The strange thing that happens when AI sits in on your interview
Here's a finding that should change how you prep for any AI interview. When people believe a machine is grading them instead of a human, they quietly rebuild their own personality. They crank up the analytical, spreadsheet-brain stuff and bury the warmth, the creativity, the gut instinct. Researchers gave it a name: the AI assessment effect.
The study behind it isn't some small lab poll. Three behavioural scientists ran 12 separate experiments with 13,342 participants and published the results in the Proceedings of the National Academy of Sciences. Same pattern, over and over. The moment people thought AI was doing the judging, they reshaped their answers to look more logical and less, well, human.
And the timing matters. This isn't a fringe worry for a handful of tech roles anymore. It's how most hiring works now.
How many interviews is AI actually judging? More than you'd think
AI screening has gone from niche to nearly universal in about two years. According to Harvard Business Review, 88% of companies now use some form of AI for initial candidate screening, and more than 90% use automated systems to filter or rank applications before a human ever sees them.
So when you sit down for a first-round video interview or a one-way recorded screen, the odds are decent that something automated is scoring at least part of it. You can feel that. And the research says the feeling alone is enough to change your behaviour.
It goes further up the ladder than screening, too.
| AI in Hiring (2025-2026) | Figure |
|---|---|
| Employers using AI for initial candidate screening | 88% |
| Using automated systems to filter or rank applications | 90%+ |
| Managers using AI to decide on raises, promotions, layoffs | 94% |
| Managers who let AI make the final call without review | ~1 in 5 |
Sources: Harvard Business Review (2025); ResumeBuilder manager survey (2025)
That ResumeBuilder data is the part that makes people nervous. 94% of managers who use AI tools said they use them for decisions about their direct reports, and roughly one in five admitted they let the AI make the final call without checking it. So the stakes are real. Which is exactly why candidates start performing.
What people actually do when they think AI is watching
Here's the heart of it. The PNAS team found that under AI assessment, people consistently push their analytical, logical, data-driven side to the front and tuck their emotional and intuitive side out of sight.
| Trait | What candidates do when they think AI is assessing them |
|---|---|
| Analytical reasoning | Play it up |
| Empathy | Hide it |
| Creativity | Hide it |
| Intuition and gut judgement | Hide it |
Source: "AI assessment changes human behavior", PNAS (2025)
The mechanism is something the researchers call a "lay belief". People just assume, without being told, that a machine cares about cold logic and has no time for the soft stuff. So they give the machine what they think it wants. Nobody instructs them to do this. They do it on their own.
The catch is that the belief might be wrong. A separate study in Nature's Scientific Reports looked at what people assume about AI judging interpersonal skills, and the gap between what candidates think AI rewards and what actually predicts good hiring is wider than most of us realise.
The problem: you're hiding the stuff that actually wins
Look, here's the part that should bother you. The traits people bury under AI assessment (empathy, creativity, the judgement calls you can't fully put into words) are the same ones the PNAS researchers point out often separate genuinely strong employees from the merely competent ones.
So you end up suppressing your best material to impress a system that may not even weight things the way you assume. It's like dressing down for a date because you've decided the other person hates nice clothes, when you've never actually asked them.
I've heard this play out more times than I can count. A candidate with a brilliant, messy, human story about talking an angry client off a ledge will flatten it into "I followed a structured escalation process and resolved the ticket within SLA." Technically fine. Completely forgettable. The version where they actually read the room and made a call is the version that gets remembered on the shortlist. They sand it off because they think the robot prefers sandpaper.
Why it gets worse: everyone's doing the same thing
Now scale that up. If most candidates believe AI rewards analytical traits, and most candidates respond by leaning hard into analytical traits, then most candidates start to sound identical.
The PNAS researchers flag this directly. They warn the effect drives a "more homogeneous" version of each person and risks flattening the whole talent pool into something uniform. When you're trying to get hired, uniform is the enemy. Blending in is how you get screened out.
Think about what the person reading the shortlist actually sees. Forty answers that all hit the same logical beats in the same flat register. The one that breathes, that sounds like a real person made a real decision, is the one that survives. Honestly, the candidates over-optimising for the algorithm are doing the recruiter's filtering for them.
There's a quieter cost too. The researchers note that when people edit themselves this hard, their "true capabilities or personalities may not be revealed". You can win a role by pretending to be a logic machine and then spend two years in a job that was hired for a version of you that doesn't exist. That's not a win.
How to stop performing for the algorithm
The fix isn't to ignore that AI might be in the room. It's to stop letting that change who you are in the room. A few things that genuinely help.
Answer like a specific person, not a category. Generic is what AI flattens you into, so fight it with detail. Names, numbers, dates, the actual decision you made and why. "We cut churn" is forgettable. "We cut churn from 9% to 5% in one quarter after I pushed to rebuild the onboarding email" is a person talking.
Keep the reasoning, not just the result. The intuition you're tempted to hide is often the most interesting thing about your answer. Why did you make that call? What did your gut say before the data confirmed it? That's signal, not fluff.
Use a structure, but don't let it strangle you. The STAR method exists to keep you concrete, not to turn you into a form. Situation, task, action, result, then a sentence of human reflection on top. Our behavioural interview questions guide walks through how interviewers actually score those answers.
Practise out loud. The robot voice creeps in when you're nervous and reading off a mental script. The more you rehearse answers as spoken language, the less you default to stiff, list-like delivery. If you're prepping for a recorded or AI video interview specifically, get used to talking to a lens without freezing up.
Don't try to game what the machine wants. You don't actually know how any given system is weighted, and the research suggests your guess is probably wrong. Spending energy second-guessing the algorithm is energy you're not spending on being clear, specific and memorable. For more on which "tells" are real and which are myth, our piece on the hidden AI test in job interviews is worth a read.
The bottom line
AI is in the hiring loop now, that part is settled. 88% of employers screen with it, and the number is still climbing. But the data also says something more useful than "be afraid of the robot". It says the biggest risk isn't the AI itself. It's what you do to yourself when you think it's watching.
Across 12 experiments and over 13,000 people, the pattern held: we make ourselves smaller, flatter and more generic for the machine. And generic is the easiest thing in the world to reject.
So don't. Be the specific, slightly imperfect, actually-human candidate. That version was always the one worth hiring.
Jacob, Instant Interview
