern: call a known number, never the one on the invoice)
Google Sheets versions (Invoice log tab; columns: A Received, B Vendor, C Invoice no, D Invoice date,
E Due date, F Currency, G Subtotal, H Tax, I Total, J Line sum, K Status, L Approver, M Approved on, N File link,
O Norm key, P Flags):
O2 =UPPER(REGEXREPLACE(B2,"[^A-Za-z0-9]",""))&"|"®EXREPLACE(UPPER(REGEXREPLACE(C2,"[^A-Za-z0-9]","")),"(^|[A-Z])0+([0-9])","$1$2")
P2 =TEXTJOIN(", ",TRUE,
IF(ABS(J2-G2)>0.01,"LINE_SUM_MISMATCH",""),
IF(ABS(G2+H2-I2)>0.01,"TOTAL_MISMATCH",""),
IF(COUNTIF($O$2:$O,O2)>1,"DUPLICATE",""),
IF(D2>TODAY(),"DATE_CHECK",""))
(Adjust P2's TOTAL check if you track discount/shipping columns.)
Conditional formatting on A2:P: custom formula =$P2<>"" → red fill.
APPROVAL RULE: Status may only change to "Approved" when P is empty (or each flag is resolved with a note),
and the approver has opened the original PDF. Payment happens in your bank/accounting system, never from this sheet.
```
### n8n Code node: validate invoice (script)
```javascript
// n8n Code node — "Validate invoice" (mode: Run Once for Each Item)
// Input: the Information Extractor's structured result (n8n puts it under `output`; we also accept a top-level object)
// plus `existingKeys` (array of normalised "VENDOR|INVOICENO" keys) merged in from a Google Sheets lookup step.
const inv = $json.output ?? $json;
const existingKeys = $json.existingKeys ?? [];
const flags = [];
const tol = (inv.line_items?.length ?? 0) > 10 ? 0.02 : 0.01;
const num = (v) => (typeof v === 'number' && isFinite(v) ? v : 0);
const close = (a, b) => Math.abs(a - b) <= tol + 1e-9;
for (const f of ['vendor_name', 'invoice_number', 'invoice_date', 'total']) {
if (inv[f] === null || inv[f] === undefined || inv[f] === '') flags.push(`MISSING_FIELD:${f}`);
}
if (inv.document_type !== 'invoice') flags.push(`NOT_INVOICE:${inv.document_type}`);
const lineSum = (inv.line_items ?? []).reduce((s, l) => s + num(l.amount), 0);
if ((inv.line_items ?? []).length && !close(lineSum, num(inv.subtotal))) flags.push('LINE_SUM_MISMATCH');
const expectedTotal = num(inv.subtotal) - num(inv.discount_total) + num(inv.shipping_total) + num(inv.tax_total);
if (!close(expectedTotal, num(inv.total))) flags.push('TOTAL_MISMATCH');
const today = new Date(); today.setHours(0, 0, 0, 0);
const invDate = new Date(inv.invoice_date);
if (isNaN(invDate) || invDate > today || (today - invDate) / 86400000 > 365) flags.push('DATE_CHECK');
if (inv.due_date && new Date(inv.due_date) < invDate) flags.push('DATE_CHECK:due_before_invoice');
const norm = (s) => String(s ?? '').toUpperCase().replace(/[^A-Z0-9]/g, '');
const key = `${norm(inv.vendor_name)}|${norm(inv.invoice_number).replace(/(^|[A-Z])0+(?=\d)/g, '$1')}`; // INV-00123 -> INV123
if (existingKeys.includes(key)) flags.push('DUPLICATE');
if (inv.confidence === 'low' || (inv.uncertain_fields ?? []).length) flags.push('HUMAN_CHECK');
if (inv.bank_details_present) flags.push('VERIFY_BANK_DETAILS_BY_PHONE_IF_NEW_OR_CHANGED');
return {
json: {
...inv,
line_sum: Math.round(lineSum * 100) / 100,
norm_key: key,
flags: flags.join(', '),
status: flags.includes('DUPLICATE') ? 'Blocked – duplicate' : 'Awaiting approval',
},
};
```
## Limitations
- Extraction errors happen, especially on scans, photos, handwriting, multi-page tables and unusual layouts. Human review before payment is mandatory.
- n8n's Extract From File node reads text from PDFs; its docs don't mention OCR, so scanned invoices need a vision-capable model or manual entry.
- No three-way matching against purchase orders and goods receipts, no approval chains by amount, no vendor master management.
- Doesn't post bills into accounting software by itself; that needs another integration step or manual entry.
- Path A is still manual (upload, paste); it saves typing, not the whole process.
- Gemini features in Sheets/Drive depend on your Google Workspace edition or Google AI plan.
## Risks
- **sensitive_data** (high): Invoices contain bank details, amounts and personal data. OpenAI states it does not train on API, ChatGPT Business or Enterprise data by default; Anthropic states it does not train on commercial-product (Team, Enterprise, API) inputs by default. Consumer plans can differ — check your plan's data settings and your contracts before uploading supplier invoices.
- **reliability** (high): A misread digit or a hallucinated total can lead to an over- or under-payment. Rule-based validation catches arithmetic and duplicates, not a wrong-but-consistent number — the human check against the PDF is the control.
- **security** (medium): Automating the inbox can speed up invoice fraud (fake invoices, changed bank details) if people trust the pipeline. Keep the bank-details verification rule and never pay from a link or number on the invoice itself.
- **compliance** (medium): Tax rules may require retaining original invoices and, in some countries, structured e-invoices. Keep the original PDFs in Drive and check local requirements; the extracted sheet is not a substitute for the original.
- **functionality_loss** (medium): Compared with dedicated AP tools you lose supplier portals, automatic accounting sync, approval routing and audit trails.
## Verification checklist (run before cancelling)
- [ ] On a test set of 30 past invoices (including at least 5 scans and 5 multi-page invoices), every field matches the PDF on at least 28, and every error was caught by a flag or by the human check.
- [ ] Feeding the same invoice twice produces DUPLICATE and the second copy is blocked; 'INV-00123' and 'inv 123' from the same vendor are treated as the same number.
- [ ] An invoice with a deliberately wrong total (edit a test PDF) is flagged TOTAL_MISMATCH.
- [ ] A credit note and a supplier statement are recognised as such and not logged as invoices to pay.
- [ ] Path B: the n8n workflow processed every test email in the AP/Inbox label exactly once over 5 days, and the Error Workflow alerted on a deliberately broken run.
- [ ] Only approvers can edit the Status/Approver columns, and no payment in the parallel month was made from a row without Status = Approved.
- [ ] You have confirmed in writing which AI plan/API is used and that its data terms are acceptable for supplier financial data.
## Savings methodology
Savings = what you pay today for invoice capture (an AP-capture subscription or a parser add-on plus the automation tasks it consumes) − incremental cost of the new path (Path A: $0 with an AI plan you already have; Path B: LLM API usage). Time saved on data entry is real but not counted in dollars. Human review time remains in both cases.
