Interviewing
What to Say When the Question Is About Something You Have Not Done
Say plainly that you have not done that specific thing, in one sentence, then spend the rest of the time on the closest thing you have done and how you would approach the real one. That gives a rater something to grade against the competency behind the question. Do not invent an example. In a mock video interview study, deceptive tactics were associated with lower ratings, and a live AI interviewer can follow up on what you say.
The takeThe advice that fails people here is "never say I do not know." It turns the answer window into a bluff, and the bluff is the part a follow-up question takes apart. The opposite advice, "just be honest," stops one sentence too early: honesty about the gap is the setup, not the answer. What earns the rest of the time is the nearest true experience and a defensible approach to the real problem.
Where Olive fits
Open a role and see what the work shows
An Olive assessment is not an interview: it is a work assignment done openly with an AI assistant, on your own clock, and a person writes or edits the findings on what happened, in words rather than a number. You are granted the same report the employer reads, free.
Assess your shortlistWhat do you say while the timer runs?
Three parts, in order. Say in one sentence that you have not done that specific thing. Give the nearest true thing you have done, with the kind of detail only someone who was there would have. Then say how you would approach the real version, and name the first thing you would need to find out before starting. That is a boundary, an example and a method, and it fits inside a short answer window.
- The boundary. "I have not led a team through a reorganisation."
- The nearest true thing. "I did coordinate four contractors through a scope change that halved the project, and two of them were about to lose most of the work."
- The approach. "For a reorganisation I would want to know who had already been told and what was still open, because that changes what you can honestly say to people."
The first sentence is short, and it does not apologise.
Most of the timer belongs to the second and third parts. A rater is looking for evidence of a competency, and the competency behind that question was probably not reorganisations. It was keeping people steady through a change nobody chose. Look for that, or something close to it, in your own history.
A missing line on the list is not the same as a missing capability.
How much room you have depends on which format you are in. A recorded one-way interview gives a fixed window and nobody to ask what the question means. A live AI round will keep talking, which is more forgiving in one way and less in another. What actually happens in each kind of AI interview is worth reading before the invitation lands.
Why not just invent an example?
Inventing a project you never ran is tempting, and the detail underneath it is fragile. In a mock asynchronous video interview taken by 582 participants across ten countries, honest tactics such as self-promotion were positively associated with rated performance 1. Deceptive tactics such as extensive image creation were negatively associated with it 1. Human raters scored it and candidates reported their own tactics, so treat that as an association rather than a scoring rule.
A borrowed story has no second layer to go to.
That is where it comes apart. The follow-up asks for the date, the name of the person who disagreed, the number that made the decision hard. An answer you lived through has those. An answer you assembled last night has whatever you assembled, and nothing underneath it.
There is some evidence that past-behaviour questions resist this. One experiment found past-behaviour questions, the "tell me about a time when" kind, harder to fake than situational ones 2. The same paper found that follow-up questioning increased faking, so the picture is not tidy, and the study ran with undergraduates in an in-person format long before AI scoring existed.
Nothing in that experiment tested a machine reading a transcript.
At least one live AI interviewer is built to follow up as well. Alex, which sells an AI recruiter for live phone and video interviews, says on its own candidate page that the AI asks follow-up questions based on your answers and summarises them back to you 3. The page is Alex's own, built from selected testimonials, so take the mechanics as a warning rather than a specification.
On the employer side the follow-up is a designed instrument. What follow-up questions expose about understanding sets out a ladder interviewers are told to run: specify, invert, falsify. A small true example has something to give on each rung, because you were there. An invented one runs out early.
Pick the nearest true thing you have done
Pick on the shape of the problem, not on the job title. The question names a situation: a deadline that moved, a decision made without enough information, a conflict nobody wanted, a system you had to learn in a week. Look for something in your history with that shape, even if it was smaller, in another industry, or in a class. The rater is testing a competency, and competencies travel further than titles do.
- Same problem, smaller scale: you have not run a migration for eight teams, but you did move one team's data over a weekend.
- Same problem, different domain: you have not managed a department budget, but you did ration a fixed grant across a term.
- Same problem, from the next seat: you were not the one deciding, but you sat beside the person who was and you know what the call cost.
- Same problem, other end: you have not built the forecast, but you have been the person handed one that was wrong.
Say the difference in scale out loud, before anyone has to ask.
Naming the scale plainly costs little and it protects the rest of the answer. "This was three people and one quarter, not a company reorganisation" keeps the example honest and lets a rater judge the reasoning on its own terms. It also removes the thing a follow-up would otherwise go hunting for.
