Drop in what customers actually said — interview transcripts, survey exports, support tickets, NPS verbatims, review dumps, win/loss notes — and get one synthesis: the themes that repeat, the jobs behind them, the pains in the customer's own words, the triggers that started the search, the vocabulary worth reusing in your copy, and personas only when the evidence actually supports them. Every quote is traceable to the asset it came from, and thin evidence is reported as a gap rather than dressed up as a finding. Adapted from the @coreyhaines31/customer-research skill on SkillSafe.
Nothing to hand? Load the — three interviews, an NPS export and a ticket digest for a fictional support-analytics tool — or the , a three-row survey that should come back as insufficient-sample with the gap analysis as the deliverable.
Paste and tag your assets
Up to twelve assets, each tagged interview, survey, tickets, NPS, reviews, win/loss or notes, with an optional segment label like "churned SMB, May". The free prescan reads them in your browser: words, independent data points, speaker labels, the NPS promoter/passive/detractor split, quote candidates, dates, and bias flags for small samples, single sources, promoter skew, duplicates and stale material.
The AI synthesizes
Themes ranked by frequency times intensity, each with verbatim quotes and their asset ids, an intensity and a confidence level; jobs to be done split functional, emotional and social; ranked pains in the customer's own words; trigger events; desired outcomes; the vocabulary to reuse; alternatives considered; contradictions between scores and verbatims. Personas appear only when a consistent segment carries five or more independent data points — otherwise the reason lands in the gaps.
Check and export
The coverage check reconciles every asset id and every prescan flag: weighed or set aside, confirmed, revised or cleared — and the app flags anything the synthesis skipped or disagreed with. Then take it away: copy the whole report, download Markdown or JSON, or pull the quote bank as CSV for the messaging doc.
Adapted from the @coreyhaines31/customer-research skill on SkillSafe.
What does a synthesis cost, and do I need an account?
The synthesis itself is a metered AI run charged to your SkillSafe credits — you are shown the reserve before you press the button and you only pay what the run actually uses. It needs a signed-in account. Everything else is free and works signed out: the browser prescan that counts your evidence, and both bundled example results, which replay in full without charging anything.
How do I know the quotes are real and not invented?
Every theme quote is checked against the raw text you pasted, in your browser, for free. Each quote is labelled found word-for-word in the asset it cites, found but sitting in a different asset than the one cited, or not found in any asset you supplied. Lightly elided quotes are matched fragment by fragment. The verdict is shown next to the quote and travels into the Markdown and the quote-bank CSV.
Why did it refuse to give me personas?
A persona needs at least five independent data points from one consistent segment. The evidence-by-segment table groups your assets by the label you gave them and shows how many data points each segment actually carries, and how many more it needs. Below that threshold the app reports the gap instead of inventing a persona, and it says which segment is closest.
Can I compare a new synthesis against an earlier one?
Yes. Every run is saved to your SkillSafe account, so past syntheses follow you to any device, with a copy mirrored in this browser. Pin one as the baseline and the next run is automatically diffed against it: posture change, data-point delta, themes and pains added or dropped, confidence upgrades and downgrades, and which research gaps closed. Comparing never overwrites the assets currently in the form.
What happens to the research I paste in?
The prescan, the word-budget clipping, the verbatim check and every export run entirely in your browser — nothing is uploaded until you press Synthesize the research. Files you drag in are read locally by the browser. Your unsent draft is kept in this browser only so a reload does not lose an hour of pasting.
What if my corpus is bigger than the send budget?
Up to twelve assets and 24,000 words are sent. Over that, the middle of the largest assets is cut on line boundaries — keeping the opening and the closing, and keeping CSV header rows — marked in-band with a clipped marker, and you are told exactly which assets lost how many words. Nothing is silently truncated from the end.