Interview Bias: The Types, Where It Creeps In, and How to Reduce It

What does your company do about bias in hiring? If the answer is self-audit, you may want to reconsider. In a 2024 study, people rated their own susceptibility to eight biases at about 4.6 on a nine-point scale, and other people's at about 6.7.

Bias is hard to see from the inside. Nobody means to judge a candidate on anything but the job, and yet interviewers are influenced by the first minute of the conversation, where a candidate sits in the day's schedule, and how much they still remember two days after the interview.

This article covers what interview bias is, the types to watch for, and how to reduce it in your hiring process.

What is interview bias?

Interview bias is any preference or mental shortcut that shapes how an interviewer judges a candidate on factors unrelated to their ability to do the job. Most of these judgments happen quickly and feel like sound instinct to the person making them. The result is that two candidates who are equally suited to the role can leave the same interview with very different outcomes.

Imagine you're interviewing two equally suited candidates for one role. The first is easy to talk to from the first minute, and the interview feels like a conversation. The second never warms up. Afterward, you're sure about the first candidate and not about the second. Was it their answers? Or was it the first minute?

Interview bias leaves companies guessing at which candidate will do the job well. Schmidt and Hunter's review of 85 years of hiring research found that a rating built on the interviewer's impressions predicts job performance poorly. Impressions reward whatever the interviewer happens to like, from a shared alma mater to an easy first minute. A company that hires this way is more likely to end up with a team of lookalikes. One consequence is reduced diversity, which can hurt the company's bottom line. McKinsey's Diversity Wins report found that companies in the top quartile for executive gender diversity were 25% more likely than bottom-quartile companies to have above-average profitability.

The types of interview bias

Below, we'll cover eight biases that feed into that rating, each with an example of what it looks like in an interview.

First impression bias

First impression bias forms the interviewer's opinion before the first question and carries it, untested, through the whole interview.

Say a candidate arrives ten minutes late, still apologizing about the train as they sit down. The interviewer has decided they're disorganized before the first question. Over the next forty minutes the candidate describes taking over a project that was a total mess, single-handedly getting it under control, and keeping it on schedule until release two years later. The interviewer still sees a disorganized candidate. The notes say "seemed scattered" and make no mention of the project.

We've written a full guide to first impression bias if you want to go deeper.

Halo and horns effect

The halo effect lets one positive trait, such as a well-known former employer or a fluent first answer, lift every other rating. Once the interviewer has decided the candidate is strong, they make justifications for weak answers.

Say a candidate spent four years at a well-known company. They give a thin answer to the core skill question, and the interviewer's notes read "probably just nervous, clearly capable." A candidate from an unknown company gives the same thin answer, and the notes read "gap in core skill."

The horns effect is the reverse. Once one thing has counted against the candidate, the interviewer stops collecting evidence the other way, and the rest of the interview turns into a search for the second flaw.

Say a candidate's cover letter had a typo in the first line. The interviewer notices it while reviewing the application and carries it into the room. In the interview the candidate answers the technical question well, and the interviewer scores it a point lower than the same answer from a candidate whose cover letter was clean. When a colleague asks about the gap in the scores, the interviewer says the answer felt careless.

Score each answer against the criteria before you rate the candidate as a whole. A thin answer on the core skill then earns the same score from every candidate, whichever company they came from, and a strong technical answer keeps its full score despite a typo in the cover letter.

Affinity bias

Affinity bias is a preference for candidates who are like us, whether in background, school, hobbies, or manner.

Picture two interviewers who played college soccer. Of the two candidates they see, one also played college soccer. That interview runs long and turns to soccer. Afterward, the notes call that candidate "great energy" and the other "a bit flat," though their answers to the interview questions were equally strong.

After every interview, reread your notes and make sure each positive judgment has a specific reason behind it. Check hardest after the interviews where you felt an easy rapport with the candidate, because that is when a note like "great energy" is most likely to stand in for evidence.

Confirmation bias

Confirmation bias is the tendency to notice and weigh evidence that fits what we already believe. In a job interview, the interviewer asks the questions that confirm an early impression and skips the ones that would test it.