Specifics are what carry a small example. Why every candidate gives the same polished STAR answer is the employer-side version of this problem. The advice there is to ask for the option that was rejected, the person who disagreed, the baseline under the number. A true small event has those. An invented one has the arc and nothing else.
A small true story carries further than a large borrowed one.
What if you freeze and say almost nothing?
On at least one platform, a very thin answer is not counted as a zero by the AI. HireVue's 2025 explainability statement says a response without enough content in it is not scored by the AI at all. It is flagged for a person to review instead, and the same applies when the timer or a microphone problem cuts an answer short 4. That is one vendor's documented process.
Flagged for review is a routing decision, not a grade. The vendor calls it a small proportion of candidates and publishes no rate for it 4. Other platforms may simply score a thin answer low. None of that makes an empty answer window a plan.
What gets scored on that platform is your words. HireVue's same statement says its AI uses no video analysis and does not assess facial expressions, body language, background or tone of voice 5. The score comes from the transcript, benchmarked against expert human raters on answers to similar competency questions 5. That account is the vendor's own, with no independent audit behind it.
Filling the time with something true is not wasted, even when it is not the thing asked for.
If something technical cut you off, say so to the recruiter in writing afterwards, naming the question and what happened. That is a checkable claim about one response rather than a complaint about the process, and it is the kind of thing a person on the other end can act on.
Silence gives a rater nothing to work with.
Common questions
Is it fine to say I have not done that in an AI interview?
Yes, and it works better as a first sentence than as the whole answer. Name the gap in one sentence, then use the rest of the time on the nearest thing you have actually done and on how you would approach the real one. That gives whoever reviews the answer something to grade against the competency behind the question. Stopping at "I have not done that" leaves the rest of the window empty, which gives a reviewer little to work with on any format.
Should I invent an example if nothing in my history is close?
No. In a mock asynchronous video interview run across ten countries, deceptive tactics such as extensive image creation were associated with lower ratings, while honest self-promotion was associated with higher ones. That study used human raters and self-reported tactics, so it shows an association rather than a rule about how a scoring model behaves. The practical problem is separate: a live AI interviewer or a human second round can ask a follow-up, and invented detail has nothing underneath it.
What if the question is about a tool I have never used?
Say you have not used it, name the closest tool you have used, and describe what you did with it in specifics. Then say what you would want to check first about the new one: what the team uses it for, what it connects to, and where its output goes. A tool question often stands in for a broader competency, though some roles do need the tool itself.
What if I run out of things to say before the timer stops?
Spend the remaining time on method rather than on filler. Say what you would need to find out, who you would ask, and what would tell you the answer was wrong. On HireVue the vendor says a response without enough content in it is routed to a person to review rather than scored by the AI, which is that platform’s documented process and not a rule everywhere. A short true answer still beats a long invented one.
Does a hypothetical answer count for anything?
It counts, though on its own it is the weakest of the three parts. Structured interviews use both situational questions, the "what would you do if" kind, and past-behaviour questions, the "tell me about a time when" kind, and one experiment found the past-behaviour kind harder to fake. So give the hypothetical when you have nothing lived to offer, and pair it with the nearest real thing you have done whenever you can.
References
- 1. Investigating impression management use in asynchronous video interviews across 10 countries doi.org Supports the claim that honest tactics such as self-promotion were positively associated with rated interview performance while deceptive tactics such as extensive image creation were negatively associated with it, in a mock asynchronous video interview taken by 582 participants across ten countries.
- 2. Measuring faking in the employment interview: development and validation of an interview faking behavior scale. pubmed.ncbi.nlm.nih.gov Supports the claim that past-behaviour questions were found harder to fake than situational questions, and that the same experiment found follow-up questioning increased faking.
- 3. Candidate Experience | Why Candidates Love Interviewing with Alex AI alex.com Supports the claim that a live AI interviewer asks follow-up questions based on the candidate’s answers and summarises those answers back, as described on the vendor’s own candidate page.
- 4. 2025 Explainability Statement hirevue.com Supports the claim that a response the system finds too short, or cut off by time or a microphone problem, is not AI-scored and is flagged for human review instead, and that the vendor describes this as a small proportion of candidates without publishing a rate.
- 5. 2025 Explainability Statement hirevue.com Supports the claim that the AI scores the transcribed words only, uses no video analysis, does not assess facial expressions, body language, background or tone of voice, and is benchmarked against expert human rater evaluations of answers to similar competency-based questions.
5 sources, numbered by first appearance. How Olive sources claims
General guidance for hiring teams. What works at one company and one volume may not transfer to yours.
Olive assesses how a person works with AI. It does not detect AI-written documents, and it never produces a score, a ranking, or a match percentage for a person. Candidates read the same report the employer reads.