Cross-study quote mining

Builds a corpus across the studies you name, sorts participants by segment, and pulls verbatim quotes with the session they came from.

Workflow 7 of 12SynthesisSeptember 30, 2026

Get the workflow

Enter your details and the prompt appears here, ready to copy.

How do I use AI to extract key quotes from user research transcripts?

Give the AI a list of studies, a segment definition and a topic, and tell it to return only direct quotes, exactly as spoken, each with the participant's role, the study and a link to the session. Connected to Great Question through the MCP, it searches the transcripts itself. Forbid it from summarising or combining quotes, and check its segment calls before you build on them.

Who runs it
Researchers. It is the workflow that most repays doing properly, because its output ends up in front of stakeholders.
Where you sign off · Read only
Nothing is written. But check the segment calls before you build on them. The agent is deciding who counts as your segment, and that decision is yours.
Schedule it · By hand
No. This is thinking work with a fetching problem attached.

How to run it in Claude, ChatGPT, Cursor or Copilot

The prompt is the same in every tool. What changes is where you connect Great Question and where you paste it.

Run it in Claude

  1. 1

    Connect Great Question to Claude

    In Claude Desktop, open Settings → Connectors → Add custom connector. Name it Great Question, paste https://greatquestion.co/api/mcp/v1, then Save and Connect and sign in with your Great Question account. Restart Claude so the tools load.

  2. 2

    Paste the prompt

    Start a new chat and paste the prompt. Claude asks before it uses each Great Question tool; allow the read tools, and read anything it proposes to write before you allow it.

Your admin has to switch the MCP on in Great Question first. The MCP guide covers every tool, the security model and a team rollout.

The tools it calls

Named so you can see exactly what the agent touches before you let it near anything.

  • search_repo_sessions
  • search_transcripts
  • get_repo_session_transcript
  • search_repo_highlights

What we never automate

Four rules sit under every workflow here, and they matter more than the workflows.

  • Nothing reaches a participant without a person

    No workflow sends an invitation, an email or a reminder. The MCP can do those things. They are left out on purpose, because a scheduled job that contacts your customers is a different category of risk from one that reads a transcript.

  • Nothing launches a study without a person

    An agent can build the screener and draft the messages. Someone still presses go.

  • Nothing reaches a stakeholder unchecked

    The synthesis workflows hand you quotes with links attached so you can check them in one click. Use the click. A misattributed quote in a board deck costs more than the hour it saved.

  • Nothing sees more than the person who connected it

    Permissions follow the individual, so an agent connected by a PM reaches exactly what that PM reaches. Your admin decides whether it is available at all.

Questions

What questions should I ask an AI tool to extract key quotes?

Ask for direct quotes on one topic, from one segment, with the source attached, and forbid summarising. “Pull direct quotes about pricing from mid-market admins, exactly as spoken, with the session link” gets far better results than “what did people say about pricing?”

Can AI misattribute quotes?

Yes, which is why every quote comes back with its session link. Click through on anything going into a deck. A misattributed quote in front of stakeholders costs more than the time the search saved.

Is quote mining the same as synthesis?

No. It finds the evidence; deciding what it means is still your job. For the analysis itself, see our guide to qualitative data analysis.

Related reading

More workflows

See all 12