AI Sheet, real tests
AI Sheet: real tests, with and without the sheet
The same question, to the same AI, three times: with no sheet, with the questionnaire, and with Mynd's usage data. The answers are the AI's own; below each one, in the full page, Claude's analysis.
How they were done
- AIs used: ChatGPT and Gemini, in the browser, with no account, on 5 November 2026, in Portuguese. Each answer in a new window, so nothing passed from one to the other.
- The question is always exactly the same. Only what comes before it changes.
- The questions: how can I be more productive; what should I focus on this week; what is the best way to back up my data; invent something that does not exist yet and that I would need; recommend 5 books to read next; and a question about the economy.
- When an answer showed a flaw in the sheet, the sheet was fixed in the app the same day, and the analysis says what changed.
- The sheet is the one of Mynd's creator, who agreed to show everything.
Claude's assessment
The AI Sheet works where it matters most. In every question tested, in ChatGPT and Gemini, the same AI gave clearly better answers with the sheet: it stopped treating the person as any student, respected their level, budget and language, and with Mynd's data it started talking about the person's real week, not textbook advice.
The tests also showed its limits, which is what makes them honest: the AIs misread some numbers, pulled personal questions towards the main project and followed some Mynd Moves only on the surface. Each failure produced a fix in the app that same day, and the repeated answer was better.
What these tests prove, and what they do not
- Proven: the sheet changes the answer, a lot. The difference was large in every question, in both AIs, and can be seen without explanation.
- Likely, not yet proven: that each fix solved its problem. Each case was tested once, and an AI never gives the same answer twice.
- The biggest limit: all tests used the sheet of one person, with almost all records in one project. With a more varied life, results may differ.
- Next tests: repeat each case two or three times; use other people's sheets; put an AI with its own memory against the same AI with no memory but with the sheet; a blind ranking of answers; and the Mynd Moves on decision questions.
Where it is worth most
In questions about yourself: your work, your week, your habits, decisions with real limits such as money, time or devices, and recommendations that should fit your tastes. There, without a sheet, the AI can only guess. It is also worth it when you use several AIs, because the same sheet serves them all. It is worth less in questions of fact: there the sheet only adjusts length, tone and language.
Seven real flaws in the sheet were found and fixed on the day. That is how the AI Sheet will keep improving.
The full answers, word for word, with the analysis of each one (in Portuguese).