A candidate's resume lists three employers in the past four years, none held longer than eighteen months. Before the interview starts, the interviewer has concluded that this candidate will leave within a year. The first twenty minutes go to why each job ended, and no question touches what the candidate accomplished in those jobs. The two promotions the candidate earned in that span go unmentioned. The notes read "flight risk." A colleague reads the notes, skips the resume, and agrees.

When the resume raises a concern, ask about what the candidate achieved in each role as well as about the concern itself. The candidate with three employers in four years then gets a question about each job's results, and the answer turns up the two promotions.

Contrast effect

With the contrast effect, the last candidate you saw sets the bar for the next one. An average candidate who interviews straight after the weakest candidate of the week looks solid. The same person, interviewed after the strongest, looks ordinary.

Suppose five people interview for one role over two days. On Monday a very strong candidate is interviewed. On Tuesday the first candidate hasn't read the job description and cannot name a project they led. The next candidate gives adequate answers, which look impressive after the interview that preceded them. At the debrief, the interviewer recommends the second Tuesday candidate and passes over the stronger one from Monday.

Score every candidate against criteria you wrote before the round began. Monday's standout and Tuesday's adequate candidate then face the same bar, and the stronger one gets the recommendation.

Recency and primacy bias

Primacy bias gives extra weight to what came first, and recency bias to what came last. In an interview round, the candidates at either end of the schedule get that weight. In a single interview, the opening and closing answers do.

Suppose five people interview on a Tuesday and the debrief is on Thursday. The interviewers discuss two of the five by name, the first and the last. The middle three have blurred together by then, and each interviewer scores them from memory. The candidate who gave the best answer on the core skill question was third.

Have each interviewer score every candidate on the day of the interview. However little of Tuesday the interviewers remember by Thursday, the third candidate's best answer is already on paper.

Stereotype bias

Stereotype bias treats an assumption about a group as a fact about the candidate. The group might be an age bracket, a gender, an ethnicity, an accent, a school, or people with a gap in their career history.

Take an interviewer who sees a 1990 graduation date on a resume. Before the candidate has answered a question, the interviewer expects them to struggle with the company's software. Nothing in the interview touches on software, and the candidate spent the last two years moving their team onto a new platform. The interviewer still gives them a low rating for learning new tools.

Trace every rating back to an answer in the notes. A low rating on learning new tools with no answer behind it came from the graduation date.

Groupthink

Groupthink is the pull toward agreement in a group. In a hiring debrief, the first confident opinion becomes the group's opinion, and the other interviewers adjust what they report to match it.

Say three people interview a candidate, and the hiring manager opens the debrief with "I think we've found our person." The second interviewer had rated the candidate average and now says "I could see that." The third had a concern about the technical answer, decides it was minor, and says nothing. The candidate is hired. Three months in, the technical gap is the first thing on their performance review.

Collect every interviewer's score in writing before the debrief starts. The third interviewer's concern about the technical answer then reaches the debrief in writing, before the hiring manager has said a word.

Where bias creeps into the interview process

Bias gets three chances at a hiring process. It can get in before anyone has met the candidate, in the room, and in the debrief afterward.

Before the interview

The interviewer usually reads the resume before the candidate walks in. A well-known former employer, a typo in the cover letter, three jobs in four years, or a graduation year can shape the interviewer's opinion before the candidate has said a word.

Write down what the interview will assess before anyone reads the resume. If something on the resume raises a concern, write it down as a question to ask the candidate.

In the room

When interviewers are free to ask whatever comes to mind, the interview changes in two ways.

• The questions drift. Different candidates get different questions. The one the interviewer likes gets the easy follow-ups, and the one they doubt gets the hard probe.
• Rapport changes the difficulty. Even the questions the interviewer planned to ask change. Twenty minutes into a conversation that's going well, the interviewer softens or skips the question about why the last job ended.

Ask every candidate the same questions in the same order, and score each answer separately.

After the interviews

The last stage is the debrief, held from memory. Two of the biases above work on it. Recency and primacy bias decide which candidates the interviewers discuss by name, and groupthink decides what the others report once the first opinion is out. If every interviewer scores on the day and sends the scores in before the debrief, the meeting has one job left. Go through the candidates whose scores differ and work out why.

In a study of about 29,000 admission interviews for a German study grant program and 8,000 hiring interviews at a consulting firm, candidates who followed a candidate with a yes vote were about 15% and 40% less likely to get one themselves. The economists who ran it, Jonas Radbruch and Amelie Schiprowski, propose independent assessments from several interviewers as a protection. There's a walkthrough of building the scorecard in our interview scoring matrix guide.

