How to Use AI in User Research Without Losing the People
AI can make research faster, and the decisions still belong to the people you serve and the team that serves them.
Research tells a team where people stop, why they stop, and what would help them keep going. AI tools can now draft interview guides, summarize transcripts, and spot patterns across hundreds of comments in minutes. Used well, that speed frees a team to spend more time with the people who use the product. Used carelessly, it produces confident answers to questions nobody asked.
Where does AI help in user research?
Desk research
AI can summarize what is already known: support tickets, app store reviews, survey comments, and published standards. Ask it to group the comments by theme and quote the original words, so you can check each theme against the source.
Planning
AI is a good first-draft partner for a research plan or an interview guide. Ask it to list the assumptions your team is making. Then turn each assumption into a question you can test.
Analysis
After interviews, AI can tag transcripts, count how often a theme appears, and pull representative quotes. A person still needs to read the transcripts, because the most important moment in an interview is often a pause or a single offhand sentence.
Simulation
Some teams use AI to simulate how different kinds of people might react to an idea before anyone's time is spent. Simulation is useful for narrowing options and finding obvious problems early. It is not evidence about real people, and it should never be reported as if it were.
Where does AI fall short?
AI cannot tell you what a real person felt when the page asked for their mother's maiden name.
AI reflects the data it learned from, which underrepresents many groups, including many disabled people and older adults.
AI can invent quotes, sources, and statistics that sound right and are wrong.
AI cannot take responsibility for a decision. A person on your team has to.
Rules that keep AI research honest
Label the source of every finding. Mark each insight as coming from real people, from existing data, or from a simulation.
Check every quote against the transcript. If a quote does not appear in the source, it does not appear in the report.
Protect participants. Remove names and personal details before any transcript goes into an AI tool, and follow your organization's data rules.
Include the people at the edges. Recruit disabled people, older adults, and people with limited time or connectivity on purpose, because a simulation will not represent them well.
Test the big decisions with real people. Use AI to narrow the options. Then put the final options in front of the people who will live with them.
How I research
I do research three ways: desk, digital, and direct. Desk research starts with what is already known. Digital research uses my own AI-assisted software to simulate the people you serve, and I check every result myself. Direct research brings in real people when the decision calls for it. You can choose one, two, or all three.
See how AI-native research and design works
For a related habit, read how to make Claude stay on brand.