How we test
Every guide says exactly how far it was tested. Here is what each label means, and what it does not mean.
The two labels
- Run once on made-up data on [date] with [model]. We pasted the guide's prompt and its made-up example into that AI model one time, and the guide shows the real output, trimmed to the useful part. One run on a made-up business says nothing certain about results with your real data.
- Documented, not yet run. The guide is written from official help pages and our own steps, but we have not run the prompt yet.
Today, 45 of 45 guides are "run once on made-up data", using Claude, Claude Opus 5.5. We never claim a guide was tested in a tool we did not use.
Made-up data only
Every example uses a made-up business, made-up people and dates in 2030. We never use real customers, companies or emails.
Steps inside real apps
28 guides include a step inside a real app, such as Gmail, Outlook, QuickBooks or Google Business Profile. We describe that step from the app's official help page and link it under Sources. We do not run it on a real account.
Time estimates
"Setup" and "Saves" times are our estimates for one typical use. They are not measured results, and we label them that way everywhere.
The human check
Every guide ends with "Check before you send": the names, numbers, dates, promises and tone you check yourself. The prompts tell the AI to use only what you give it, to flag what is unclear instead of guessing, and not to send anything.
Dates and re-checks
Each guide shows the date it was last checked. We re-check a guide when a tool changes or a reader reports a problem, then update that date and log real fixes on the corrections page.
Drafts and approval
Guides are drafted with AI help (see how we use AI). A guide marked "Draft" has not yet been approved by the owner.