Creating an interview
The quality of your Insight study depends mostly on how clearly you define goals before publish. These are the areas worth spending time on in the builder.
Start with one primary goal
Ask: “What decision will this conversation inform?” Write one primary goal — for example “Understand why trial users churn before connecting their data.” Vague goals (“get feedback”) produce vague interviews. Use the in-builder AI assistant to tighten wording.
Objectives: tone, depth, and guardrails
- Tone — friendly check-in vs. formal research; match your brand and audience.
- Depth — how many follow-ups on pain points; when to move on.
- Boundaries — topics to avoid, max length, or sensitive areas.
- Anchors — must-ask themes so nothing important is skipped.
What the AI needs to know
Add approved answers the agent may say out loud — response times, privacy, recording policy, contact paths. This is not internal notes; participants may ask these questions mid-chat. Context assets (links, screenshots) can be surfaced when relevant.
Insight types
Insight types control how the assistant labels notes (praise, complaint, feature request, friction, etc.). Keep the set aligned with what you will act on — defaults work for general feedback; narrow the list for focused studies.
Common mistakes
- Too many competing goals in one interview — split into separate studies if needed.
- Internal jargon in objectives that participants will not understand.
- Skipping live preview — always run a test conversation before sharing widely.
- No clear stop condition — say when the agent should wrap up.