Interview transcripts are where the truth is and where nobody looks. Forty pages each, twelve interviews, one person with a highlighter. The research hub exists to make that afternoon shorter without making the result less trustworthy.
What goes in
Paste text or upload .txt, .md or .docx files into an investigation. An investigation is one study: the transcripts, the insights that came out of them, and the maps they inform.
What comes out
For each transcript, the model proposes candidate insights: a short statement, the passage it came from with a line reference, and a confidence level. That is all it does. It proposes.
Why a person accepts every one
We decided early that nothing reaches a map without a human decision. Three reasons.
First, models are confidently wrong often enough that silent publishing would poison the map. Second, the act of reviewing is how the researcher builds the mental model they need in the workshop anyway. Third, when a stakeholder asks "who says so", the answer should be a person, not a setting.
So every insight waits: accept, edit, or dismiss. Edit is the most used button. The model finds the passage; the researcher writes the sentence.
Where it attaches
An accepted insight becomes evidence. Attach it to a portfolio item and the bubble for PAT-2 grows, because four customers said the same thing about shipping costs. Anyone can click through from the bubble to the quotes.
What we do not do
We do not train on your transcripts; content goes through the Anthropic API to return a result and nothing more. There is a monthly quota per workspace so costs stay predictable, and an off switch for admins who would rather not use AI at all. The rest of Journima works identically with it off.
Try it on one interview
Take the transcript that you remember least and run it through. The insight you had forgotten is usually the useful one.