How to Write an AI-Moderated Interview Discussion Guide (2026)
Qualitati Research Team · 2026-06-23 · 10 min read
Short answer: A discussion guide for AI-moderated interviews is the structured brief that tells an AI interviewer what to learn, how to probe, and when to move on. Unlike a human moderator’s guide, it must encode instinctive decisions as explicit rules: section objectives, probing depth, tone, and stop conditions. In 2026, the quality gap between rich transcripts and shallow ones is almost always the guide — not the model. Keep it to 6–10 core questions, write objectives not just questions, and define when the AI should dig deeper versus advance.
Last updated: June 23, 2026.
Why the discussion guide is the bottleneck in AI-moderated interviews
An AI-moderated interview is a qualitative interview run by a conversational AI that asks your questions, listens to each answer, and decides in real time whether to probe deeper or advance. The discussion guide for AI-moderated interviews is the brief that drives those decisions. It is the single highest-leverage artifact in the whole workflow, because the AI will faithfully execute whatever you give it — including its gaps.
A human moderator improvises. They sense hesitation, notice an unprompted detail worth chasing, and quietly abandon a question that is not landing. An AI moderator does none of this unless you tell it to. As the Nielsen Norman Group put it in its 2026 assessment, AI interviewers can run adaptive conversations at scale but inherit every ambiguity in their instructions (NN/g, 2026). Practitioner guides converge on the same conclusion: the instructions you give the AI are the bottleneck, not the technology (Great Question, 2026).
Key takeaways
- Write objectives, not just questions. For each section, state what you are trying to learn so the AI can recognize a complete answer.
- Keep it to 6–10 core questions. AI moderators handle longer guides, but participant fatigue and answer quality drop after roughly 10 questions (Great Question, 2026).
- Encode probing rules explicitly. Define how deep to dig and what a good answer looks like, because the AI cannot read body language.
- Define stop conditions. Tell the AI when a topic is exhausted, or it will either loop or move on too early.
- Semi-structured is the sweet spot. Fixed core questions plus latitude to follow up beats both rigid scripts and fully open prompts.
- Pilot before scale. AI amplifies a flawed guide across every session at once, so a bad guide is a bigger risk than in human research.
How an AI-moderated guide differs from a human moderator guide
If you hand an AI interviewer a traditional moderator guide — a list of questions and a few bracketed “[probe as needed]” notes — you will get a transcript that reads like a survey administered out loud. The decisions a human makes by instinct have to become written rules.
| Decision | Human moderator | AI moderator needs |
|---|
| When to probe deeper | Reads hesitation, curiosity, contradiction | Explicit rule: probe until a concrete example is given |
| When to move on | Senses the topic is exhausted | A defined stop condition per section |
| Tone and vocabulary | Adjusts to the participant naturally | Stated tone rules and banned jargon |
| Handling vague answers | Gently presses for specifics | Instruction: do not advance on abstract answers |
| Following an unexpected thread | Improvises a relevant follow-up | Permission and boundaries for off-script probing |
| Sensitive disclosures | Slows down, shows empathy, may pause | Safety and empathy instructions, escalation rule |
The anatomy of a strong AI-moderated discussion guide
A complete guide has six layers. The first three set the frame; the last three control the conversation turn by turn.
1. Study objective and participant context
One paragraph: what the study is for, who the participant is, and what decision the research will inform. This grounds the AI’s judgment about what is relevant when a participant goes off-script.
2. Tone and language rules
State the register explicitly. For consumer research, instruct plain language: say “how you feel about” instead of “your perception of,” and never use research jargon. Specify formality, sentence length, and whether the AI may use the participant’s name.
3. Section objectives
For each topic block, write the learning objective before the questions. “Understand how the participant currently solves X and where it breaks down” tells the AI what a complete answer contains — far more useful than a bare question.
4. Core questions (6–10)
Open, non-leading, one idea each. These are the anchors the AI must cover. Everything else is adaptive probing around them.
5. Probing rules
The part most guides omit. Specify probe depth, what counts as a concrete answer, and the maximum number of follow-ups before moving on. Example: “Probe until the participant describes at least one specific recent example. Do not move on if answers are only abstract or general.”
6. Stop conditions and edge cases
Tell the AI when to advance (“move on once the participant gives a concrete example or says they have nothing to add”), how to handle refusals or distress, and what is out of scope.
Original asset: the AI Interview Guide Readiness Checklist
Run your draft guide through this before any session goes live. Score each item Yes (1) or No (0). 11–13 is launch-ready; 7–10 needs a revision pass; below 7 will produce shallow transcripts at scale.
