How AI Turns User Interviews Into Usable Design Data

A user interview or a usability test starts out as raw material, valuable, but raw all the same. Hours of recording, hesitations, small details that can slip by in the moment. The real value doesn't come from collecting it, it comes from what happens next.

That's where AI earns its place in this process, not to replace analysis, but to speed it up without losing precision.

Capturing without losing anything

Interviews run on Google Meet, with the Gemini extension transcribing live. The summary generated automatically by Gemini is never used. What matters is the raw, word-for-word transcript.

Gemini also records the session on video. That's the safety net for the process. AI can misread a word or miss a shift in tone, and reviewing the recording in parallel often surfaces small details that change how a response should be read.

Structuring with Claude

Once the transcript is in hand, it gets sent to Claude, which proves more reliable for text analysis and writing. Several interviews can be passed at once, along with the interview guide.

That guide holds the questions, but also checkboxes built to track whether a user understood a step or reacted quickly during a natural flow. Claude cross-references these checkboxes with the transcript and returns an interactive folder for each participant: answers organized as bullet points, and observations tied to each checked box. The output is a synthesis already ready to work with.

For usability tests, Claude also brings together, in a single view, the observations from the natural flow and the details gathered during the step-by-step questions. That way it's possible to understand not just that there's a friction point, but why. And to back up every friction or validation, the matching verbatims are pulled out as well.

Turning text into visual data

Those interactive folders are then copied and pasted onto a FigJam board, as sticky notes, one color per participant. They get grouped by topic or question, which makes it possible to instantly stack every participant's input on a single point.

At a glance, it becomes possible to see how many users confirm a friction point or a need, since each color corresponds to a participant. This stays qualitative, but it makes reading almost immediate: the most shared friction or validation points stand out visually, which helps identify what needs the most improvement or iteration.

The Takeaway

This process makes it possible to go from raw interviews to a clear priority list, whether that means moving into ideation after interviews or into iteration after usability tests. AI doesn't replace the designer's judgment, it frees up time to use it where it matters.

The next step, not yet tested: connecting Figma to Claude Cowork so the sticky notes generate themselves. One more way to save time on the mechanical part, and keep more of it for the analysis.