Example (illustrative): A small agency processing ~60 supplier invoices a month with a paid invoice-capture add-on, switching to Path B on an n8n instance it already runs. $30/mo → $5/mo. Assumptions: ILLUSTRATIVE: $30/month for the current invoice-capture tool is an assumed figure, not a verified vendor price — use your own invoice. ILLUSTRATIVE: $5/month is a conservative allowance for LLM API usage at ~60 invoices; actual cost depends on model and pages per invoice and was not verified against a price list today. n8n and Google Workspace are already paid for; no new subscriptions. Human review (~1 minute per invoice) continues in both setups.
## Sources
- n8n docs — Information Extractor node: https://docs.n8n.io/integrations/builtin/cluster-nodes/root-nodes/n8n-nodes-langchain.information-extractor/ (accessed 2026-10-13)
- n8n docs — Extract From File: https://docs.n8n.io/integrations/builtin/core-nodes/n8n-nodes-base.extractfromfile/ (accessed 2026-10-13)
- n8n docs — Gmail Trigger: https://docs.n8n.io/integrations/builtin/trigger-nodes/n8n-nodes-base.gmailtrigger/ (accessed 2026-10-13)
- Gemini in Google Sheets: https://support.google.com/docs/answer/14218565 (accessed 2026-10-13)
- Google Workspace pricing / plan comparison: https://workspace.google.com/pricing (accessed 2026-10-13)
- OpenAI — Business data privacy: https://openai.com/business-data/ (accessed 2026-10-13)
- Anthropic — Is my data used for model training?: https://privacy.claude.com/en/articles/7996868-is-my-data-used-for-model-training (accessed 2026-10-13)
# Stop paying for Otter: use the transcription and AI notes already in Meet, Teams or Zoom
> Many business plans already include meeting transcripts and AI notes: Google Meet transcripts and 'Take notes for me' on eligible Workspace editions, Teams transcription (with AI notes via Teams Premium or Microsoft 365 Copilot), and Zoom's built-in meeting summaries on eligible paid plans. Check what you have, fill gaps with a transcript-to-action-items prompt, and handle recording consent properly.
- Outcome: Get a transcript, a summary and a clear list of action items from every meeting without a separate note-taker subscription.
- Replaces: Otter.ai (full)
- Uses: Google Workspace (optional) — Google Meet transcripts (Business Standard and up, among others) and Gemini 'Take notes for me' on eligible editions.; Microsoft 365 (optional) — Teams live transcription and Recap (availability depends on licence and admin policy).; Microsoft 365 Copilot (optional) — AI-generated notes, tasks and Q&A over Teams meetings (intelligent recap is part of the Copilot licence; also in Teams Premium).; Claude (optional) — Turn any transcript file into decisions and action items with the prompts below.; ChatGPT (optional) — Alternative for transcript-to-action-items.; Google Gemini (optional) — Alternative for transcript-to-action-items, already inside Google Docs on eligible plans.
- Extra monthly cost: $0 — $0 if your meeting platform plan already includes transcription/AI notes (see the eligibility matrix) and you already pay for an AI assistant for the gaps. If your plan doesn't include AI notes (e.g. Google Workspace Business Starter, Teams without Premium or Copilot), upgrading may cost more than Otter — do the comparison before cancelling.
- Difficulty: easy; skill: none; setup: 30–60 minutes (check plan eligibility, admin settings, invite template)
- Evidence: editorial; illustrative example: false; last reviewed 2026-10-13
- Canonical: https://onelesstool.com/workflows/meeting-notes-with-built-in-ai
## Good fit for
- Teams that meet mostly on one platform (Meet, Teams or Zoom) and already pay for an eligible business plan
- People using Otter mainly for summaries and action items of internal meetings
- Organisations that prefer meeting data to stay inside their existing Google or Microsoft tenant
## Not a fit for
- People who meet across many platforms or join other companies' meetings and need one bot that works everywhere
- Workflows built on Otter-specific features: shared searchable conversation library across all meetings, CRM sync, in-person recording from a phone app, importing audio files at volume
- Teams on plans without built-in AI notes (e.g. Google Workspace Business Starter, Microsoft 365 without Teams Premium or Copilot) where upgrading costs more than the note-taker
- Meetings where recording or transcription isn't permitted or consent can't be obtained
## Before
An Otter seat per person joins meetings (via calendar integration or a bot), records and transcribes them, writes a summary and action items, and keeps a searchable library of conversations.
## After
The meeting platform does the transcription natively: Meet saves a transcript Doc and Gemini notes to the organizer's Drive; Teams shows the transcript (and AI notes with Premium/Copilot) in Recap; Zoom produces a meeting summary on eligible paid plans. For anything the platform doesn't summarise well, the transcript file goes through a saved 'transcript → action items' prompt in your existing AI assistant. Consent wording is in every invite.
## Steps
### 1. Check exactly what your plan includes
Use the **Eligibility matrix** artifact, then confirm in your own account — vendors change these often.
- **Google:** start a test meeting as organizer. If the meeting tools offer **Transcripts** and **Take notes for me**, you have them. Transcripts are listed for Business Standard and above (not Business Starter). 'Take notes for me' supports 8 spoken languages, one per meeting.
- **Microsoft:** in a Teams meeting, open **More** and look for the **Record and transcribe** options. If it's missing, your admin's meeting policy or licence doesn't allow it — **check your plan with IT**. AI notes and tasks in Recap need **Teams Premium** or a **Microsoft 365 Copilot** licence.
- **Zoom:** in settings, look for **meeting summary** and **transcription** options. Zoom says these (formerly 'AI Companion') remain included with **eligible paid plans** — **check your plan**. Free Basic accounts get a small monthly allowance.
If none applies, compare the cost of upgrading against Otter before going further.
### 2. Turn the features on (admin + user)
- **Google Workspace admin:** make sure Meet transcripts and Gemini features are enabled for the right organisational units. Users then click **Take notes for me** (or set it to start automatically in the meeting options) and choose who receives notes under **Send notes to**.
- **Microsoft 365 / Teams admin:** in the Teams admin center meeting policy, allow transcription (and recording if you want it). For Copilot, organizers choose **During and after the meeting** so Copilot can use the transcript later.
- **Zoom admin:** enable meeting summaries and transcription at account or group level, and decide who summaries are sent to.
Test with a colleague: run a 5-minute meeting and confirm the transcript/notes land where you expect (Drive, OneDrive/Recap, Zoom summary email).
### 3. Put consent in the workflow, not in someone's memory
Work through the **Consent checklist** artifact. The short version:
- US federal law is one-party consent, but **several states require all-party consent** (the Digital Media Law Project lists California, Florida, Washington, Pennsylvania, Maryland and others). With mixed locations, get everyone's consent.
- In the EU/UK, transcripts are personal data under GDPR: tell people why, who sees it and how long you keep it.
