AI in UX research: faster synthesis, same need for real users
Language models can sort interview notes and surface themes in minutes. They can't replace talking to the people you design for.
Creative Pouch · · 1 min read

Research synthesis used to mean days of sticky notes on a wall. AI can now do a first pass on a stack of interview transcripts in minutes. That’s useful, but it comes with a trap.
Where AI helps
Transcription and tagging
Automatic transcription is now good enough for most interviews. Models can also tag quotes by topic, which turns hours of listening into minutes of reviewing.
First-pass clustering
Give a model your notes and ask it to group them into themes. You’ll get a reasonable starting structure that you can then challenge, merge and rename. It’s like having a colleague do the first round of affinity mapping.
Finding the quote
“Did anyone mention pricing?” is a question a model can answer across twenty transcripts instantly. That makes it much easier to back up a finding with evidence.
Where it goes wrong
Synthetic users
Some tools offer AI “personas” you can interview instead of real people. They’re fluent and confident, and they tell you what the average internet text would say. Real users surprise you. That surprise is the whole point of research.
Flattened nuance
Models tend to summarise toward the majority view. The one participant who used your product in a completely unexpected way, often the most valuable insight, can disappear in a tidy summary.
Invented evidence
Always check that quotes in an AI summary actually exist in the transcript. Models sometimes produce plausible paraphrases presented as direct quotes.
A healthy workflow
- Run the interviews yourself. Nothing replaces being in the room.
- Use AI for transcription, tagging and a first clustering.
- Review the clusters as a team and rebuild them where they feel wrong.
- Pull every key finding back to real quotes from real people.
AI makes research faster. It doesn’t make it optional.


