Confirmation bias in interviews: how first impressions steer the questions, and the fixes

Confirmation bias in interviews is the tendency to look for, ask about and remember evidence that fits what the interviewer already believes about a candidate, whether that belief came from the resume, a referral or an earlier interviewer's note. In a field study of 79 audiotaped interviews, Dougherty, Turban and Callender (1994) found that interviewers with a positive first impression showed more positive regard, spent more time selling the company and gathered less information. The structural fixes are the same questions and planned probes for every candidate, evidence written down before scores, and interviewers who do not see earlier scores until they have submitted their own.

This page covers the research, where the first impression comes from, what confirmation bias looks like in transcripts and notes, and the fixes. Rating errors that spread one impression across scores are covered in the halo effect in interviews.

What is confirmation bias?

In his review in the Review of General Psychology, Raymond Nickerson (1998) describes confirmation bias as seeking or interpreting evidence in ways that are partial to existing beliefs, expectations or a hypothesis in hand. In an interview it works through three channels:

• Search: which questions get asked, and which follow-ups.
• Interpretation: whether an ambiguous answer is read as modesty or as a lack of substance.
• Memory and notes: which parts of a 45-minute conversation make it onto the scorecard.

Ratings are the last step. By the time the interviewer scores, the evidence has already been selected.

What does the research show?

One caution in reading these. A first impression from a good resume is often partly right, so a link between pre-interview and post-interview views is not proof of bias on its own. The problem the studies show is behavioral: candidates with different first impressions got different interviews, so the evidence collected was not comparable. Snyder and Swann's result is the uncomfortable one, because it means the candidate's behavior can shift in response to the questions, making the confirming evidence real.

Where does the first impression come from?

• The resume: famous employers and schools, or a gap and a short tenure.
• The referral: who sent the candidate, and how strongly.
• The screen note: "excellent, must hire" or "a bit flat" from the recruiter.
• Earlier rounds: a colleague's score, or hallway talk between interviews.
• The first minutes: a late start, a handshake, a joke that lands. OPM's structured interview guide (September 2008) lists relying on first impressions as the first common interviewing mistake, and warns that looking at resumes during the interview may lead interviewers to rate candidates with strong resumes more favorably.

What does confirmation bias look like in a real interview?

An invented example: one hiring manager, two operations analyst candidates, questions taken from the transcripts.

CANDIDATE A  (screen note: "strongest we've seen", referred by a VP)
"Tell me about the dashboard project, that sounds great."
"And the team loved it, I assume?"
[Manager talks about the company for 9 minutes]
Notes: "Strong as expected. Clearly data-driven."

CANDIDATE B  (resume: 14 months at last job)
"Why did you leave after just over a year?"
"What went wrong on that project?"
"What would your old manager say you need to improve?"
[Manager talks about the company for 2 minutes]
Notes: "Tenure concern confirmed. Some gaps in depth."

Candidate A never got a question that could show a weakness; candidate B got three that could show nothing else. The notes then confirm the starting points. Both scorecards may look reasonable on their own. Side by side, they are not measuring the same thing.

What are the signs in transcripts and notes?

• Question asymmetry. Count questions that seek negative information ("what went wrong", "why did you leave") per candidate. Large differences between candidates for the same role are a signal.
• Leading follow-ups. "I assume that went well?" invites agreement, not evidence.
• Unprobed answers. Vague answers from the favored candidate pass without "what did you personally do?"
• Talk and selling time. More interviewer talk and more selling with some candidates, the pattern Dougherty and colleagues observed.
• Uneven explanations. A weak answer is "nerves" for one candidate and "lack of depth" for another, the attribution pattern in Phillips and Dipboye.
• Words that give it away. "As expected", "confirmed", "just as [recruiter] said".
• Remembered quotes. Evidence on the scorecard that the candidate did not quite say. Checking notes against a transcript catches this. Interview Signal, which we make, drops any quote it cannot find in the transcript before it reaches the scorecard; with handwritten notes, reread the verbatim lines before scoring.
• Scores that match the previous round exactly across every competency.

Is a first impression ever useful?

Yes, as a hypothesis to test rather than a conclusion to confirm. A resume that shows three short tenures is a fair reason to ask about tenure; a famous employer is a fair reason to ask what the candidate personally did there. The difference is whether the question can come back either way. A useful habit is to plan one question per candidate that could prove your first impression wrong:

Which structural fixes reduce confirmation bias?

1. The same questions and planned probes for everyone, including the hard ones. If "what went wrong?" is in the kit, every candidate gets it. See what is a structured interview.
2. Read before, not during. Use the resume to plan where to probe, then close it.
3. Hand over questions, not verdicts. Between rounds, pass on what is still unknown rather than an overall rating. Interview handoff notes between rounds has a format.
4. Hide earlier scores until you submit. Interviewers should not see other scorecards, or hear hallway verdicts, before their own is in. See independent interview feedback before the debrief.
5. Evidence first, then the score. Write what the candidate said under each competency, then rate against the anchor. Evidence-based interview feedback explains the method.
6. Consider the opposite before scoring. Ask yourself one question: "What did this candidate say that cuts against my first impression?" Lord, Lepper and Preston found this kind of instruction worked better than simply resolving to be fair. Write the answer down.
7. Sell last, and the same for everyone. Keep the pitch for the final minutes, with similar time for each candidate, so selling does not crowd out questions for the candidates you already like.
8. Audit a few transcripts. Once a quarter, compare two interviews for the same role using the signs above. The patterns are easier to see side by side than in the moment.

None of these steps asks interviewers to stop forming impressions, which nobody can do. They make sure the impression does not decide which evidence gets collected. For where confirmation bias sits among the other rating errors, see how to reduce bias in interviews.