Use AI to Rehearse a Customer Interview

Before a customer interview, an independent creator may want to practise asking a difficult follow-up, explaining the session or resisting the urge to pitch. An AI conversation offers a place to attempt those moves before taking up someone else’s time.

The useful boundary is what the exercise can establish. A simulated customer can supply material for a rehearsal. The resulting dialogue is not testimony from someone who has used your service, faced the problem or decided to buy. Keeping that boundary visible lets you use the exercise without turning an imagined conversation into a business finding.

Why a convincing persona needs careful interpretation

In research published online on 17 May 2024, James Bisbee and colleagues tested ChatGPT-generated personas against responses from the 2016–2020 American National Election Study. Synthetic average opinions could resemble the survey averages, while variation and relationships between respondent characteristics and answers differed. Results also changed with prompt wording and across collection dates.

This was political-opinion research using models available in 2023. It was not an evaluation of every current AI system or a test of customer-interview rehearsal. Its relevance is methodological: convincing responses and plausible averages do not, by themselves, demonstrate that generated people represent a target population.

For a creator, a practical inference follows. Treat a fictional buyer’s objection as a possibility to investigate. Do not count it as evidence that a particular share of your audience has that objection. Generating more imaginary respondents does not turn their answers into independent observations of customers.

Choose one interviewing skill to practise

Make the rehearsal objective something you control. You might want to ask a neutral opening question, follow up on a vague answer or give the other person room to finish. Trying to validate the entire business in the same exercise makes the result harder to interpret.

The GOV.UK guide to in-depth interviews, published on 21 February 2017, recommends preparing a discussion guide, trying the structure with a colleague, using open questions and seeking real examples. An AI rehearsal can be an additional preparation exercise. The guide’s recommendation to practise with a colleague should not be presented as an endorsement of replacing participants with AI.

Write down what you intend to improve. “Ask for a concrete example before proposing a solution” gives you a behaviour to review afterwards. “Get positive feedback about my idea” would train a different kind of conversation.

Give the simulation a limited, fictional brief

Describe the situation you want to rehearse and label it as invented. Keep the brief narrow enough to inspect: a fictional freelance photographer managing enquiries, for example, rather than a claim to reproduce all photographers’ needs.

Here is a starting instruction you can adapt:

Act as a fictional photographer for interview practice. I am researching how enquiries are handled. Give brief answers, and sometimes leave a detail unclear so I can practise follow-up questions. Do not validate my product or claim to represent real photographers. After five exchanges, leave the role and identify where I asked a leading question, interrupted the topic or missed a follow-up. Label all scenario details as fictional.

This is a proposed practice exercise, not a research protocol validated by the survey study. A model may not follow every instruction reliably. Review what actually happened rather than assuming that the persona or its feedback performed as requested.

Review your questions, rather than scoring the imaginary buyer

After the exchange, examine the interviewer’s behaviour. Useful review questions include:

• Did I ask about a specific situation, or invite a broad opinion?
• Did my wording suggest the answer I hoped to hear?
• Did I pursue an unclear detail before moving to another topic?
• Did I start describing features before understanding the situation?
• Which question would I rewrite before speaking to a real participant?

The model’s debrief can suggest issues to inspect, but it should not be the sole judge of the exercise. Read the exchange yourself or ask a colleague to review it. Record one change to your discussion guide, then try the revised question.

Turn fictional answers into open research questions

Suppose the fictional photographer says that enquiries become difficult when several customers ask for the same weekend. That is a generated scenario. It might prompt you to ask an actual photographer about the last time competing enquiries occurred.

Avoid carrying the scenario into the real interview as an assumed fact. “How frustrating is it when customers compete for the same weekend?” presupposes both the event and the frustration. A more open starting point is to ask the participant to describe how their most recent enquiry was handled.

Maintain a short separation between three kinds of material:

• Simulation: a detail invented during the rehearsal.
• Research question: what you now want to learn from relevant people.
• Observation: what an actual participant said or did, recorded with its context.

A detail should enter the observation record because it was encountered in research. Agreement between several AI personas is still agreement within generated material.

Gather evidence from people with relevant experience

GOV.UK’s research-planning guidance advises turning assumptions into research questions, identifying the user groups involved and choosing activities that can answer those questions. That provides a useful next step after rehearsal: decide who you need to learn from and what evidence would inform your decision.

MaryChuks’s guide to customer interviews that investigate a real problem explains how to explore recent behaviour and workarounds. Use your improved discussion guide in those conversations, staying open to answers the simulation never suggested.

If the decision concerns whether someone can use a prototype, observe relevant people attempting the task. If it concerns willingness to buy, recognise that an interview answer and an actual purchasing decision are different forms of evidence. Match the activity to the question rather than asking the rehearsal to settle everything.

Stop practising when it is time to learn

Set a modest endpoint: one improved opening, a few clearer follow-ups and a discussion guide ready to try. More simulated conversations can always produce more material. The useful test is whether the rehearsal has prepared you to listen better in the next real session.

AI can have a defined role in preparation while customer evidence keeps its own identity. Practise the interview, document the assumptions it reveals and let real encounters change what you believe about the product.

Sources

• Bisbee and colleagues, Synthetic Replacements for Human Survey Data? The Perils of Large Language Models, Political Analysis, online publication 17 May 2024.
• GOV.UK Service Manual: Using in-depth interviews, published 21 February 2017.
• GOV.UK Service Manual: Plan user research for your service.