- Add a consent line to your **default calendar invite**, say it out loud at the start of external calls, and **don't transcribe** HR, health, legal or performance conversations by default.
All three platforms show participants that transcription is running; Google also lets admins require explicit participant consent before transcription starts.
### 4. Fill the gaps with a saved prompt
Built-in notes are generic. For client calls and project meetings, download the transcript (Meet: the transcript Doc; Teams: Recap → Transcript → Download .docx; Zoom: the transcript file) and run the **Transcript → decisions and action items** prompt in Claude, ChatGPT or Gemini.
Save the prompt as a Claude/ChatGPT Project instruction or a Gemini Gem so it's one paste per meeting. The prompt forces **owners and due dates**, flags **unclear** commitments instead of inventing them, and quotes the transcript line for each action so you can verify it.
### 5. Replace Otter's library with a simple filing rule
Otter's searchable library is the hardest thing to replace. A good-enough version:
- Meet and Teams already file transcripts in the organizer's Drive/OneDrive and link them to the calendar event — search there.
- Paste each approved action-item summary into one running doc per client/project (e.g. `Client – Meeting log`), newest first. That doc becomes the searchable history.
- Before cancelling Otter, **export** the conversations you need to keep (check Otter's export options for your plan) and store them in that folder with the same retention rule.
### 6. Run both for two weeks, then cancel
For two weeks, let Otter and the built-in feature both run on the same internal meetings (with consent). Compare action items side by side. Cancel when the built-in notes plus the prompt catch every action item Otter caught on at least 5 meetings, and your exports are done.
## Prompts
### Transcript → decisions and action items
Turn a raw transcript into a verified action list with owners and dates.
```text
Below is the transcript of a {{meeting_type}} held on {{meeting_date}}. Attendees: {{attendees}}. Our team: {{our_team}}.
{{transcript}}
Produce:
1. **Decisions made** — one line each, with the transcript quote that shows the decision.
2. **Action items** — table: Action | Owner | Due date | Quote from transcript. Use only owners and dates actually stated. If an owner or date is missing or ambiguous, write "UNCLEAR" — do not guess.
3. **Open questions / parked items**.
4. **Risks or concerns raised** by anyone outside our team.
5. A follow-up email to {{recipient}} (max 150 words) summarising decisions and actions.
Transcripts contain recognition errors: if a name, number or product term looks wrong, flag it in a final "Check these" list rather than correcting it silently. Do not include anything said after someone asked to go off the record.
```
### Compare built-in notes with a transcript
Check what the platform's AI notes missed during the parallel run.
```text
Here are the AI-generated notes from {{platform}} and the full transcript of the same meeting.
AI notes:
{{platform_notes}}
Transcript:
{{transcript}}
List: (1) action items or decisions in the transcript that the notes missed, (2) statements in the notes not supported by the transcript, (3) items attributed to the wrong person. Quote the transcript for each. End with a one-line verdict: are these notes reliable enough to send without edits?
```
### Client meeting log entry
Create a consistent entry for a running per-client meeting log (replacing Otter's library).
```text
Create a meeting-log entry for {{client_name}} from these approved notes:
{{approved_notes}}
Format exactly:
### {{meeting_date}} – {{meeting_title}}
**Attendees:** …
**Decisions:** bullet list
**Actions:** - [ ] Action — Owner — Due
**Next meeting:** date or "not set"
**Tags:** up to 5 keywords for search (products, projects, topics)
Use only information in the notes.
```
## Templates and scripts
### Eligibility matrix: what Meet, Teams and Zoom include (checklist)
```
WHICH BUILT-IN FEATURE DO YOU ALREADY HAVE? (checked 2026-10-13 — vendors change this often; confirm in your admin console)
GOOGLE MEET (Google Workspace)
Transcripts (verbatim, saved as a Google Doc in the organizer's Drive "Google Meet" folder, linked to the Calendar event)
Editions listed by Google: Business Standard, Business Plus, Enterprise Starter, Enterprise Standard, Enterprise Plus,
Teaching and Learning Upgrade, Education Plus, Workspace Individual.
→ NOT listed: Business Starter.
"Take notes for me" (Gemini AI summary + next steps, saved as a Doc in the organizer's Drive and attached to the event)
Requires an eligible Workspace edition or Google AI plan; the ORGANIZER's edition must include it.
Google's plan comparison shows Gemini in Meet in Business Standard and Plus, not in Business Starter.
Spoken languages: English, French, German, Italian, Japanese, Korean, Portuguese, Spanish — one language per meeting.
Who gets notes: host setting "Send notes to" (all invited guests incl. external / internal invitees / hosts only).
MICROSOFT TEAMS (Microsoft 365)
Live transcription: participants are notified; transcript appears in the calendar event and the meeting chat's
Recap tab, downloadable as .docx/.vtt; stored in the organizer's OneDrive. Availability depends on your license
AND your admin's meeting policy → CHECK YOUR PLAN / ask IT if the Transcribe button is missing.
Intelligent recap with AI-generated notes and tasks: a Teams Premium feature (per Microsoft's Teams Premium
licensing page) and also part of the Microsoft 365 Copilot license (per Microsoft's Copilot-in-Teams support page).
Copilot in Teams meetings: transcription must be on for Copilot to work during and after the meeting
(organizer setting "During and after the meeting").
ZOOM WORKPLACE
Meeting summaries and transcription (formerly branded "AI Companion"): Zoom's June 2026 blog says these remain
"included with eligible paid plans" and are now simply named by what they do. → CHECK YOUR PLAN for eligibility.
Free Basic accounts: limited AI meeting summaries (Zoom's page lists 3 hosted meetings/month) and 40-minute meetings.
ZoomMate (agentic assistant, custom summary templates at scale, workflows) is a separate paid product.
NEITHER? Record with the platform's own recording (with consent), download the transcript/captions file, and use the
"Transcript → action items" prompt in Claude/ChatGPT/Gemini. That covers summaries, not live note-taking.
```
### Recording & transcription consent checklist (checklist)
```
RECORDING & TRANSCRIPTION CONSENT CHECKLIST (general information, not legal advice)
Know the rule that applies
[ ] US federal law (18 U.S.C. 2511(2)(d)) is ONE-party consent, but several states require ALL parties to consent —
the Digital Media Law Project lists California, Connecticut, Florida, Illinois, Maryland, Massachusetts (bans
secret recording), Montana, New Hampshire, Pennsylvania and Washington. With participants in several states,
follow the strictest rule: get everyone's consent.
[ ] EU/UK: recordings and transcripts are personal data under GDPR/UK GDPR — you need a lawful basis, a clear notice
(purpose, retention, who sees it) and must honour access/erasure requests. Employee monitoring rules may apply.
[ ] Other countries: check local law; when unsure, get explicit consent from everyone.
