AI Interviews May Be
Less Biased.
Why Don’t Candidates Believe It?

That response is interesting because most people have experienced bias during recruitment. They know interviewers form quick impressions. They know two interviewers can assess the same candidate differently. Yet they still trust a human interviewer more than a machine.

Interviewers also assess candidates through their own life experiences.

Someone who built a career by staying with one company may see frequent job changes as a lack of loyalty. Someone who progressed by changing companies may see the same career as ambitious.

People value strengths similar to their own. They are also more likely to notice weaknesses they have personally struggled with.

Then comes the pressure to break the ice.

Interviewers are told to begin with small talk and make the candidate comfortable. Sometimes that works. Sometimes the conversation becomes awkward. The candidate gives a short answer, does not smile enough or fails to respond enthusiastically to an irrelevant joke.

The interviewer switches off within two minutes.

The remaining interview may last another 30 minutes, but the decision has largely been made. The questions then become a search for evidence supporting that first impression.

Interviewers observed candidates completing a group exercise and predicted who would become an effective officer. They formed clear impressions and felt highly confident about their predictions.

The predictions had little relationship with what happened later.

Kahneman redesigned the process. Instead of asking interviewers to form an overall impression, he asked them to assess candidates separately against defined traits. Each trait had specific questions and an independent score. Combining those scores produced better predictions than relying on intuition.

You can see a more entertaining version of this in The Traitors. Participants spend days living together, observing one another and discussing who might be the traitor. They then confidently vote out another innocent person.

This happens repeatedly.

Research now places well-designed structured interviews among the strongest available predictors of future job performance.

Most recruiters and managers already know this.

The problem is following it.

Designing good questions and scoring criteria requires serious preparation. Listening carefully, recording evidence and scoring every answer requires mental effort. Repeating the same process across ten candidates becomes tiring.

Human beings also get bored easily.

They want some variety and fun, even during an interview. They change the questions, follow interesting tangents, skip areas already discussed or spend ten minutes talking about a common interest.

It also does not compare a candidate’s career with its own career.

This concern will become stronger for senior roles.

AI already has a reputation for hallucinating and confidently producing incorrect answers. People naturally question whether that machine should assess their capabilities or career.

There is also no real small talk, rapport or personal connection. Candidates may finish the interview without feeling heard, understood or validated.

But this creates another difficult question:

Potential is not clearly defined. Human interviewers often assess it through confidence, communication style, familiarity and resemblance to successful people they already know. That can feel more personal without necessarily being more accurate.

AI can also contain bias. Poor questions and biased criteria will produce poor assessments. But AI interviews leave transcripts, scores and results that can be reviewed across hundreds of candidates. Patterns can be detected and criteria can be corrected.

Human bias is usually hidden behind statements such as “not the right fit” or “something did not feel right.”