Skip to content

Responses and transcripts

Open a survey to see its responses. Each one has a transcript: the ordered record of what the survey asked and how the participant answered.

Every response has one, whether the survey is Static or Agent. That matters for two reasons.

It preserves context. For an Agent Survey the questions differ per participant, so a bare answer is meaningless without knowing what prompted it. The transcript keeps them together.

It’s the canonical evidence. Structured answers — the values behind charts, filters, and summaries — are derived from the transcript, not stored separately. So anything the product asserts can be traced back to what someone actually said.

When you see a theme or a count you don’t believe, open the transcripts behind it. That path always exists.

During the edit window, a participant revising their answer replaces the current transcript. There’s no revision history.

This is deliberate: the transcript represents what the participant currently stands behind, not a record of them changing their mind. If you need to know that answers shifted, a recurring series is the tool for that.

Static Surveys show basic counts and analysis automatically.

They don’t receive an Agent Results Brief by default — but on a paid plan you can explicitly generate one from Static Survey data, while the survey is live or after it closes.

Agent Surveys show derived signals that update as responses arrive: counts, themes, confidence, citable evidence, and conflict signals. These don’t require regenerating the Agent Results Brief each time.

Two rules govern what you see:

  • Signals come from real evidence — actual transcript text or model-extracted values. Nothing is seeded or invented.
  • Where a signal isn’t available yet, you get a placeholder, not a plausible-looking number.

So an empty state means “not enough evidence yet”, which is information in itself.

Participants can download their own transcript.