Before the meeting
[ ] Calendar invite says: "This meeting will be transcribed and AI-summarised for note-taking. Tell the organizer if
you'd prefer not." — include it in your default invite template
[ ] External meetings: ask at the start, out loud, and note the answer in the agenda doc
[ ] Sensitive topics (HR, health, legal, M&A, performance reviews): DON'T transcribe by default
During
[ ] Platform indicators are on (Meet/Teams/Zoom all show participants that transcription/notes are running)
[ ] If anyone objects, stop transcription and take manual notes
After
[ ] Restrict who receives auto-generated notes (Meet: "Send notes to"; Teams: transcript download policy)
[ ] Review AI notes before forwarding — they can misattribute statements
[ ] Retention: set a deletion rule (e.g. delete raw transcripts after 90 days; keep the approved action-item summary)
[ ] Pasting transcripts into a separate AI assistant? Use a business plan whose data terms you've checked, and
remove anything you wouldn't put in an email to that vendor
```
## Limitations
- Each platform only covers its own meetings; a dedicated note-taker works across Zoom, Meet and Teams and in-person recordings.
- Plan eligibility varies and changes: Google Workspace Business Starter lacks transcripts; Teams AI notes need Teams Premium or Microsoft 365 Copilot; Zoom features depend on an eligible paid plan.
- Google 'Take notes for me' handles one spoken language per meeting and a fixed list of languages.
- No cross-meeting searchable library like Otter's; you rely on Drive/OneDrive search and your own meeting logs.
- Transcription errors with names, jargon and accents carry into summaries; AI notes can misattribute statements.
- Notes usually land with the organizer; when you're a guest in someone else's meeting you may get nothing unless they share.
## Risks
- **compliance** (high): Recording or transcribing without the consent required by law (all-party consent in several US states; GDPR transparency in the EU/UK) can be unlawful. Put consent wording in invites, ask at the start of external calls and stop if anyone objects.
- **sensitive_data** (medium): Transcripts capture everything said, including confidential and personal information, and may be emailed to all invitees (including external guests) depending on settings. Restrict 'Send notes to' and set retention rules.
- **functionality_loss** (medium): Losing a cross-platform note-taker, CRM integration and a searchable conversation library.
- **data_portability** (low): Export the Otter conversations you need before cancelling; after cancellation, access may be limited.
## Verification checklist (run before cancelling)
- [ ] In a test meeting on your actual plan, the transcript (and AI notes, if expected) were generated and saved where you expected within 30 minutes of the meeting ending.
- [ ] On at least 5 real meetings run in parallel with Otter, the built-in notes plus the action-items prompt captured every action item Otter captured.
- [ ] Notes/transcripts reached only the intended recipients (test with an external guest invite to confirm external sharing behaviour).
- [ ] Your default calendar invite contains consent wording, and the team knows to stop transcription if someone objects.
- [ ] A retention/deletion rule for raw transcripts is set (admin policy or a calendar reminder) and documented.
- [ ] All Otter conversations you must keep have been exported and filed before the subscription ends.
## Savings methodology
Savings = Otter seats × price per seat per month − any upgrade needed to unlock built-in features (often $0 if you're already on an eligible plan). If you'd have to upgrade every user's meeting-platform plan or buy Teams Premium/Copilot just for notes, include that cost — it can exceed Otter's price.
Example: A 3-person consultancy on Google Workspace Business Standard paying for Otter Pro on monthly billing, switching to Meet transcripts and 'Take notes for me'. $50.97/mo → $0/mo. Assumptions: Otter Pro at $16.99 per user/month on monthly billing ($8.33/user/month billed annually), per otter.ai/pricing on 2026-10-13: 3 × $16.99 = $50.97. Business Standard includes Meet transcripts (Google's transcript help page) and Gemini in Meet (Google Workspace plan comparison) — no upgrade needed. They already pay for an AI assistant used for the action-items prompt. If they were on annual Otter billing the saving would be 3 × $8.33 = $24.99/month.
## Sources
- Google Meet — Use transcripts: https://support.google.com/meet/answer/12849897 (accessed 2026-10-13)
- Google Meet — Take notes for me: https://support.google.com/meet/answer/14754931 (accessed 2026-10-13)
- Google Workspace plan comparison: https://workspace.google.com/pricing (accessed 2026-10-13)
- Microsoft — Live transcription in Teams meetings: https://support.microsoft.com/en-us/office/view-live-transcription-in-microsoft-teams-meetings-dc1a8f23-2e20-4684-885e-2152e06a4a8b (accessed 2026-10-13)
- Microsoft Learn — Teams Premium licensing: https://learn.microsoft.com/en-us/microsoftteams/teams-add-on-licensing/licensing-enhance-teams (accessed 2026-10-13)
- Microsoft — Copilot in Teams: https://support.microsoft.com/en-us/copilot-teams (accessed 2026-10-13)
- Zoom blog — Meet ZoomMate (June 24, 2026): https://www.zoom.com/en/blog/zoom-ai-companion-zoommate/ (accessed 2026-10-13)
- Zoom — ZoomMate product page: https://www.zoom.com/en/products/ai-assistant/ (accessed 2026-10-13)
- Otter.ai pricing: https://otter.ai/pricing (accessed 2026-10-13)
- Digital Media Law Project — Recording phone calls and conversations: https://www.dmlp.org/legal-guide/recording-phone-calls-and-conversations (accessed 2026-10-13)
# Research ~500 target accounts a month with an AI assistant and Google Sheets
> Replace the account-research columns you'd build in Clay with batched, cited web research in Claude or ChatGPT, a strict JSON output schema, a scoring rubric and a 10% hallucination spot-check. It does not replace verified contact data — you still need Apollo, ZoomInfo or similar for emails and phones.
- Outcome: Each month, turn a list of ~500 target companies into a qualified, cited, de-duplicated account list with a fit score and a personalisation angle for the best ones.
- Replaces: Clay (partial)
- Uses: Claude (optional) — Web search / Research mode with citations to research each batch and return JSON.; ChatGPT (optional) — Alternative: web search or deep research with citations to research each batch.; Google Workspace — Google Sheets as the master account list, de-dup formulas and QA log.; Apollo.io (optional) — Verified business emails/phones for Tier A accounts only. Any licensed contact-data provider works (ZoomInfo, etc.).
- Extra monthly cost: $0 — $0 extra if you already pay for Claude or ChatGPT with web search and already have a contact-data source. Usage limits on research features vary by plan, so heavy months may hit caps. Contact data is NOT included: if Clay was also your email-finding tool, budget for a contact-data provider separately.
