There’s a piece of advice about AI interviews that I hear repeated all the time, and it makes people far more anxious than they need to be. It goes something like this: the AI is watching your face, reading your body language, working out whether you’re nervous or lying. So sit still, hold eye contact with the camera, and don’t fidget.

I understand why people believe it. A few years ago, it was closer to the truth. But the tools have moved on, and in some places the law has moved on too. If you’re preparing for an AI interview based on the old picture, you’re preparing for the wrong thing.

Let me clear this up, because the reality is actually easier to prepare for.

The face-reading era is largely over

The best-known name in this space is HireVue. For a while, it did analyse candidates’ facial expressions as part of its scoring. That’s the origin of the whole “the AI is judging your face” idea.

Then, in early 2021, HireVue dropped facial analysis altogether. An independent audit and a good deal of public pressure led to it. Its own research had found that the visual data added almost nothing to the result. What the tool relies on now is what you actually say, analysed through language processing, not the expressions you make while saying it.

So the thing people are most worried about, a camera silently marking them down for looking nervous, is not how the mainstream tools work anymore. Your words carry the weight.

In some places, face-reading is now banned outright

This is worth knowing, because it tells you the direction things are heading. Under the EU AI Act, using AI to infer someone’s emotions in a hiring process has been prohibited since February 2025. Not discouraged. Prohibited, with serious fines attached.

The line the law draws is a useful one to hold in your head. Turning your spoken answers into a transcript is fine. Claiming to read your inner emotional state from your face or voice is not. Words, yes. Feelings, no.

Put the two things together, the tools stepping back from face analysis and the law stepping in to ban emotion detection, and a clear message emerges. What you say matters far more than how composed you look while saying it.

So what should you actually prepare?

This is the good news. Preparing strong answers is something you can control, practise and get better at. Managing your micro-expressions on camera is not, and thankfully you don’t need to.

The tools are listening for substance. They map what you say against the role and the company: the skills, the experience, the keywords that show you understand what the job needs. Short, thin answers give them very little to work with. Vague ones do the same.

The structure I coach my clients to use is CPR: Context, Process, Results. Set the scene briefly, explain what you did, then land on the outcome. And make that outcome real. A number, a percentage, a before and after. “I improved the process” tells the tool almost nothing. “I cut the reporting time from three days to four hours” gives it something solid to record and match.

The human still reads the transcript

Here’s the part people forget. The AI interview usually produces a transcript and a summary of your skills and experience, and a human being reads it. Often that same human wants to meet you afterwards to judge whether you’d fit the team.

So you’re not really performing for a machine. You’re creating a clear, well-structured record that a person will use to decide whether to take you further. That’s a much less intimidating way to think about it, and it points you at the right preparation: clear stories, real results, and answers that show you understand the role.

Stop rehearsing your eye contact. Start rehearsing your examples.