What AI can't do for your user research

I use AI tools in research every week. They transcribe interviews, cluster notes and draft summaries faster than I ever could. But after two years of using them on real projects, I’ve learned where to stop trusting them.

Good at patterns, bad at silence

A model will tell you that six out of eight people mentioned price. It won’t tell you that nobody mentioned the onboarding, even though that’s where everyone got stuck in the session. Noticing absence is still your job.

Summaries flatten tone

“The user found the flow confusing” could mean a mild shrug or a person close to giving up. The transcript has the words but not the pause before them. Always go back to the recording for the moments that matter.

A workflow that holds up

Let the tool do the first pass: transcripts, tags, a rough cluster. Then read the raw notes yourself and look for what’s missing. We walk through this exact workflow in AI in your design workflow, with real interview material.