- Difficulty: moderate; skill: basic; setup: 2–3 hours (rubric, prompt tuning on 2 test batches, sheet setup)
- Evidence: editorial; illustrative example: true; last reviewed 2026-10-13
- Canonical: https://onelesstool.com/workflows/prospect-research-with-ai-assistant
## Good fit for
- Founders, SDRs and small sales teams researching tens to a few hundred accounts per week
- Account-based outreach where a cited 'why now' matters more than list size
- Teams that used Clay mainly for AI research and scoring columns, not for waterfall contact enrichment
- Anyone who can spend a few hours a week on batched research and spot-checks
## Not a fit for
- Teams that need verified emails, mobile numbers or firmographics at scale — that's a contact-data product, not an AI assistant
- Thousands of accounts per month or fully automated enrichment on a trigger — manual batching doesn't scale; consider an API pipeline or keep Clay
- Workflows that depend on scraping LinkedIn or other sites whose terms forbid automated collection
- Anyone not willing to spot-check sources — unchecked AI research will contain fabricated or misattributed facts
## Before
A Clay workbook imports a company list, runs enrichment and AI research columns per row (company summary, recent news, hiring signals, a scoring formula, an AI-written first line) and pushes results to a CRM or sequencer. It's fast and automated but billed in credits/actions, and the bill grows with every column and re-run.
## After
A Google Sheet holds the master list. Each working session you paste 10–25 new, de-duplicated companies into a saved research prompt in Claude or ChatGPT with web search on. The assistant returns strict JSON (one object per company, every claim with a source URL), which you paste into the sheet. Formulas flag duplicates and score totals; you open the sources for 10% of rows (and every Tier A) before trusting the batch. Only Tier A accounts go to Apollo/ZoomInfo for verified contacts.
## Steps
### 1. Write your rubric before you research anything
Copy the **Qualification rubric** artifact and replace the bracketed descriptions with your real ICP: core industries, employee sweet spot, the signals that actually predicted past wins.
Test it on **10 past customers and 10 lost deals**. If the rubric doesn't rank most past customers above the losses, fix the rubric now — no prompt will rescue a vague ICP.
Keep `reachability` partly human: warm paths come from *your* CRM and network, not from the AI.
### 2. Set up the master sheet
Create a Google Sheet with tabs **Input**, **Accounts**, **Manual**, **QA**.
- **Input:** `Company name | Domain | Normalised domain | Already researched?` with the formulas from the *De-duplication and QA* artifact.
- **Accounts:** one column per field in the JSON schema, plus `Batch ID`, `Researched on`, `QA result`, `Owner`, `Status`.
- Use **Data → Data validation** dropdowns for `tier` and `Status` (`New / Checked / Contacted / Disqualified`).
To paste JSON into columns, ask the assistant to also output the batch as **TSV** in a second code block (the batch prompt below does this), then paste into the first empty row of Accounts.
### 3. Run research in batches of 10–25
In Claude (turn on **web search**, or use **Research** on Pro/Max/Team/Enterprise) or ChatGPT (web search or **deep research**), paste the **Batch account research** prompt with 10–25 companies.
Why small batches:
- Quality drops and the model starts skipping or merging companies in long lists.
- Research modes consume plan limits faster; Anthropic notes Research uses the same limits as normal chats but can use them up faster.
- A bad batch of 20 is cheap to re-run; a bad batch of 200 isn't.
**Throughput reality:** 500 accounts ≈ 25 batches of 20. At 15–30 minutes per batch including QA, budget **8–20 hours a month**. Save the prompt in a Claude Project or ChatGPT Project/custom GPT so every batch uses identical instructions.
### 4. Validate structure, de-duplicate, then spot-check sources
Follow the *De-duplication and QA* artifact for every batch:
1. **Structure:** row count in = out; `fit_total` = sum of scores; every row has ≥1 source; no emails/phones.
2. **Identity:** if `resolved_domain` ≠ input domain, the model may have researched a namesake — move to **Manual**.
3. **Spot-check 10%** (minimum 3 rows, plus *every* Tier A): open each source URL and confirm the claim is on the page and the date matches. Log PASS/FAIL in **QA**.
4. If more than 1 in 10 fails, re-run the batch through the **Verify claims** prompt or redo it — don't fix rows by hand and trust the rest.
The rule *“no URL, no claim”* is what makes this usable: an uncited signal is capped at 1 point by the rubric.
### 5. Get contact data the compliant way — only for Tier A
An AI assistant **does not provide verified contact data**. Emails it 'finds' are often guessed patterns. For Tier A accounts only:
- Look up the buyer in **Apollo, ZoomInfo or another licensed provider**, or use contact details the company publishes for business enquiries.
- **Do not scrape LinkedIn.** LinkedIn's User Agreement (section 8.2) prohibits scraping or copying profiles and using bots or unauthorised automated methods. Read profiles manually if you must; don't automate it.
- **GDPR (EU/UK contacts):** document your lawful basis (usually legitimate interests) and give the Article 14 notice — who you are, why you have their data, where it came from, their rights — **at first contact** at the latest.
- **CAN-SPAM (US):** applies to B2B email too. Accurate headers and subject, a postal address, a clear opt-out honoured within 10 business days. Other countries (e.g. Canada's CASL) can be stricter — check where your recipients are.
### 6. Write openers from cited signals and close the loop
Use the **Personalisation from cited signals** prompt on Tier A rows only. It must quote the signal and URL it used, so you can check it in seconds.
After a month, compare reply and meeting rates by tier. If Tier B converts like Tier A, your rubric is too strict; if Tier A doesn't convert, re-weight `need_signal` and `timing`. Then decide what to cancel: if you only used Clay's research columns, downgrade it; if you also relied on its contact waterfall, keep or replace that part with a contact-data subscription.
## Prompts
### Batch account research (strict JSON)
Research 10–25 companies with citations and score them against your rubric.
```text
You are a B2B research analyst. Use web search. Research each company below for a seller of {{our_offer}} whose ideal customer is: {{icp_description}}.
Companies (name, domain):
{{company_list}}
Scoring rubric (score each 0–3):
{{rubric}}
Rules:
1. Every factual claim must come from a page you actually opened. Put its URL in source_url / sources. If you cannot find a source, write "unknown" and add the gap to `unknowns`. Never guess employee counts, funding or dates.
2. Prefer the company's own site, press releases, job boards and reputable news from the last 12 months. Today is {{today}}; ignore signals older than 18 months.
3. If a name is ambiguous, research the company at the given domain; set match_confidence accordingly. If you can't confirm it's the same company, use "low" or "not_found" and don't score it.
4. Do NOT output personal email addresses, phone numbers or home locations. Do not use LinkedIn content obtained by scraping.
5. reachability: score only 2 (buyer role publicly identifiable) or 1 (unclear); I'll add warm paths myself.
6. fit_total must equal the sum of fit_scores. Tier: A 11–15 with need_signal ≥2; B 7–10; C 3–6; otherwise disqualified with a reason.
Output exactly two code blocks and nothing else:
(a) a JSON array that validates against this schema:
{{json_schema}}
(b) the same rows as TSV with a header row, columns in schema order, signals flattened as "type: claim (date) URL" joined by " | ".
Return one object per input company, in input order, even if not found.
```
### Verify claims (second pass)
Have the assistant re-check a batch whose spot-check failed, claim by claim.