How to reduce interview bias in a small business

At a small business, anyone might be doing the hiring. A larger business has a written process, an approval at each step, and a recruiting department that decides who does. On a hiring team of two or three, the process is whatever the team does each time, and the fixes below are habits.

• Write the scorecard before the first interview. Our behavioral interview scoring matrix template is a ready-made one.
• Put the questions in writing and ask them in a fixed order. Our problem-solving interview questions are a place to start.
• Hold the discussion after every interviewer's scores are written down. With a small team, a message with a number in it before the debrief starts will do.
• Book candidates on different days. In the Radbruch and Schiprowski study, the previous candidate's pull halved with an hour or more between interviews and was nearly gone by the next day.
• Get a second opinion. In Polymer, our hiring platform, you can email a colleague a candidate's review kit, and they can review the candidate from it without being added to the hiring team.

How to reduce interview bias on a larger hiring team

On a larger hiring team, someone in the recruiting department owns the process. They set it up once, and it runs the same way for every role.

Start with the structured interview. Schmidt and Hunter's review also found that adding a structured interview to a test of general mental ability improved the prediction of job performance by 24%, and adding an unstructured one improved it by 8%. Then set up the rest of the process around it.

• Build diverse interview panels. A candidate who shares a background with one interviewer shouldn't share it with all of them. With three interviewers from three backgrounds, one interviewer's affinity bias is one score out of three.
• Assign each interviewer a focus area. One interviewer covers the skills the job needs, and another covers teamwork. Each interviewer scores that one area. Nobody gives an overall impression.
• Run calibration sessions. Before a hiring round, the interviewers score the same set of sample answers and compare. The point is to agree on what earns a 3 before any real candidate gets one.
• Don't schedule similar candidates back to back. In the Radbruch and Schiprowski study, the previous candidate weighed more when the two candidates were alike, in gender or field of study. If two of the candidates studied the same thing, put a day between them.
• Debrief from the written scores. The meeting doesn't start until every score is in, and the first item on the agenda is any candidate the panel split on.
• Use training to teach the process. Interviewers should know what the scorecard is for and how the debrief runs. That's a different job from training interviewers out of their biases (more on this in the next section).

What doesn't work

Three fixes have the research against them, and knowing which three saves you the effort of trying them.

Training on its own. The fix here is a diversity training course, a session meant to leave the people who take it less biased. According to a Harvard Business Review article by Frank Dobbin and Alexandra Kalev, nearly half of midsize companies run one, and so do nearly all the Fortune 500. The same article says its positive effects rarely last beyond a day or two, and that a number of studies suggest it can activate bias or spark a backlash.

In the authors' data from more than 800 U.S. firms, the share of Black women in management fell by 9% on average in the five years after a firm required its managers to take the course, and every other group either held steady or lost ground. Companies got better results when they engaged managers in solving the problem, put them in more on-the-job contact with women and minority workers, and held them accountable. As the article puts it, "When people know they might have to explain their decisions, they are less likely to act on bias."

Awareness alone. The hope behind awareness is that if a person's bias can be lowered, their decisions will change with it. A review of 492 studies looked at ways of changing how people score on a bias test. The scores moved, but not by much. In the studies that also measured behavior, the review found no evidence that a change in the score led to a change in what people did. The Radbruch and Schiprowski study above tested awareness directly, and telling the evaluators about the effect didn't shrink it.

Auditing yourself. Let's go back to the bias blind spot from the introduction. In the 2024 study, people put their own bias at about 4.6 and other people's at about 6.7. One explanation the researchers give is the introspection illusion. We judge ourselves by looking inward and trust what we find there too much, and we judge everyone else by what they do. A self-audit relies on the same inward look, so it runs into the same problem.

Does AI make interview bias better or worse?

For a growing number of candidates, the first interviewer is a machine. Few studies have tested AI interviewers so far. There is a growing body of research on AI resume screening, and much of it can be applied to the interview. The research so far has found the same biases in AI screening as in human interviewers, and the same fixes work on both. The rest of this section lays out that evidence and what it means for anyone building an AI interviewer.

