Run unbiased interviews

What an unbiased interview actually is

"Unbiased interviewing" is often sold as a state you can reach — remove the bias, and what remains is objective assessment. That is not achievable, and chasing it tends to produce theatre rather than fairness.

A more useful definition: an unbiased interview is one where the differences in how you evaluate two candidates are traceable to differences in what they demonstrated, rather than to how they made you feel. You will still have reactions. The question is whether those reactions are doing the deciding.

That reframing matters because it makes the problem tractable. You cannot stop yourself warming to someone who reminds you of a colleague you like. You can make sure that warmth is not what your scorecard records.

The biases that actually show up in interviews

Most bias training covers demographic bias and stops. Those matter and they are legally consequential — but in practice the effects that most often change a hiring outcome are subtler, and interviewers rarely notice them happening.

The first-impression lock-in

Interviewers commonly form a view within the first few minutes, then spend the rest of the conversation gathering support for it. A candidate who opens badly gets harder follow-ups and less benefit of the doubt; a candidate who opens well gets their vague answers charitably interpreted. Same interviewer, same questions, different interview.

Fluency mistaken for competence

Articulacy is easy to observe and feels like a signal. For most roles it is a weak predictor of performance and a strong predictor of confidence. It systematically disadvantages candidates interviewing in a second language, candidates who are anxious, and candidates who think before speaking — none of which relate to whether they can do the job.

Similarity as a proxy for judgement

When a candidate has made the same technical choices you would have made, their reasoning feels sound. When they made different ones, it feels questionable. This is the mechanism behind most "culture fit" rejections, and it is how teams become homogeneous without anyone intending it.

Contrast against the previous candidate

The same answer scores differently depending on who you interviewed an hour earlier. Interviewers rate against the recent pool rather than against the bar for the role — which is why the third candidate on a bad day is often overrated.

What actually reduces interview bias

The evidence base here is unusually clear, and it points at structure rather than awareness. Four changes do most of the work.

Ask every candidate the same core questions

Unstructured interviews let each conversation drift somewhere different, which makes candidates uncomparable and gives bias maximum room. Fix a core set per role and ask it of everyone. Follow-ups should still be responsive — that is where the actual information is — but the spine stays constant.

Score against defined anchors, not a general impression

A 1–5 scale with no definitions collects sentiment. Write down what a 2 looks like and what a 4 looks like for each competency, in role-specific terms, and require the evidence that produced the score. If an interviewer cannot cite the moment, the score is an impression wearing a number.

Score during the interview, not afterwards

A score written twenty minutes later is a memory of a feeling, and memory reorganises itself around your conclusion. Scoring each competency as it is covered is the single highest-leverage habit change available, and it also surfaces coverage gaps while you can still fix them.

Collect assessments independently, before any discussion

Verbal debriefs are dominated by whoever speaks first and whoever is most senior. Once an interviewer hears that the hiring manager was impressed, their own recollection starts to shift. Independent submission before the debrief preserves information that would otherwise be destroyed in the first ninety seconds.

Notice what is absent from that list: reminding people to be objective. Awareness alone has a poor track record, because the biases above operate below the level you can observe in yourself.

Where AI helps — and where it introduces its own risk

It would be convenient for us to claim that an AI assessment is inherently neutral. It is not, and you should be sceptical of any vendor who says otherwise.

Models learn from data that carries human judgement, and a system trained to predict past hiring decisions will reproduce whatever bias produced them. There are well-documented cases of exactly this. Any tool that scores candidates deserves the same scrutiny you would apply to a human interviewer with an unexplained pattern in their decisions.

What AI genuinely helps with

• Consistency of coverage — every candidate is actually assessed against the same competencies, rather than whichever ones the conversation happened to reach.
• Evidence capture — the candidate's own words, attached to the score, so a debrief argues about what was said rather than who remembers it best.
• Removing the memory gap — scores generated during the interview cannot drift toward the conclusion you reached afterwards.
• Surfacing what you missed — an inconsistency between an answer and something said twenty minutes earlier, which no interviewer reliably catches while also holding a conversation.

What it should not do

• Make the decision. A tool that outputs hire/no-hire has moved from assisting judgement to replacing it, and it is now the thing nobody audits.
• Score on delivery. Assessment should rest on the substance of an answer, not on tone, accent, fluency or perceived confidence — those are the exact channels through which bias travels.
• Be unauditable. You should be able to see which answer produced which score. If you cannot inspect the reasoning, you cannot detect the bias.

A practical checklist

If you do nothing else, do these. They cost no budget and can be in place for your next interview.

• Write four or five competencies per role, with a defined description of what a weak and a strong answer look like.
• Fix a core question set and ask it of every candidate for that role.
• Score each competency as it is covered, not at the end.
• Record the evidence — what the candidate actually said — alongside every score.
• Submit assessments independently before any debrief happens.
• Open the debrief on the widest disagreement, not the overall verdict.
• Decide against the bar for the role, not against the other candidates you happened to see.
• Write down the reason for the decision, so you can calibrate against how the hire actually performs.

The honest catch is that doing all of this while running a conversation the candidate finds human is genuinely demanding — which is why structured processes decay back into impressions within a couple of quarters. That decay is the problem worth solving, and it is a tooling problem more than a discipline one.