```text
Act as a fact-checker. For each row in the JSON below, open every source_url and check whether the page explicitly supports the claim, for the right company and date.
{{batch_json}}
Return a table with columns: input_name | claim | source_url | verdict (SUPPORTED / NOT FOUND ON PAGE / WRONG COMPANY / WRONG DATE / PAGE UNAVAILABLE) | corrected claim with new URL, or "remove".
Then return the corrected JSON array with unsupported claims removed, fit_scores lowered where a removed claim was the basis (cap uncited criteria at 1), fit_total and tier recomputed. Do not add new claims without a URL.
```
### Personalisation from cited signals
Draft a first line for Tier A accounts that a human can verify in seconds.
```text
Write a first line for a cold email from {{sender_name}} at {{our_company}} to the {{buyer_role}} at {{company_name}}.
Use ONLY these cited signals:
{{signals_with_urls}}
Our relevant proof point: {{proof_point}}
Rules: max 30 words; reference exactly one signal factually (no flattery, no 'I noticed', no invented details); no claims about the person. Give 3 options, and under each write the signal and URL it relies on.
```
## Templates and scripts
### JSON output schema for a research batch (schema)
```json
{
"$schema": "https://json-schema.org/draft/2020-12/schema",
"title": "AccountResearchBatch",
"type": "array",
"items": {
"type": "object",
"additionalProperties": false,
"required": ["input_name", "input_domain", "resolved_domain", "match_confidence",
"what_they_do", "industry", "hq_country", "employee_range",
"signals", "fit_scores", "fit_total", "tier", "personalization_hook",
"unknowns", "sources"],
"properties": {
"input_name": { "type": "string", "description": "Exactly as given in the input list" },
"input_domain": { "type": "string" },
"resolved_domain": { "type": "string", "description": "Domain of the company actually researched. Empty if not found." },
"match_confidence": { "enum": ["high", "medium", "low", "not_found"] },
"what_they_do": { "type": "string", "maxLength": 300, "description": "One or two plain sentences" },
"industry": { "type": "string" },
"hq_country": { "type": "string", "description": "ISO country name or 'unknown'" },
"employee_range": { "enum": ["1-10", "11-50", "51-200", "201-500", "501-1000", "1001-5000", "5000+", "unknown"] },
"signals": {
"type": "array", "maxItems": 5,
"items": {
"type": "object", "additionalProperties": false,
"required": ["type", "claim", "date", "source_url"],
"properties": {
"type": { "enum": ["hiring", "funding", "leadership_change", "product_launch", "expansion", "tech_stack", "partnership", "other"] },
"claim": { "type": "string", "maxLength": 200 },
"date": { "type": "string", "description": "YYYY-MM or YYYY-MM-DD; 'unknown' if not stated" },
"source_url": { "type": "string", "format": "uri" }
}
}
},
"fit_scores": {
"type": "object", "additionalProperties": false,
"required": ["icp_industry", "size", "need_signal", "timing", "reachability"],
"properties": {
"icp_industry": { "type": "integer", "minimum": 0, "maximum": 3 },
"size": { "type": "integer", "minimum": 0, "maximum": 3 },
"need_signal": { "type": "integer", "minimum": 0, "maximum": 3 },
"timing": { "type": "integer", "minimum": 0, "maximum": 3 },
"reachability": { "type": "integer", "minimum": 0, "maximum": 3 }
}
},
"fit_total": { "type": "integer", "minimum": 0, "maximum": 15 },
"tier": { "enum": ["A", "B", "C", "disqualified"] },
"disqualify_reason": { "type": "string" },
"personalization_hook": { "type": "string", "maxLength": 240, "description": "Must reference one of the cited signals" },
"unknowns": { "type": "array", "items": { "type": "string" } },
"sources": { "type": "array", "minItems": 1, "items": { "type": "string", "format": "uri" } }
}
}
}
NOTE: this schema intentionally has NO fields for personal email addresses or phone numbers.
Verified contact data comes from a contact-data provider, not from an AI assistant.
```
### Qualification rubric (0–15) (template)
```
QUALIFICATION RUBRIC — edit the descriptions to match YOUR ideal customer profile (ICP).
Score each criterion 0–3. Total 0–15. The AI proposes scores; a human confirms Tier A.
1. icp_industry
3 = core target industry ([your core industries])
2 = adjacent industry we have won before
1 = plausible but no references
0 = outside our market → disqualified
2. size (employees, from the company's own site or a cited source)
3 = inside our sweet spot (e.g. 51–500)
2 = just outside (e.g. 20–50 or 501–1,000)
1 = far outside but has a relevant division
0 = unknown after research (do NOT guess)
3. need_signal (evidence they have the problem we solve)
3 = explicit, dated, cited signal in the last 6 months (job post for the role we replace, public complaint, initiative announced)
2 = indirect signal (growth, new market, tech stack that implies the need)
1 = generic fit only
0 = signal suggests they've already solved it (e.g. announced a competitor)
4. timing
3 = trigger event in the last 90 days (funding, new leader in the buyer role, expansion)
2 = trigger event 3–12 months ago
1 = none found
0 = known freeze (layoffs, acquisition pending) — note it
5. reachability
3 = we have a warm path (customer, investor, partner in common) — fill from YOUR records, not the AI
2 = clear buyer role exists and is publicly identifiable
1 = buyer role unclear
0 = no way in found
TIERS
A = 11–15 and need_signal ≥ 2 → human review, then contact-data lookup, then personal outreach
B = 7–10 → nurture list / lighter-touch sequence
C = 3–6 → park; re-check in 6 months
disqualified = icp_industry 0, or a hard exclusion (existing customer, competitor, sanctioned country, under legal minimum size)
RULE: any score based on a claim without a source URL is capped at 1.
```
### De-duplication and QA checklist with Sheets formulas (checklist)
```
DE-DUPLICATION AND QA — run on every batch before it enters the master sheet
A. De-duplicate (Google Sheets, master tab "Accounts")
1. Normalise the domain in a helper column (assume raw domain/URL in column B):
=LOWER(REGEXREPLACE(REGEXREPLACE(TRIM(B2),"^(https?://)?(www\.)?",""),"[/?#].*$",""))
2. Flag duplicates against everything already researched:
=IF(COUNTIF($C$2:$C, C2)>1, "DUPLICATE", "")
3. Before each run, remove input rows whose normalised domain already exists in Accounts
(=ISNUMBER(MATCH(C2, Accounts!C:C, 0)) on the input tab). This avoids paying for the same research twice.
4. Also check resolved_domain vs input_domain: if they differ, the AI may have researched a
different company with a similar name → set status "CHECK MATCH".
