Clarity's AI session summary confidently described my own dev-testing traffic as a distracted user. It wasn't.

Quick disclosure since this is a tools group: no affiliate/sponsorship angle here - Microsoft Clarity has no paid tier at all, so there's nothing to earn a commission on even if I wanted to. This is just a finding from testing it firsthand against my own site.

Clarity (Microsoft's free heatmap/session-recording tool) has an AI feature that summarizes session recordings for you - useful in theory, since watching every recording yourself doesn't scale. I ran it on a session that happened to be me testing my own dev server, not a real visitor.

The summary was fluent and confident:

"The user repeatedly refreshed the Activepieces review page, possibly indicating issues with content loading or display."

"Frequent 'Page hidden' events suggest that the user may have been distracted or switched tabs often, potentially impacting their engagement with the content."

Neither was true. That was normal dev-server testing behavior (reloading a page, switching tabs to check code) getting narrated as a frustrated, distracted visitor. Clarity's own UI includes the disclaimer "Your insights are powered by AI, so mistakes are possible" - which is honest of them - but this is what that mistake actually looks like in practice, not a hypothetical.

The part that stuck with me: the summary wasn't vague or hedged. It picked a specific, plausible-sounding narrative and stated it as fact. If I hadn't known exactly what that session actually was, I'd have believed it - it reads exactly like a real UX insight would.

Not a knock on Clarity specifically - I'd guess most "AI summarizes your data" features have the same failure mode, since the model has no way to tell "normal dev activity" from "confused user" apart from pattern-matching on the same raw events. Curious if anyone here has hit the same thing with a different tool's AI summary feature - and if so, did you catch it the same way (knowing the ground truth yourself), or did something else tip you off?