Build your question guide

The question guide is the script the AI moderator follows. It's an ordered list of steps. The AI runs them in order, asks dynamic follow-ups where you've enabled them, and skips conditional sections that don't apply.

Steps

1. Open the interview region

   In Build, the question guide appears below the screener divider in the Guide panel.

2. Add your first step

   Click + Add element to open the step menu. Steps are grouped into five categories:

   • Ask: Open question, Task, Statement.
   • Survey: Multiple choice, Ranking, Open text, Slider, Rating scale, Media upload, Matrix, NPS.
   • Structure: Section, Randomization (shown as Variant comparison in the editor).
   • Usability: Prototype (Figma), Website, Image, Video, Embed, Text.
   • Test: Card sort, Tree test.

   The most common step is Open question, a question the AI moderator will ask in your own words.

   

3. Write the question

   In Properties, type the question text. Write it the way you'd say it aloud. By default the moderator paraphrases it conversationally rather than reading it word-for-word. Turn on Ask question verbatim (below) when the exact wording matters.

   To format what participants read on screen, use the buttons above the text box: bold, italic, underline, strikethrough, bulleted and numbered lists, a heading, a subheading, a quote, and links. Formatting works on questions, tasks, statements, and survey questions, but not on answer options, section titles, or media. The moderator still gets the words as one plain line, so a heading and its bullets are asked as part of the question.

   Optional fields depend on the step type:

   • Follow-up depth (on open questions): Off, Low, Medium, High (roughly 0, 1, 2, 3 follow-ups the moderator may ask).
   • Follow-up guidance (on open questions): tell the moderator what to probe ("dig into pricing objections").
   • Ask question verbatim: force the moderator to read the question word-for-word instead of paraphrasing.
   • Blur recording for privacy: blur the screen recording while just this item is active. Useful for moments when the participant enters personal or sensitive information.
4. Add a task (when participants should do something)

   A Task is for when the participant needs to do something: walk through a prototype, visit a website, open a file. Where an open question asks the participant to talk, a Task asks them to do something while thinking aloud.

   Tasks pair naturally with a stimulus: drop a Figma prototype or other media into the same Section, and it stays visible the whole time. Task completion behavior (the moderator decides, a button click, or a verbal "done") is set globally in Task completion handling.

   Tasks do not support dynamic follow-ups. Add an open question after the task to explore the participant's experience.

5. Add a statement (when you want to say something without asking)

   A Statement is the AI saying something to the participant: context for the next question, a transition, a thank-you. No response is collected. Statements are read aloud exactly as written.

   Common use: put a Statement as the first item to add your own framing right after the moderator's greeting, or as the last item for a thank-you in your own words right before the moderator's built-in closing.

6. Organize with sections

   Add a Section to group related steps. Sections can be:

   • Conditional: only shown to participants who match a rule. See Conditional sections.
   • Limited to a percentage of participants: show this section to only N% of participants (sticky per participant).
   • Randomized: pick one of multiple paths at random per participant (see Randomization).

   Drag steps into a section to nest them.

7. Reorder, duplicate, or remove

   Drag steps to reorder. Right-click for duplicate or delete.

What can go inside what

The builder enforces these containment rules and blocks invalid drops.

Common rules people hit:

• Sections can't nest. For sub-topics, use separate consecutive Sections. Participants never see section boundaries anyway.
• Media blocks can't nest and hold exactly one piece of media. For a second piece of media, add a second media block.
• Media blocks can't contain Sections or Randomization blocks: just questions, tasks, statements, and survey questions.
• Only Variant options go inside a Randomization block.
• A card sort holds follow-up questions but not media or another card sort. Its follow-ups are open questions, tasks, statements, and survey questions.
• A tree test holds follow-up questions the same way. Its follow-ups are open questions, tasks, statements, and survey questions, asked while the outline stays on screen.

Best practices

A few rules keep responses genuine and the conversation natural. See Best practices for the full checklist.

• Don't lead or load. "What problems do you have with X?" plants the answer. Ask "How do you feel about X?" and let the participant supply it.
• Ask about what people did, not what they'd do. "Tell me about the last time you…" beats "Would you use a feature that…" because people predict their own behavior poorly. (Reactions to something they're looking at right now are fine.)
• One thing per question. "What do you like and dislike about X?" is two questions. Split it.
• Keep questions open. If a participant can answer in one word, the question isn't working: "Walk me through…" beats "Did you…".
• Cover every research objective with at least one question.
• Follow a sequence: warm-up → context → deep dive → exploration → wrap-up, with a short statement at each section transition.
• Keep the intro to project context (don't mention recording, time, or future contact), and close with a simple thank-you.

Tips

• Write questions like you'd say them out loud. The moderator's voice will deliver them. Long, jargon-heavy questions sound stilted when spoken.
• Use follow-ups generously on open questions. They draw a fuller explanation from a participant who answers "yes".
• Use Statements between sections to give participants context and warmth. The moderator will sound less robotic.
• Tasks need clear success criteria. Tell the moderator what "done" looks like: completing a flow, finding a button, voicing a reaction.

Related