The weak criteria effect

Two AI models can give the same resume very different scores. Researchers at the Indian Institute of Technology Jodhpur gave the same resumes, the same job description, and the same instructions to Claude, GPT, and Gemini. One resume averaged 50 out of 100 from GPT and 73 from Gemini. Adding details about the hiring company moved GPT's score for that resume to 76, and cutting the job description to a minimum spread the three models furthest apart. With a detailed job description, the three models agreed closely enough to be useful. The score was only as good as the criteria behind it, which is what the scorecard does for a human interviewer.

The stereotype effect

A detailed job description does not remove every error. The ones left follow a pattern, and it is the pattern a biased human interviewer follows. In a second study, researchers gave a screening tool that ranks resumes by how closely they match a job description a set of resumes that differed only in the name at the top, and tested it across nine occupations. In 85% of the 27 tests the screening favored white-sounding names, and it favored women's names in only 11%. The tool had taken an assumption about a group and applied it as a fact about the candidate, which is stereotype bias as defined above.

The order effect

The models favored the resume they saw first. A third study gave 22 language models a job description and two matching resumes and asked which candidate suited the job better. Each pair was then run again with the resumes in the other order. Twenty-one of the 22 models favored whichever resume was listed first, choosing it 63.5% of the time. When the models scored each resume alone, with no pair to compare, the preference disappeared. This is the order effect, the contrast effect from above in a machine, and scoring one candidate at a time removed it as it does for people.

Four fixes for AI screening

Each of the three studies found one change that made the scoring more accurate and more consistent. Whoever sets up the AI screening controls all of them.

• Give the model the full job description. Scores turned to noise when the description was thin.
• Remove the name and other identifying details. Removing them takes away one signal the model was using.
• Score one candidate at a time. With no pair, there is no first-listed candidate to favor.
• Require the evidence behind every score, so a person can check it.

The Plato example

Plato, Polymer's AI candidate review, is one way to build to those four rules.

• It draws the criteria from the job description once, in a separate step, so every candidate is measured against the same list. That is the fix for the weak criteria effect.
• Its first step for each candidate removes the name, location, and contact details, and the review that follows never sees them. That is the fix for the stereotype effect.
• It reviews one candidate at a time, so each candidate is the only one in view, with nothing listed first or second. That is the fix for the order effect.
• It searches each application for evidence on every criterion and records what it finds or marks the criterion as missing, so a person on the hiring team can check the reasoning. That is the fix for confirmation bias, since every criterion gets looked for, including the ones that would test the first impression.

Why AI interviewers are harder to build

So, does AI make interview bias better or worse? The studies above show it can carry every weakness a human interviewer has. The difference is what can be done about it. A human interviewer cannot be forced to use the scorecard. An AI reviewer can be built so the scorecard is the only thing it sees, and it follows the structure more reliably than a person does, though not perfectly. Built that way, AI can make interview bias better. Built without it, AI makes the same mistakes faster and at scale.

An AI interviewer with a set number of questions is AI screening with a microphone. It asks the same list in the same order to every candidate and scores each answer against criteria set up beforehand, and the fixes that correct the issues with AI screening should correct the issues with this kind of AI interview.

The adaptive interviewer is harder to design without bias, and it is the kind most vendors now sell. It listens to the candidate's response and alters its next question based on what it heard. An AI reviewer avoids bias only when it scores against criteria written before the interview begins. The model writes its follow-up during the interview. The company could not have written criteria beforehand for a question it never knew would be asked.

The only way around it is to write the follow-ups beforehand too. That means a whole set of branching interview questions must be written for the AI interviewer. For every question it asks, it must have a set of follow-up questions it may select from, and each of those must have its criteria written beforehand. Two candidates who take different paths through that set still have to be scored on the same competencies, or their scores can't be compared. The set gets large fast, and building it is the hard part of an adaptive interviewer. Before buying one, ask the vendor how they built theirs.

Reduce interview bias by changing the process

The eight biases above are ordinary, and every interviewer has them. The process around the interview is what you can change. Write the criteria down before you read the first resume, ask every candidate the same questions, get the scores in before the debrief starts, and leave a day between candidates.

These steps base the hiring decision on each candidate's answers, scored against the same written criteria.

If you're ready to run your hiring this way, give Polymer a try. Sign up for free.

Frequently asked questions