PM Mapped
Working with AI DISCOVERY · User Interviews

User Interviews with AI

AI helps at the edges of an interview — sharpening your guide beforehand and speeding synthesis afterward — but it cannot be in the room. The value of an interview is the unscripted human moment: the surprise, the contradiction, the thing you didn't think to ask. Use AI to prepare and to process, never to fabricate.

← Back to the User Interviews tool
1How to use AI for this

A step-by-step way to work

1

Pressure-test your discussion guide

Paste your questions and ask AI to flag leading or closed questions and suggest open, neutral rephrasings.

2

Rehearse difficult moments

Have it role-play an evasive or rambling interviewee so you can practise staying neutral and probing deeper.

3

Transcribe, then read

Use transcription so you're present in the conversation, not scribbling — but read the transcript yourself afterward.

4

Get a first-pass on themes

After several interviews, ask AI to cluster observations, then verify every cluster against the actual quotes.

5

Never let it invent

If AI 'summarises' a quote into something cleaner, discard it. Interview data must be what was actually said.

2Worked examples

Real prompts, and what good output looks like

Fixing leading questions before the interview
Your prompt

Critique these interview questions for bias and rewrite them to be open and neutral: 1) 'Don't you find the current checkout frustrating?' 2) 'How much do you love the new dashboard?'

What good output looks like

1) Leading — it assumes frustration and invites agreement. Neutral: 'Walk me through the last time you checked out. What was that like?'

2) Loaded — 'love' presumes a positive feeling. Neutral: 'Tell me about how you've been using the dashboard. What stands out?'

Both rewrites are open and past-/behaviour-focused, which gets you stories rather than opinions.

High-value, low-risk: catching leading questions before you ask them protects the whole study. This is AI at its best for interviews.
First-pass theme clustering (to verify)
Your prompt

Here are 12 observations from 4 interviews [pasted]. Cluster them into themes and label each. Mark anything that only one person said.

What good output looks like

Theme A – Onboarding feels long (4 mentions): …

Theme B – Unclear pricing (3 mentions): …

Theme C – Wants mobile access (1 mention — single source, treat cautiously): …

Note: clusters are a starting layout — confirm each grouping against the original quotes.

Watch the merges: AI will confidently combine things that don't belong together. The single-source flag is useful; the clusters are a draft, not a finding.
3Copy-paste template

A prompt you can reuse

Fill in the highlighted parts and paste it into your AI tool of choice. Edit the output — it's a starting point, not a finished answer.

Reusable prompt
Help me prepare for and synthesise user interviews. I'll use you before and after — not as a substitute for the conversation.

[Before] Here is my discussion guide:
[paste your questions]
Please flag any leading, closed, or double-barrelled questions and rewrite them as open, behaviour-focused questions. Suggest 2–3 good follow-up probes.

[After] Here are my raw observations/quotes from [N] interviews:
[paste notes]
Please cluster them into themes, label each, count mentions, and flag anything said by only one person. Do not rewrite or 'clean up' any quote — keep them verbatim.
4Common pitfalls

What AI gets wrong here

Fabricated or 'tidied' quotes

Asked to summarise, AI may produce a cleaner quote that no one actually said — corrupting your data.

Do this instead: Keep quotes verbatim. Use AI to cluster and count, never to reword what a participant said.

Over-merged themes

AI groups by surface similarity and will merge distinct issues into one tidy theme.

Do this instead: Check every cluster against the raw quotes; split anything that lumps different problems together.

Treating clusters as conclusions

A first-pass clustering looks authoritative but is just a starting layout.

Do this instead: Do the sense-making yourself — the insight comes from your reading of the transcripts, not the AI's grouping.
The judgment that stays yours

An interview's value is the live, human exchange — reading hesitation, following a surprise, asking the question you only thought of because of how someone's face changed. AI can make your guide sharper and your synthesis faster, but it can't conduct the conversation and must never invent what was said. You stay in the room; AI helps at the doorways.