The Tell

You ask the user if the task was simple.

They pause. Mouth slightly open, no sound yet. Eyes drop to the floor for a moment. Small crinkles appear at the temples. Then the smile comes, and the answer: "It was easy."
You know it wasn't.
That half-second before the answer is where the real data lives. Not in what they said - in what their face did before they decided what to say. The embarrassment of struggling with something that was supposed to be intuitive. The social instinct to not appear incompetent in front of a stranger. The performance of confidence that AI reading the transcript will record as a positive signal.

This isn't an argument against AI in research. Synthesis, pattern recognition, theme extraction across large volumes of sessions - these are tasks AI handles faster and more consistently than manual analysis.

But there is a category of insight that only exists in the room. The hesitation before the answer. The user who has developed a workaround for a broken flow and navigates it so automatically they no longer register it as friction - and therefore never mention it. The complaint, framed as a compliment because the participant doesn't want to seem difficult. These signals were never captured because they were never spoken, or not spoken explicitly.

As AI takes over more of the analysis layer, the skill of being present enough to catch what doesn't get said becomes rarer and more valuable.