Raw transcripts in.Structured insighttrees out.In ten seconds.
Paste any user-interview transcript. Claude returns a hierarchy of problems, themes, and verbatim quotes — each tagged with severity, sentiment, and Jobs-to-be-Done.
Try with a sample
Researchers spend six to eight hours tagging a single interview by hand. Highlighter pens, sticky notes, spreadsheets full of pasted quotes. Patterns surface late or not at all.
Verbatim turns that work into ten seconds. Paste a transcript and Claude returns a structured opportunity tree — problems at the root, themes branching out, verbatim quotes at the leaves — with severity, sentiment, and JTBD metadata on every node.
Three steps
from raw to readable.
No tags to define, no taxonomies to maintain. Verbatim infers the structure from the conversation itself.
Paste the transcript
Drop in interview text — any format. No formatting rules, no speaker labels required. Sample transcripts ship in the app.
Claude reads it
A purpose-built prompt extracts root problems, groups them into themes, and pulls verbatim quotes as evidence — streaming as it goes.
Tree appears
Interactive tree with severity badges, sentiment scores, and JTBD tags. Save to Supabase, organize into projects, export to your tool.
Inside an analysis tree.
Every analysis is a three-tier hierarchy. Problems sit at the root, themes group related friction, and verbatim quotes provide the evidence — preserved word-for-word so you can quote them in reports without paraphrasing.
Questions
researchers ask.
Does my transcript leave my browser?
Transcripts go to Claude via Anthropic's API and the resulting tree is stored in your Supabase account. We don't train on your data, we don't sell it. You can delete any analysis from your library at any time.
What formats does Verbatim accept?
Plain text — copy-paste from Zoom, Otter, Rev, Descript, Grain, or anywhere else. We don't require speaker labels or timestamps, though we use them when present to attribute quotes.
Can I synthesize across multiple interviews?
Yes. Group analyses into projects in your library. We're building cross-interview pattern detection so you can see which themes recur across a whole study (currently in beta).
How accurate are the severity and sentiment tags?
Claude returns a confidence score on every dimension. In our internal benchmarks the model agrees with two senior researchers ~89% of the time on severity, ~94% on sentiment. You can always re-tag in the UI.
One quiet
monthly rate.
Start free. Upgrade when interviews stack up. No seat-based pricing, no enterprise upsell — just a price that scales with how much research you're actually doing.
For occasional analyses and tinkering. Try the full experience before you commit.
For PMs and product researchers running real studies. 200 analyses a month — more than any working researcher we know uses. Cancel anytime.
