Reducing bias in hiring: shortlisting, interviews & AI

Why bias is built into most hiring processes

The CV was only invented as a standard document in the 1980s, when fax machines made it practical. Before that, we filled in application forms. Neither format was designed with bias in mind — and both carry it. Research into shortlisting consistently shows that identical credentials attract different scores depending on the name at the top of the page, the language used to describe achievements, or whether a photograph is included.

The interview is no better in its traditional form. As I often say, the typical hiring process is "two people lying to each other about the demo mode of the game, rather than actual game footage" — the candidate performs their best self, the hiring manager describes an idealised role. Neither side learns much about whether the person will actually thrive in the work.

And panels introduce their own dynamics. Without structure, the loudest or most senior voice shapes the outcome. Affinity bias pulls people towards candidates who remind them of themselves. Halo and horn effects mean a strong or weak answer early in an interview colours everything that follows.

Structured shortlisting: what it means in practice

The single most effective change most organisations can make is to define their criteria before opening applications — not after. Once you have seen a strong candidate, your brain subtly rewrites what the role requires to fit them. Criteria set in advance resist that drift.

• Name-blind and photo-blind screening. Remove names, photos and addresses at the shortlisting stage. This is straightforward to implement and reduces the impact of name-based bias and affinity.
• Score against the criteria, not the person. Each shortlister scores independently before discussing. Comparing scores first — rather than impressions — surfaces disagreement and forces it to be reasoned through.
• Check what the criteria are actually measuring. Ask yourself whether each requirement genuinely predicts success in this role, or whether it is a proxy for something else. Demanding a degree for a role that does not need one, or penalising informal English in written applications, can be a form of indirect discrimination dressed up as a standard.
• Watch the language in job adverts. Gender-coded vocabulary — words like "aggressive targets" or "nurturing environment" — nudges certain groups away before they even apply.

Structured interviews: removing the luck of the draw

The same logic applies in the room. Forward-looking organisations are investing in training their interviewers and working to avoid groupthink or dominance within an interview panel — where each person rates and scores responses against objective criteria independently, before any discussion. Only then do scores get consolidated and compared.

• Same questions, same order, for every candidate. This is the foundation of a fair interview. It means you are comparing like with like.
• Behavioural and skills-based questions over hypotheticals. "Tell me about a time you..." gives you evidence of actual behaviour. "What would you do if..." mostly tests composure under questioning.
• Practical tasks and work samples. Ask candidates to solve a real problem, write something, review something, or demonstrate a skill. A day's work-trial or a structured task tells you far more than an hour of conversation — and is considerably harder to fake.
• Diverse panels. A panel that shares the same background, seniority and experience will share many of the same blind spots. Bringing in different perspectives — and agreeing in advance that each panellist will name bias when they see it — makes for better decisions.
• Score before you discuss. Have each panellist record their score independently immediately after the interview. Collective deliberation before individual scoring allows one strong voice to anchor the whole panel.

One pattern to be especially alert to: we tend to go for the candidate with the extrovert personality, or the person who reminds us of whoever we liked in a similar role before, or someone who fits the existing team dynamic. None of those instincts tells you much about whether someone will do the job well. The question to keep returning to is: what does this role actually need?

AI in hiring: promise, risk and the human in the loop

Artificial intelligence is now present across the hiring pipeline — in applicant tracking systems that filter CVs, in tools that score video interviews, in chatbots that screen candidates before a human is involved. The promise is efficiency and consistency. The risk is that AI trained on biased historical data will reproduce and entrench that bias at scale, faster than any human panel could.

The most cited example is the kind of screening tool that learns from past hiring decisions. If your organisation historically hired more men into technical roles, the algorithm learns that the features associated with male applicants predict success — because that is what the data shows. The bias is not in the code. It is in the data.

• Keep a human in the loop. No AI tool should make an employment decision without human review. This is not just good practice — in many jurisdictions there are legal requirements around automated decision-making in employment, and organisations need to understand what applies to them.
• Audit your algorithms. If you are using AI screening, ask your vendor what the tool was trained on, how bias is tested for, and what the outcomes look like across different demographic groups. Demand that transparency — it is reasonable.
• Ensure diverse training data and diverse oversight. The people configuring, auditing and acting on AI hiring tools matter. Homogeneous teams asking homogeneous questions of their data will miss the same things they always missed.
• Do not over-index on AI-generated signals. A tool that scores candidate video for "confidence" or "communication style" is encoding a cultural and neurotype preference. Enthusiasm, directness, eye contact — these are not universal markers of competence.

The AI CV question: should candidates be penalised?

Candidates are increasingly using AI tools to polish their CVs, prepare for interviews and structure their application answers. Some hiring managers are uneasy about this. My view is straightforward: we did not criticise people for using calculators, autocorrect or predictive text. We should not penalise candidates for using AI either.

Candidates have always sought help — from careers advisers, professional CV writers, mentors and friends who happen to be good writers. AI is a new form of that support, and one that is far more accessible than expensive one-to-one coaching. Penalising its use would simply disadvantage people who cannot afford human alternatives.

More importantly, the future skill you want in your organisation is someone who can think critically and use AI to augment — not replace — their judgement. A candidate who uses AI sensibly to present their experience clearly is demonstrating exactly the capability you need. The interview and the work-sample stage exist precisely to test the thinking behind the application.

What is worth thinking about is whether your shortlisting criteria are picking up on polish rather than substance. If the most AI-assisted CVs are always scoring highest, that might be a signal that your criteria need revisiting — not that candidates need restricting.

The deeper pattern: recruiting in your own likeness

Most hiring bias is not about active prejudice. It is about comfort. We gravitate towards people who remind us of ourselves, who use the language we use, who went to places we recognise, who present confidence in the way we were taught to present it. This is what social scientists call homophily — recruiting in your own likeness — and it is one of the most powerful forces keeping organisations homogeneous.

The antidote is not to become suspicious of everyone who feels like a "good fit." It is to make fit mean something precise: fit for the role's requirements, fit for the team's gaps, fit for the organisation's future — not fit in the sense of "feels familiar."

See also the companion guide Inclusive Recruitment and the speaking topic Inclusive Recruitment for the broader strategic picture, or explore more on the Inclusion Bites podcast.