B. Structural checks (instant, every row)
[ ] JSON parsed without edits (if not, re-run the batch; do not hand-fix silently)
[ ] Row count out = row count in
[ ] fit_total equals the sum of fit_scores
[ ] Every signal has a source_url; every row has ≥1 source
[ ] No email addresses or phone numbers anywhere in the output (search for "@" and digit runs) — delete if present
[ ] match_confidence "low"/"not_found" rows are moved to a Manual tab, not scored
C. Hallucination spot-check (10% of rows, minimum 3 per batch, ALWAYS include every Tier A)
For each sampled row, open every source URL and confirm:
[ ] The page loads and is about this company (not a namesake)
[ ] The claim is actually stated on the page (not inferred)
[ ] The date matches
Record result in a "QA" column: PASS / FAIL + note.
If >1 of 10 sampled rows FAIL → re-run the whole batch with the verification prompt; don't trust the rest.
D. Compliance gate before any outreach
[ ] Contact details came from a licensed provider or the person's own published business contact page — not scraped from LinkedIn
[ ] EU/UK contacts: lawful basis (usually legitimate interests) documented; GDPR Art. 14 notice given at first contact
[ ] US commercial email: CAN-SPAM basics in the template (accurate From/subject, postal address, working opt-out honoured within 10 business days)
[ ] Suppression list (unsubscribes, existing customers, competitors) applied
```
## Limitations
- No verified contact data: AI assistants don't maintain a B2B contact database. Emails they produce are often guessed patterns; use a licensed contact-data provider.
- Manual batching: 10–25 companies per run. 500 accounts a month is 25–50 runs and roughly 8–20 hours of human time, versus near-zero in an automated Clay table.
- Research quality depends on what's public. Private companies, small firms and non-English markets return many 'unknown' fields.
- AI research makes mistakes: namesake companies, outdated news presented as current and claims not actually on the cited page. The 10% spot-check reduces but doesn't eliminate this.
- Research features have plan-based usage limits that can change; a heavy month can hit them.
- No native CRM/sequencer push — results move by copy/paste or CSV import.
## Risks
- **reliability** (high): Hallucinated or misattributed facts can end up in outreach and damage credibility. Mitigation: no URL, no claim; spot-check 10% plus every Tier A; re-run failing batches.
- **compliance** (high): Outreach to people still triggers GDPR (EU/UK: lawful basis, Article 14 notice by first contact) and CAN-SPAM (US: applies to B2B; FTC lists penalties of up to $53,088 per violating email). Using the AI doesn't change who is the data controller — you are.
- **compliance** (medium): Scraping LinkedIn (directly or by asking an agent to) breaches LinkedIn's User Agreement section 8.2 and can get accounts restricted. Keep research to public web pages and licensed data.
- **sensitive_data** (low): Company lists and ICP notes are commercially sensitive. Use a business/team plan or check your consumer plan's training settings before pasting customer or pipeline data.
- **functionality_loss** (medium): You lose Clay's automation: scheduled refreshes, enrichment waterfalls across providers and pushes to CRM/sequencers.
## Verification checklist (run before cancelling)
- [ ] The rubric ranks at least 7 of 10 known past customers above at least 7 of 10 known lost/unfit deals in a back-test.
- [ ] Two full test batches (≥40 companies) came back as valid JSON with one row per input, in order, with no emails or phone numbers.
- [ ] Spot-check of those batches: at least 90% of sampled claims are SUPPORTED when you open the source URL; every Tier A row passed a full source check.
- [ ] Side-by-side on 25 accounts: the AI's tiering agrees with your existing Clay scoring (or a human's judgement) on at least 20.
- [ ] The de-dup formula catches a deliberately re-pasted batch (all rows flagged DUPLICATE) and normalises 'https://www.example.com/about' to 'example.com'.
- [ ] Contact data for Tier A accounts is coming from a licensed provider or published business contacts, and your outreach template includes the CAN-SPAM elements and (for EU/UK) an Article 14 notice.
- [ ] You timed a real month's workload (or a week × 4) and the hours are acceptable versus what you pay for Clay.
## Savings methodology
Savings = Clay subscription you can drop or downgrade − any extra spend (normally $0 if you already pay for Claude/ChatGPT with web search). Contact-data costs are excluded on both sides: if Clay was also your email finder, you need a contact-data subscription either way, so subtract that from the savings. Labour is the hidden cost: at 15–30 minutes per 10–25 companies, 500 accounts is roughly 8–20 hours a month — price your time before deciding.
Example (illustrative): A two-person outbound team that used a Clay Launch plan mainly for research and scoring columns on ~500 accounts a month. $167/mo → $0/mo. Assumptions: ILLUSTRATIVE: Clay's pricing page on 2026-10-13 showed $167/month as the Launch headline price but also lower 'starts at' prices and different figures in its FAQ, so we treat $167 as an example, not a verified price. Use your own invoice. They already pay for Claude or ChatGPT with web search; no new subscription. Contact data comes from an Apollo/ZoomInfo seat they already have; that cost is unchanged and excluded. 8–20 hours a month of research and QA time is not converted to dollars.
## Sources
- Clay pricing: https://www.clay.com/pricing (accessed 2026-10-13)
- Using Research on Claude: https://support.claude.com/en/articles/11088861-using-research-on-claude (accessed 2026-10-13)
- LinkedIn User Agreement: https://www.linkedin.com/legal/user-agreement (accessed 2026-10-13)
- FTC CAN-SPAM Act compliance guide: https://www.ftc.gov/business-guidance/resources/can-spam-act-compliance-guide-business (accessed 2026-10-13)
- GDPR Article 14: https://gdpr-info.eu/art-14-gdpr/ (accessed 2026-10-13)
- ChatGPT deep research FAQ: https://help.openai.com/en/articles/10500283-deep-research-faq (accessed 2026-10-13)
# Design a print-ready restaurant menu with AI instead of Canva
> Keep your menu as a simple list, have Claude or ChatGPT turn it into a clean, printable HTML menu (Letter or A4), and save it as a PDF from your browser. Includes a Google Docs route for owners who don't want to touch code, plus the allergen and print-shop caveats that matter.
- Outcome: I want a good-looking, up-to-date menu I can print in-house or send to a print shop, without paying for Canva Pro.
- Replaces: Canva (task_specific)
- Uses: Claude — Clean up the menu list, write short dish descriptions and fill the HTML menu template; it can show the result as a live preview.; ChatGPT (optional) — Alternative to Claude for the same prompts.; Google Workspace (optional) — No-code route: a Google Docs menu template exported to PDF; Google Sheets as the master price list.
- Extra monthly cost: $0 — Uses an AI assistant you already have (free tiers work for occasional menu updates), a web browser's built-in Save as PDF, and optionally Google Docs. Printing costs are the same as before.