| # | Checklist item | Why it matters |
|---|
| 1 | Each section states a learning objective, not just questions | Lets the AI recognize a complete answer |
| 2 | Core questions number 6–10 | Guards against fatigue and quality drop |
| 3 | Every core question is open and non-leading | Prevents the AI from priming answers |
| 4 | Probe depth is defined per section | Stops the AI advancing on thin answers |
| 5 | “What a good answer looks like” is specified | Gives the AI a concrete bar to hit |
| 6 | A maximum follow-up count is set | Prevents probing loops and frustration |
| 7 | Stop conditions are explicit per topic | Controls pacing without a human present |
| 8 | Tone and banned jargon are stated | Keeps language natural for participants |
| 9 | Off-script threads have permission and boundaries | Allows useful tangents, blocks irrelevant ones |
| 10 | Distress / refusal handling is written in | Protects participants and research integrity |
| 11 | The AI moderator is disclosed before consent | Ethical and increasingly regulatory baseline |
| 12 | Out-of-scope topics are named | Keeps sessions focused and comparable |
| 13 | The guide was piloted on 3–5 real participants | Surfaces failure modes before they scale |
Writing probing rules the AI can actually follow
Probing is where AI-moderated interviews live or die. The qualitative-methods literature distinguishes several probe types — continuation (“then what happened?”), elaboration (“can you give an example?”), clarification (“what do you mean by…”), and steering (“you mentioned X…”) — each used relative to what the participant has already said (NN/g, 2026). Your guide should name which probes the AI may use and tie them to triggers.
A 2025 study using an AI-driven Wizard-of-Oz setup (GPT-4o suggesting follow-ups to a co-interviewer) found participants valued AI-generated probes as a “springboard” for deeper questioning — but flagged poor timing as the main weakness, with badly placed follow-ups disrupting flow. The researchers strongly preferred keeping human gatekeeping over which AI suggestions to use (arXiv 2509.12709, 2025). The lesson for guide writing: do not just tell the AI to “probe deeply” — tell it what triggers a probe and when to stop, so its timing matches the conversation.
A probe rule template
- Trigger: Participant gives an abstract or general answer with no concrete example.
- Probe: Ask for a specific recent instance (“Can you walk me through the last time that happened?”).
- Depth: Up to two follow-ups per question.
- Stop: Advance once one concrete example is captured or the participant says they have nothing to add.
Where Qualitati fits
Qualitati is an AI user research platform that runs AI-moderated interviews in text and voice, plus AI-moderated focus groups, conversational surveys, and ThemeLens thematic analysis. Its AI moderator is built to act on exactly the kind of structured guide described here: you define section objectives, probing depth, and stop conditions, and the moderator probes for specifics and advances when a section’s objective is met. For teams that want a human in the loop, Active Listener mode feeds a live interviewer real-time prompts and section tracking — the “springboard” pattern the 2025 study endorsed, with the researcher keeping gatekeeping control.
Because the platform also analyzes the resulting transcripts, a well-written guide pays off twice: cleaner, more comparable sessions make downstream thematic analysis and coding more reliable. You can start free with 30 credits, no credit card required, and pilot a guide on a handful of interviews before scaling.
Limitations and methodology cautions
A discussion guide does not make AI moderation appropriate for every study. Three cautions:
- AI amplifies guide flaws at scale. A leading question or a missing stop condition contaminates every session simultaneously — the error gets amplified, not averaged out (Pearson, 2025). Pilot first.
- Sensitive and exploratory topics still need humans. AI moderators struggle with grief, conflict, deep emotion, and genuinely novel terrain where the most valuable follow-up is unanticipated. NN/g recommends AI moderation for structured, well-bounded studies, not the hardest qualitative work (NN/g, 2026).
- Probing depth has a ceiling. Even a good guide cannot make an AI improvise a brilliant, context-specific probe a skilled human would. Treat the guide as a floor on quality, not a guarantee of depth.
Human-review note: claims about model behavior and probing efficacy reflect the cited 2025–2026 sources and should be re-validated against your own pilot data before you rely on them for high-stakes research.
Frequently asked questions
How long should an AI-moderated interview discussion guide be?
Aim for 6–10 core questions. AI moderators can technically handle more, but participant fatigue and answer quality drop after roughly question 10 (Great Question, 2026). Add depth through adaptive probing, not more scripted questions.
What is the difference between a discussion guide and a script?
A script fixes the exact wording and order. A discussion guide for AI-moderated interviews sets objectives, core questions, and rules, then lets the AI adapt phrasing and follow-ups — the semi-structured approach most practitioners recommend.
Do I still need to pilot the guide if the AI is doing the interviews?
Yes — more so. Because AI applies your guide identically across every session, a flaw is replicated at scale rather than caught mid-study by a human. Pilot on 3–5 participants and refine before launch.
How do I stop an AI moderator from asking leading questions?
Write every core question as open and non-leading, and add an explicit instruction not to suggest answers or agree with the participant. Leading phrasing in the guide is the most common source of biased AI follow-ups.
Can the AI follow up on something I did not anticipate?
Only if you allow it. Give explicit permission for relevant off-script probing, with boundaries (stay on the study objective; do not pursue out-of-scope topics). Otherwise the AI will stick rigidly to your questions.
Should participants know they are talking to an AI?
Yes. Disclose the AI moderator before consent. Beyond being an ethical baseline, transparency is increasingly expected under emerging AI regulations governing automated interactions.
Bottom line
The discussion guide for AI-moderated interviews is the lever that determines whether you get rich, comparable data or a pile of shallow transcripts. Write objectives instead of bare questions, cap core questions at 6–10, encode probing and stop rules the AI can follow, and pilot before you scale. The model is rarely the limiting factor — the guide is. Start free with 30 credits, build your guide, and run an AI-moderated interview, focus group, or conversational survey on Qualitati. Or view transparent pricing first.