The Weakness Question, Answered Into a Camera

Why Does the Disguised Strength Fail Here?

Because everyone gives it, and a transcript makes that obvious in a way a room sometimes does not. “I care too much” and “I work too hard” are claims about character with no event attached, so there is nothing in them for a reviewer to check and nothing a follow-up can open up. The employer-side complaint is already written down: why every candidate now gives the same polished STAR answer.

Evidence sits nearby, and it comes with real limits. One study put 582 participants from ten countries through an eight-question asynchronous video interview for a mock bank manager position ¹.

Self-promotion, an honest tactic, was positively associated with rated interview performance, while extensive image creation, a deceptive one, was negatively associated with it ¹.

Read that as a direction, not a dial. The mock interview put nothing at stake, the tactic use came from self-report, and the ratings were human rather than machine, so nothing there describes how a scoring model treats invention ¹.

What it does not support is the idea that a polished fiction beats a plain truth.

The question itself is doing something narrower than candidates assume. A reviewer is checking whether you can describe your own work accurately, which is a small proxy for whether your other answers can be trusted. An answer with no flaw in it fails that check without ever getting to the flaw.

Naming the flaw is only the price of admission to the rest of the answer.

What the Recording Actually Carries

The words, and then the video, to two different readers. HireVue's 2025 explainability statement says its AI scores only the transcribed words of an answer, and specifically that it does not assess facial expressions, body language, background and surroundings, or tone of voice ². Vervoe's candidate help article says its AI reviews only the transcript for audio and video questions ³. Both statements describe the scorer, not the eventual reader.

Both are vendors describing their own products rather than audits, and each covers only its own platform. Vervoe's page adds the half that matters here: the video and audio are transcribed and provided to the employer along with the recording ³.

So the delivery is out of the algorithm, and not out of the picture.

A rueful half-smile does not travel into a transcript. Neither does the pause you would use in a room to signal that you know the thing you just said is uncomfortable. Whatever work that softening was doing, the sentences have to do instead, which is why the correction belongs in the same answer rather than in the follow-up you are hoping for.

Being specific matters more here than sounding composed.

That cuts both ways. Sounding nervous while admitting something is a delivery problem, and the vendors above say their scorers are not reading delivery ²³, though a person at the employer can still watch the recording ³. What an AI interview scores and what it does not goes through the rest of what is and is not being read.

Pick Something Adjacent to the Job, Not Central to It

Choose something true that sits next to the work rather than under it. A project manager whose deadlines slip, or an auditor bored by detail, is naming the job requirement itself, and no correction rescues that. An adjacent weakness is one a competent person can carry while still doing the role well, which is most real weaknesses. The test is whether saying it out loud would cost you the role or only a little face.

• Applying for an analyst role: you put off the writeup once the interesting part of the analysis is done.
• Applying for a design role: you take blunt feedback as a verdict on the whole direction and need a night before you can use it.
• Applying for a support role: you escalate late because you would rather solve it yourself.

Two categories are poor choices for this answer. Anything that reads as a reliability risk, meaning attendance, honesty or temper, gives a stranger reading a transcript nowhere safe to land. And anything still unresolved leaves the answer as a bare admission, which is the version this format handles worst.

The answer also outlives the conversation. HireVue's candidate FAQ says it stores your personal data for a retention period chosen by the company you interviewed with, and purges it automatically when that period expires ⁴.

The same page says you can request deletion at any point in the process, and that the company applied to decides it rather than the vendor ⁴.

The page names no default period and no maximum, so the length is unknown to you unless the employer says. How long a company keeps your AI interview video covers what to ask for and who to ask for it.

A weakness nobody could ever act on is not a weakness, it is a hedge.

Pick something you would still be comfortable seeing quoted back a year from now.

How to Close the Answer, and What Comes After

Give one sentence to the flaw and the rest to the change. Name it plainly, say what you now do differently, then give the evidence that it held: a date, a habit that stuck, a specific thing that stopped happening. The evidence sentence is the part most candidates leave out, and it is the part no list of safe weaknesses can supply. Two sentences about the change are worth four about the flaw.

A working version runs about like this. “I used to wait until an analysis was finished before telling anyone where it was heading, which cost my last team a week on a pricing model. Since March I send a two-paragraph status every Friday whether the work is done or not. The last three projects got redirected early instead of at the end.”

One flaw, one change, one piece of evidence, in under a minute.

Write the change down before the interview, and let the flaw take one line.

If the interviewer is a live AI rather than a recorder, the second half may be asked back to you. Alex, formerly Apriora, markets an AI recruiter that runs live phone and video interviews, and its candidate page describes the AI summarising answers back and asking follow-up questions based on them, and a testimonial there says the answers then go to a human recruiter ⁵.

That description is the vendor's own, drawn from selected testimonials, and what any employer switches on will vary ⁵.

A follow-up asking what changed is only hard if you picked a weakness you have not actually worked on.

Interviewers get the mirror-image instruction, which is to push on an answer rather than take it as given. What follow-up questions expose about whether someone understands their own answer is written for interviewers rather than for you, and it asks for the same kind of specifics your evidence sentence should already hold.

See how it works >