- Difficulty: easy; skill: basic; setup: 1–2 hours for the first menu
- Evidence: editorial; illustrative example: true; last reviewed 2026-10-13
- Canonical: https://onelesstool.com/workflows/restaurant-menus-with-ai
## Good fit for
- Independent restaurants, cafés and food trucks that pay for Canva Pro mostly to update menus
- Menus that change often (seasonal dishes, price updates) where retyping in a design tool is the main cost
- Owners who want one master list of dishes, prices and allergens that every menu is generated from
## Not a fit for
- Menus with heavy illustration, photo collages or custom die-cut shapes
- Large-run professional print jobs that need CMYK files, bleed and crop marks (use a designer or the print shop's own template)
- Businesses that also use Canva Pro for social posts, signage and video — cancelling Canva would remove those too
- Anyone not willing to proofread every price and allergen marker personally before printing
## Before
Open the menu design in Canva Pro, find each changed dish, retype prices and descriptions in separate text boxes, nudge alignment, re-export a PDF for the printer and a PNG for the website, and repeat for the A4 and Letter versions or the takeaway menu.
## After
Update one master list (a Google Sheet or a text note) of dishes, prices and dietary tags. Paste it into Claude with the fill prompt and the HTML template; open the result in Chrome, check it on screen, and Print > Save as PDF. Non-technical owners do the same in a Google Docs template and download a PDF.
## Steps
### 1. Make one master list of dishes, prices and dietary tags
Put every menu item in a single list — a Google Sheet works well — with columns: **section, dish name, description, price, dietary tags (V, VG, GF, N…), allergens confirmed by kitchen (yes/no), last checked date**.
This list is the source of truth. The design is regenerated from it, so prices never drift between the printed menu, the website and delivery apps.
**Allergen and dietary information is your legal responsibility, not the AI's.** Fill the allergen columns from recipes and supplier specs with your head chef. Labelling rules differ by country and city; check what your local food authority requires.
### 2. Clean up names and write short descriptions
Paste the list into Claude or ChatGPT with the **Menu clean-up prompt**. It standardises capitalisation, trims descriptions to one line and flags inconsistencies (duplicate dishes, prices without decimals in one section but not another).
Tell it **not** to add or remove ingredients. Review its suggested descriptions against the recipe: a model may make up a garnish that changes an allergen.
### 3. Generate the printable HTML menu
Use the **Fill the menu template prompt** with the template below. Choose `letter` (US) or `A4` in the `@page` line.
Ask Claude to show it as a preview (an artifact) or save the HTML to a file and open it in Chrome. On screen, the page shows a white sheet with the printable margins so you can see whether everything fits on one page.
If something overflows: ask the AI to shorten descriptions, reduce the base `font-size` from `10.5pt` to `10pt`, or move a section to a second page block.
### 4. Save as PDF and print a test copy
In Chrome: **Print > Destination: Save as PDF > Margins: Default > More settings: tick "Background graphics" > Save.**
Print one copy on your normal printer at **100% / Actual size** (not "Fit to page"). Check prices, spelling and the dietary legend on paper, where errors are easier to spot.
**Print-shop caveat:** browser PDFs are RGB, with no **bleed** or **crop marks**. Designs with colour running to the paper edge will show a thin white line after trimming. Either keep a white margin (the template does), or ask the print shop whether they can add bleed or work from your PDF.
### 5. No-code route: Google Docs menu
If you'd rather not handle HTML:
1. In Google Docs, **File > Page setup** and choose Letter or A4 with 0.5 in margins.
2. Ask the AI for the **Google Docs menu layout prompt** output, then right-click > **Paste from Markdown** (turn on Markdown in Tools > Preferences first) or paste plain text.
3. Use a 2-column table with hidden borders for "dish — price" alignment, a heading style per section, and one accent colour.
4. **File > Download > PDF.**
It looks plainer than the HTML version, but anyone on staff can update it.
### 6. Run the final proof pass and publish
Run the **Proofreading and allergen cross-check prompt** with your master list and the final menu text. It returns a mismatch table (price differs, tag missing, dish missing).
The AI can only compare the two lists. It cannot know whether a dish actually contains nuts. Fix every mismatch at the source list, regenerate, and keep the PDF with a date in the filename (e.g. `menu-2026-10-13.pdf`) so you know which version is on the tables.
## Prompts
### Menu clean-up prompt
Standardise the master list and write short descriptions without inventing ingredients.
```text
Here is the menu for {{restaurant_name}}, a {{cuisine_style}} restaurant. Each line is: section | dish | ingredients or notes | price | dietary tags.
{{raw_menu_list}}
Tasks:
1. Standardise dish names to Title Case and keep them under 5 words where possible.
2. Write a description of at most {{max_description_words}} words per dish using ONLY the ingredients I listed. Do not add garnishes, sauces or ingredients.
3. Format prices as {{price_format}} (e.g. "14" or "14.00"), consistently.
4. Keep my dietary tags exactly as given. Do not add or remove tags.
5. After the cleaned list, add a "Check these" list: duplicates, missing prices, dishes with no tags where the name suggests meat/dairy/nuts, and anything ambiguous.
Return the cleaned list in the same pipe-separated format.
```
### Fill the menu template prompt
Turn the cleaned list into a complete printable HTML menu using the template.
```text
Below is an HTML menu template and my cleaned menu list. Produce one complete HTML file:
- Keep the template's CSS and structure. Replace the sample sections and items with mine, in my section order.
- Restaurant name: {{restaurant_name}}. Menu title: {{menu_title}}. Service times: {{service_times}}.
- Paper size: {{paper_size}} (set the @page line accordingly and remove the other).
- Accent colour: {{accent_color}}.
- Footer allergen statement: "{{allergen_statement}}". Contact line: "{{address_phone_website}}".
- Only include dietary tags that appear in my list, and make the legend match the tags used.
- It must fit on {{page_count}} page(s). If it won't, tell me which descriptions to shorten rather than shrinking text below 9pt.
Template:
{{html_template}}
Menu list:
{{clean_menu_list}}
```
### Google Docs menu layout prompt
Produce Markdown that pastes cleanly into Google Docs for the no-code route.
```text
Format this menu for Google Docs. Output Markdown only:
- "# {{restaurant_name}}" then one italic line: "{{menu_title}} · {{service_times}}".
- "## Section" headings.
- For each dish, a two-column Markdown table row: "**Dish name** (tags) – description" | "price".
- A final line with the dietary legend and this statement: "{{allergen_statement}}".
Menu list:
{{clean_menu_list}}
```
### Proofreading and allergen cross-check prompt
Catch mismatches between the master list and the final menu before printing.
```text
Compare the MASTER LIST with the FINAL MENU TEXT. Report every difference in a table with columns: dish | field (name, price, tags, description, missing) | master value | menu value.
Also list spelling mistakes and inconsistent price formats. Do not judge whether tags are correct for the ingredients; only compare the two texts.
MASTER LIST:
{{master_list}}
FINAL MENU TEXT:
{{final_menu_text}}
```
## Templates and scripts
### Printable HTML menu template (Letter or A4, two columns, text dietary markers) (template)
```html
<!doctype html>
<html lang="en">
<head>
<meta charset="utf-8">
<title>{{restaurant_name}} – Menu