A user test where AI preps and sorts, and you do the watching
Five participants, tasks that do not give the answer away, a session you moderate yourself, and a synthesis where every quote can be checked against the source.
Why
AI writes the report of a test it never saw. It transcribes fast, groups fast, and invents a quote with the same confidence it copies a real one. So the quote check comes before the readout, not after: the method is in Research synthesis. The rest of the problem is not in the text. A participant who hesitates, clicks the wrong thing and backs out almost never says so out loud. The transcript is clean, the journey failed. Hand off the support work. Keep the part that teaches you something: watching.
When
You build this kit when a product decision is waiting on real users reacting to a prototype or a product. Audience: product designers, PMs, POs. ✅ You know it's your kit when You have a prototype and one precise question to put to it You will record sessions and plan to synthesise them with AI Someone wants a report fast, and you can see the unchecked summary coming ❌ Not for this kit You have…
Tools included
- Claude
- Claude Code
- Granola
- Dovetail