Create an Agent Survey

In an Agent Survey you don’t write questions. The AI interviews you about what you’re trying to learn, then interviews each participant individually — adapting each turn to what that person just said.

Agent Surveys are a paid-plan capability. Static Surveys remain unlimited on every plan.

The Design interview

Creating an Agent Survey has a single authoring step, called Design. It is a live conversation, not a form.

1. Go to Surveys → Create → Agent.
2. The agent starts interviewing you: what outcome do you want, who is the audience, what tone, what’s off-limits, how much of a participant’s time can you spend.
3. As the conversation uncovers each piece, the structured brief assembles in real time beside it.

What you end up with is the Agent Survey Brief — your intent, audience, tone, off-limits, and time budget, together with its Coverage.

Note

Participants never see the brief. It is the instruction set the interviewing agent works from, not a set of questions waiting to be asked.

Coverage

Coverage is the structured part of the brief that steers each interview. It holds:

• An opening prompt — how the conversation starts
• Topics, each with:
  • a rationale — why this topic matters
  • a priority — must-cover, should-cover, or if-natural
  • a probe depth — light, standard, or deep
• Hard guardrails — what the agent must not do or ask
• A closing prompt — how the conversation ends

Priority is what makes Agent Surveys behave sensibly under time pressure: a must-cover topic gets reached even in a short conversation, while if-natural topics only come up if the participant’s answers lead there.

Probe depth controls persistence. A deep topic keeps asking “why” until it bottoms out; a light one takes the first answer and moves on.

The Agent Persona

Participants don’t meet “the AI” — they meet a named interviewer. The Agent Persona is that participant-facing identity: its name, which defaults to Spoke, and its conversational tone. It’s configured as part of Coverage.

Reading the results

Agent Surveys produce an Agent Results Brief — an AI-written synthesis of the response transcripts.

Two things are worth understanding about it:

Live signals don’t wait for the brief. Counts, themes, confidence, citable evidence, and conflict signals update as responses arrive, without regenerating the polished brief each time.

The brief is a snapshot, and regeneration is explicit. It represents the transcripts available when it was generated and is stored with that source metadata. As new responses arrive it becomes stale — it does not refresh itself. Regenerate it when you want it current.

Where a signal isn’t available yet, you’ll see a placeholder rather than a number. Derived signals always come from real transcript evidence.

Static Surveys can have one too

On a paid plan you can explicitly generate an Agent Results Brief from Static Survey data as well, while the survey is live or after it closes. Static Surveys just don’t receive one automatically.