Before I share research findings, I ask AI to attack them first. Not summarize. Attack. It's easy to walk away from a few interviews feeling confident about a pattern. AI is good at making that confidence look even more solid — clean summary, clear themes, done. That's exactly the problem. Confidence isn't the same as evidence. So before I write up any findings, I run them through this: "Here are my research findings: [paste]. Play devil's advocate. What would a skeptical reader say is wrong with this conclusion? What's the weakest part of my evidence?" It's caught things like a "pattern" that was really just one participant's opinion repeated across my notes. Takes 2 minutes. Cheaper than being wrong in public. What's one thing AI has pushed back on when you've tried something like this? #UXResearch #UserResearch #AIinResearch