Write effective prompts for the AI interviewer

Write clear prompts that guide the AI interviewer’s follow-up questions and tone.

Use practical prompt-writing strategies. Specify tone, structure, and length. Add examples and clear boundaries for the AI interviewer.

Last reviewed
aipromptswriting5-20 minutesIntermediateResearchersPrompt engineers

Before you start

Decide the tone, depth, and format you want before you edit the prompt. Prepare one or two sample responses to test against.

Steps

  1. Specify tone and follow-up depth

    State the tone you want, such as friendly, professional, or academic. Say how deeply the AI should follow up, such as 1-2 follow-ups or deeper probing. State the response format.
  2. Add examples

    Give positive and negative examples of the follow-up questions you want the AI to ask.
  3. Define boundaries

    Tell the AI which topics to avoid. Say when it should stop probing. Explain how it should handle off-topic responses.
  4. Test and refine

    Try your prompt on sample responses in Preview or with synthetic answers for testing. Revise the prompt until the AI behaves the way you want.

Details and limitations

Be specific about the tone, follow-up depth, and examples of acceptable responses. Short examples help the model match the format and style you want.

Avoid vague instructions. Use concrete tasks like "Ask one clarifying question, then ask for a specific example."

You can review the full prompt that drives the AI interviewer, edit its context, and add your own sections. In a response, open Prompt used to see the prompt and model behind each interview.

We improve the default interviewer instructions over time, and a running study picks up each improvement. For research where every participant must get the same interviewer, turn on Keep the interviewer prompt fixed during a study in Settings > Account preferences. Each study then stays on the instructions its first interview used. Sections you edit yourself never change.

Check the result

Your prompt should clearly tell the AI how to ask follow-up questions, what to avoid, and when to stop probing.

If something goes wrong

If the AI asks questions that feel too broad or too deep, replace vague instructions with one concrete action and one clear limit.