The say–do gap
Why asking people isn't measuring them
Ask someone whether they'd use your product and they will tell you something. It will be articulate, considered, and delivered in good faith. It will also be a weak predictor of what they do next.
This is not cynicism about users. It is one of the most replicated findings in social psychology.
The number
Sheeran's review of the intention–behavior literature aggregated ten prior meta-analyses covering 422 studies and found an average correlation between stated intention and later behavior of r ≈ 0.53.
Squared, that is roughly 28% of the variance in behavior. Intention is a real signal, not noise, but it leaves about three quarters of what people actually do unexplained.
The experimental evidence is sharper. When researchers successfully change someone's intentions, the resulting change in behavior is reliably smaller: a medium-to-large shift in intention produces only a small-to-medium shift in behavior. Moving what people say they will do is not the same as moving what they do.
What this means for product research
Most user research, synthetic or human, is built on asking. Interviews, surveys, concept tests, "would you pay for this." All of it collects the stated layer, and the stated layer is the one with the weaker link to outcomes.
That is also where the current generation of synthetic-user tools operates. They convene AI participants and interview them. The output is what a simulated person says about your product. Even if the simulation of speech were flawless, you would have automated the half of research that predicts behavior least well.
The problem isn't that the answers are synthetic. It's that they're answers.
What we do instead
Loop doesn't interview anyone. Human-like synthetic users move through your real screens, and we record what happened: where attention went, what got clicked, which step lost them, whether they finished. Then we hand you the session so you can watch it back.
Heatmaps per screen. Task funnels with exact drop-off points. Feature friction, ranked. Session replays as proof behind every finding.
Not a transcript of opinions. A record of behavior.
The honest part
The gap cuts both ways, and we won't pretend otherwise.
The literature says stated intention is a weak-to-moderate predictor of behavior, not a useless one. r = 0.53 is a real relationship. Interviews also tell you things behavior cannot: why someone hesitated, what they thought your pricing meant, what they expected to happen. Those are worth knowing. If you need them, go ask a person.
Our claim is narrower, and we think more useful: if the question is what people will actually do, watch them do it. That is the question Loop answers, and on the evidence it is the one that asking answers worst.
Sources
- Sheeran, P. (2002). Intention–Behavior Relations: A Conceptual and Empirical Review. European Review of Social Psychology. overview
- Sheeran, P. & Webb, T. L. (2016). The Intention–Behavior Gap. Social and Personality Psychology Compass. Wiley
- Webb, T. L. & Sheeran, P. Meta-analysis of experimental evidence on changing intentions. PMC