What it is
Using large language models or other AI tools to help cluster, tag, summarize, and surface themes across qualitative research data (interview transcripts, open-ended survey responses, support tickets), with a researcher directing the process and validating the output.
When to use it
During Synthesize, when you have more raw qualitative data than you can read closely by hand and want a faster first pass at patterns before human verification.
Why (value)
Speeds up the slow, mechanical parts of synthesis (first-pass coding, deduplication, summarization) so researchers spend their time on interpretation and judgment. Helps surface candidate themes you might overlook in a large corpus.
How to run it
- Clean and anonymize the source data; remove anything you cannot share with the tool.
- Give the AI a clear task and your coding frame or research questions.
- Have it propose clusters, tags, or theme summaries with supporting quotes.
- Trace every claimed theme back to the actual source data.
- Correct, merge, or discard AI groupings using human judgment.
- Write the final synthesis yourself; treat AI output as a draft, not a conclusion.
Inputs and outputs
- Inputs: Cleaned, anonymized research data; clear research questions; an AI tool you are permitted to use with that data.
- Outputs: A verified set of themes with traceable evidence, plus a faster path to a human-authored synthesis.
Time and effort
Low to medium. The AI pass is fast; the required human verification and interpretation are where the real time goes.
Evidence and limitations
AI-assisted output is inference, not evidence. Models hallucinate quotes, invent patterns, and flatten nuance, so nothing is trustworthy until a human traces it to source data. Never let an AI summary stand in for direct-user evidence, and never feed sensitive or identifiable data into a tool without clearance. See Evidence and limitations.
Tips and pitfalls
- Verify every quote and theme against the raw data; assume fabrication until checked.
- Anonymize and get clearance before uploading research data anywhere.
- Keep the human as author and decision-maker; AI drafts, you decide.
- Watch for confident-but-wrong summaries that smooth over contradictory signals.
Variations
Automated transcript coding, open-ended survey clustering, support-ticket theme extraction, AI-assisted affinity grouping as a first pass before a human affinity session.
Related methods
Need help choosing and applying a method?
Gumshoe helps teams connect research, collaborative workshops and service design to the decisions they need to make.
Sources and further reading
- UX Tigers, “100 UX methods.”
- Nielsen Norman Group, “Using AI to analyze qualitative UX research data.”
Explore related methods
- Affinity diagramming — Cluster many observations into themes to make sense of qualitative data as a group.
- AI-assisted ideation — Use generative AI as a tireless brainstorming partner to expand and challenge the idea space.
- Journey mapping — Visualize a person's end-to-end experience across steps, emotions, and touchpoints to find pain and opportunity.
