AI Interview in UX: Prepare the Script and Review the Evidence

An AI interview can conduct questions and probes with real participants. To produce useful research, it needs an objective, a script, and review. Conversation automation does not transform every answer into sufficient evidence for a decision.

Define What the Conversation Should Clarify

Imagine a fictional team management software. The research aims to understand how managers perceive a delay and decide when to intervene.

A topic like “tell me about the last time a deadline changed” opens space for an episode. Asking only “would you like smart alerts?” tends to produce opinions about an abstract proposal.

Write down what each topic should investigate and which subjects are out of scope. In UXTap, the AI interview block allows you to configure context, topics, probes, and duration.

Choose Conversation by Topics or by Screens

A topic-based interview can explore routines without presenting an interface. A screen-based interview adds visual material to the investigation.

In UXTap, screen-based conversation can follow a script or track navigation through image links, depending on the configuration. It can also relate questions to specific screens.

Choose this format when the evidence depends on what the person is seeing or trying to do. If the question is about the work process prior to the product, showing the solution too early can limit the conversation.

Conduct a Moderation Pilot

Respond to the study as a member of the audience, using short answers, questions, and plausible deviations. Check if the AI asks for useful explanations or insists on a topic that has already been clarified.

Review especially:

• Questions that presuppose a problem.
• Probes that suggest an answer.
• Topic changes without context.
• Lack of opportunity to conclude.
• Instructions that do not match the presented screen.

In UXTap, voice and text depend on the interview configuration. The real-time voice mode has its own operating conditions and a turn-based alternative. Test the intended device and connection, without promising an identical experience in every environment.

Treat Transcription and Synthesis as Reviewable Material

After data collection, consult the conversation turns and available evidence. A transcription error in a product term can alter the meaning of a passage.

Separate what the participant stated from the interpretation produced in the synthesis. If the AI summarizes “the manager wants more alerts,” go back to the account and check if they asked for notifications or just more clarity on responsibilities.

AI features depend on the plan, configuration, and usage limits. Do not present the interview as an automatic benefit of the Free plan nor as a complete replacement for the researcher's work.

Look for the Episode that Supports the Recommendation

In the delay example, perhaps the manager reports only noticing changes by asking individuals. This might suggest a need for shared visibility.

Before proposing a new alert, investigate how they track work today and what information is missing. A conversation can open a hypothesis; a task with a prototype can help evaluate a solution.

Keep divergent accounts. One manager might want frequent tracking, while another needs to reduce interruptions.

A Script Centered on a Delay Episode

In the fictional team management tool, start with a concrete experience: “Tell me about the last time an important delivery was delayed.” The answer situates people, information, and timing. Asking first “what do you think of smart alerts?” would shift the conversation towards a solution suggested by the team.

Organize topics that can deepen the episode without presupposing the explanation. Ask how the person perceived the delay, what information they consulted, who they talked to, and what they decided to do. If the account is generic, ask for a specific situation. If it's already detailed, avoid repeating the same question with different words.

Also prepare boundaries: the conversation should investigate work tracking, without asking for colleagues' names, client data, or unnecessary internal documents. Explain the expected type of participation and use demonstration situations when presenting screens. This helps the participant understand what they can report without turning the interview into access to company information.

Conduct a Pilot that Challenges Moderation

Test short answers, contradictions, and requests for clarification. Respond “I don't remember,” change the subject, and mention an episode that doesn't involve the imagined functionality. Observe if the AI accepts the limitation, carefully returns to the topic, or insists on obtaining a confirmation that the participant did not offer.

Check if the probes maintain neutrality. After “I noticed the delay in a meeting,” a useful question is “what happened in that meeting?” Asking “would an alert have prevented the problem?” anticipates a solution and creates a hypothesis that might not have existed in the account. Review context and topics if this behavior repeats.

In screen mode, check if the question matches what is visible. A conversation about a previous state can confuse the person after navigation. The pilot should include forward, backward, and interruption, in addition to a passage through the expected path. Record limitations of the voice environment and connection before inviting the audience.

Analyze the Conversation, Not Just the Final Answer

Read the question that preceded each relevant statement. A spontaneous declaration and an agreement after an AI suggestion have different contexts. Preserve this distinction in the synthesis, especially when the passage will be used to advocate for a feature.

Check transcription, important terms, and the sequence of turns when in doubt. An omitted negative or an incorrectly transcribed technical name can invert the interpretation. Do not treat a long conversation as richer evidence by definition: it may contain repetitions, deviations, or probes that did not answer the research question.

Transform the episode into a product hypothesis with explicit scope. In the case of a delay, the difficulty might be in comparing information distributed across tools, rather than receiving more notifications. The user journey map can help position the account between stages and dependencies before proposing a solution.

Start with a Small Script

Choose a few topics linked to a decision and prepare criteria for reviewing the conversation. Conduct the pilot before recruiting.

In UXTap, configure the context and questions, test the participation mode, and read the generated turns. After data collection, select an episode with sufficient evidence and turn it into a next investigation. Quality comes from the link between question, answer, and decision.

Prepare for review with the guide on analyzing open-ended responses with AI and check the UXTap interview modes before choosing how to conduct the conversation.

Frequently Asked Questions about AI Interviews

Does AI Replace Researcher Review?

No. The team remains responsible for formulating the question, evaluating the moderation, and interpreting the material. Automation can help conduct conversations within a script, but it does not transform every answer into reliable evidence nor eliminate the need to review questions that may have led the participant.

Is it Better to Ask About the Future or About a Lived Experience?

Recent experiences often offer events that can be explored in detail. Questions about future use can reveal expectations, but they do not demonstrate what the person will actually do. Identify this difference in the analysis and avoid presenting a declared intention as observed behavior.

When Should I Prefer a Human Interview?

Consider human moderation when the situation requires delicate adaptation, complex clarifications, or careful monitoring of sensitive topics. The decision depends on the context and the risk of inappropriate moderation. A pilot can reveal that automation does not meet the depth or care required for that investigation.

Create your UXTap account to prepare your first study.