AI Interviews That Ask Why: Follow-Up Questions That Reach the Real Example

Short answer

The first answer to an interview question is usually a summary: "communication is poor", "the system is slow", "the shelf was empty". The useful part comes after the follow-up: when it last happened, what the person did next, who else was involved and what it cost. A good follow-up uses the person's own words, asks for one recent example, and stays within the topics the organiser approved. An AI interviewer can ask those follow-ups consistently in every conversation. It should never lead, argue, or wander into topics nobody agreed to cover.

Why a form stops too early

A survey or form records the first answer and moves on. That works for asking a comparable question across a large population, as the employee survey limitations guide explains. It does not work for finding out why something happens. "The shelf was empty" is where a form stops. It does not say which shelf, how often, what the system showed, or what happened to the order.

The GOV.UK Service Manual guide to in-depth interviews suggests simple follow-up questions such as "you said… when/why/who was that?", "can you tell me more about…?" and "in what way…?". It also advises focusing on stories and real examples rather than generalities. Those three prompts do most of the work.

Five follow-ups that reach the example

A sixth is often the most valuable: "Is that typical?" It separates a memorable incident from a regular pattern without assuming either.

Use the person's words, not yours

Follow-ups work best when they repeat what the person said. "You said orders miss the 4 pm truck. Where do they get stuck?" shows the interviewer was listening. It also keeps the conversation on the participant's account. Swapping their words for the organiser's vocabulary ("so it's a capacity issue?") puts a conclusion in their mouth.

What a follow-up must not do

1. Lead. "Would you say management is to blame?" suggests an answer. Ask "what happened next?" instead.
2. Argue. If one account contradicts another, the interviewer records both. Correcting the participant is not the interviewer's job.
3. Leave the approved topics. A conversation about missed deliveries should not drift into someone's health, their colleagues' performance or their pay. The exception is a topic the organiser approved and the person raises.
4. Push past "I don't know". "I don't know" and "that's not my area" are findings. Repeating the question until something comes out produces invented detail.
5. Pry into other people's private matters. A follow-up can ask who else would know about a process. It should not invite gossip about a named colleague.

A fictional example: from complaint to evidence

A fictional warehouse operator asks pickers why orders miss the afternoon truck. One conversation runs like this:

• Interviewer: What usually makes an order late?
• Picker: Stock problems, mostly.
• Interviewer: You said stock problems. When did that last happen?
• Picker: Tuesday. The shelf was empty but the system said there were fourteen.
• Interviewer: Walk me through what you did then.
• Picker: I flagged it and waited for a recount. That took about forty minutes.
• Interviewer: Is that typical, or was Tuesday unusual?
• Picker: It happens most days in that aisle.

The first answer, "stock problems", could mean anything. Four follow-ups later there is a date, a location, a system discrepancy, a workaround and a delay. Each of these can be checked against a record. If seven of ten pickers give similar accounts, the organiser has a finding with evidence behind it, not an impression.

Consistency is the main advantage of an AI interviewer

Human interviewers vary. One probes every answer; another accepts the first summary when the day runs late. When the organiser compares ten conversations, some contain examples and others contain opinions. It becomes hard to tell a real difference between people from a difference in the interviewing.

An AI interviewer can ask for the last occurrence, the sequence and the consequence in every conversation, at whatever time each participant chooses to answer. It can also adapt: if a participant mentions an export or a photo, it can ask to see it. What it cannot do is decide what matters to the organisation. That is why the topics, and the limits on them, come from a person.

Check the follow-ups before anyone else sees them

Before inviting anyone, answer the conversation yourself. Watch for three things:

• follow-ups that lead;
• follow-ups that ask about something you did not intend;
• follow-ups that keep asking after a clear "I don't know".

If the topics produce any of these, change the wording. The pulse survey vs conversation guide covers when a conversation is the better tool in the first place.

Where Lontra fits

Lontra asks why in every conversation. Follow-ups adapt to what each person says, use their own words and stay within the topics you approved. It asks when something last happened, what people did next and what it cost. People can share documents, screenshots or photos, and Lontra asks about what it sees in them.

You can try the conversation yourself first, and your practice answers never appear in the findings. Every finding links back to the conversations behind it, so you can read the follow-ups that produced it. How it works shows the steps.

The test of a good follow-up

After the interview, a reader should be able to point to a specific event, a date or period, and what happened as a result. If a conversation produced only opinions, the follow-ups did not do their job, whoever or whatever asked them.