Paste interview transcripts or notes. Get themes, pain points and insights, each linked to the exact quotes behind it, so you can see what's evidence and what's the AI's reading.
Example: five fictional interviews about booking a GP appointment.
AI wording People struggle to get through by phone at the morning rush and often lose the appointments they need.
Their wordsI redialled for twenty minutes.P1
Their wordsI phone. And that's a whole story because everyone tells you to ring at eight o'clock.P2
Their wordsI rang at eight like everyone else. Engaged. Rang again, engaged.P3
AI wording Logging in, resetting passwords and verifying identity regularly block people from using the online booking route.
Their wordsI'd also want the login to just work.P1
Their wordsIt wanted a photo of my passport, then a video of my face.P3
Their wordsThe problem is logging in. It sends a code to my phone by text, and the code expires after five minutes.P4
AI wording People do not know which clinician suits their need or what the appointment type names mean, so they guess.
Their wordsNobody asked me what I wanted.P1
Their wordsI didn't want to waste a doctor's time if a nurse could do itP3
Their wordsI don't know what triage means.P4
AI wording Being asked about symptoms quickly by a receptionist feels awkward or like a test, making people unsure what to say.
Their wordsI felt a bit awkward telling a stranger.P1
Their wordsThe receptionist is lovely, but she asks me things quickly, like it's a test.P2
AI wording People want to see available times and clinicians directly and choose a slot themselves rather than wait for a call back.
Their wordsIf I could see the free slots myself, like booking a restaurant.P1
Their wordsTell me who I'd see, what they can do, and when. And let me pick a time.P3
AI wording People want clinician roles and booking options explained in simple words, with pictures, large text or sign language.
Their wordsI'd like someone to explain it in plain words, on a leaflet or a website with proper large text.P2
Their wordsI would like it written in simpler words with a picture, or a short video in sign language explaining the options.P4
AI wording Booking forms with unlabelled controls leave screen reader users unable to tell what each option is for.
Their wordsIt doesn't say what they're for. So I can't tell which one is which.P2
AI wording Booking on behalf of another person raises concerns about seeing private medical records and who else can view them.
Their wordsI don't want to read things she hasn't told me.P5
AI wording There is no reliable way to request a sign language interpreter during booking, which affects privacy and access.
Their wordsYou're supposed to be able to ask for one when you book, but there's no box for it.P4
AI wording Clear text message reminders with a set time and one-tap cancellation are seen as a working part of the process.
Their wordsText messages. When they text me a reminder with a clear time, that's good.P4
Access barriers at the login, phone and form stages stop people before they reach any choice about who or when to see.
The AI's reading, not evidence. Rests on: , ,
Unclear clinician options combined with rushed receptionist questions leave people guessing and losing appointments.
The AI's reading, not evidence. Rests on: , ,
People want to see and choose their own slots, but privacy and interpreter gaps make the online route hard to trust.
The AI's reading, not evidence. Rests on: , ,
Scan these for anything the themes missed or that contradicts them.
Kept in this tab until you close it. Download or copy anything you want to keep. Your text is sent to Anthropic, the company behind Claude, to produce the synthesis. Under Anthropic's commercial terms it isn't used to train their models. We don't store it.
Raw words, names removed
Transcripts or notes, one source per participant.
Count the people behind a theme
Trust themes with quotes from several people. Check what was left out.
Interpretation is a lead
Re-read quotes in context, then test the AI's reading.
Built from an idea by Jonathan Hill, UX strategy and research trainer.
Every quote on a theme is checked against the text you pasted. If it appears word for word, it is shown exactly as you wrote it and counts as evidence. The AI's one-line summary of each theme, and all insights, are labelled as the AI's wording. A theme with no quote that checks out is marked Interpretation.
The theme has verified quotes from exactly one source. It may still be true and important, but only one person said it, so treat it as something to test with more people.
The AI proposed words that are not in your text: a paraphrase, an edited quote, a merge of two sentences, or something the interviewer said. We drop those and show you how many. Rejected quotes are never shown, because they aren't your participants' words.
Raw words, not your summary: the tool can only check what is in the text. Use one box per participant so that "how many people said it" means something, and keep speaker labels like Interviewer: and Sam: at the start of each turn. You can also load .txt or .vtt files. Add your research question to steer the grouping.
Up to 10 sources and about 120,000 characters in one run, roughly two one-hour interviews or eight short sets of notes. Three to eight interviews is the sweet spot. One interview can only ever give "One participant" themes. If you go over, split it into two runs; we never cut text off silently.
Your text is sent to Anthropic, the company behind Claude, to produce the synthesis. Under Anthropic's commercial terms it isn't used to train their models. We don't store your transcripts or the result on our servers. In your browser they stay in this tab's session storage until you close the tab or press Reset.
Yes. Replace participants' names, employers, places and anything else that identifies them before you paste. We mask emails and phone numbers and swap speaker names for P1, P2 and so on, but we can't catch everything. Don't paste health, religion, ethnicity or other sensitive details unless your participants agreed to it.
Read each Supported theme's quotes in context. Treat Interpretation and One participant themes as hypotheses. Check who you interviewed and who is missing, scan what the AI left out for anything that contradicts a theme, and don't present counts as percentages of users.
Yes. Tick the themes you want and open the persona builder with them filled in: pain points, goals and behaviours from the theme titles, and a few of your participants' own words as quotes. Nothing is invented, so you add the name and role. You can also send themes to a card sort as cards.
Yes, with no account. There is a daily limit on runs so that it stays free for everyone.
Tools help you do the work. They don't teach you the judgement behind it. Beginner UX (AI) Design is taught live online by working UX professionals, built for career-changers with no design background.
Working in a team rather than learning on your own? We run in-house UX training built around your own product.