Learning how to conduct an interview well is mostly learning to stop doing the thing that comes naturally. The natural version is a conversation: you build rapport, you follow what is interesting, you form an impression, and you leave feeling you have a good read on the person. That version is close to unreliable as a predictor of job performance, and decades of research have been consistent about it.
The problem is not that interviewers are bad at reading people. It is that in a free-form conversation, every candidate gets a different interview, so there is nothing to compare. You end up comparing your conversations rather than their capabilities.
Decide what you are assessing, and stop there
Four or five criteria from the scorecard. That is the whole scope of the interview.
If a criterion is not on the list, it does not enter the decision, however much you noticed it. This is harder than it sounds, because the things you notice are vivid and the criteria are abstract. But an interviewer who scores on unlisted impressions is not assessing, they are rationalising.
Split the criteria across the panel so each interviewer owns one or two and goes deep, rather than four people skimming the same ground. It also spares the candidate answering the same question four times, which is the most common complaint candidates have about panel processes.
Ask about what they have actually done
The most reliable question format asks for a specific past situation and then digs into it.
"Tell me about a time you had to deliver something with incomplete information. What was the situation, what did you do, and what happened." Then follow up until you understand it: what was your part specifically, what did you consider and reject, what would you do differently.
The follow-ups are where the value is. Anyone can prepare a polished opening story. Very few people can sustain three layers of detail on something they did not actually do. You are not trying to catch anyone out. You are trying to get past the summary to the actual decisions.
Avoid hypotheticals. "What would you do if a client escalated" tests whether someone can describe good practice. "Tell me about a time a client escalated" tests whether they have done it. These produce different information and only one of them is evidence.
Behavioural interview questions covers the question design in more depth.
Ask everyone the same things
Same core questions, same order, for every candidate on the same role.
This is the part interviewers resist most, because it feels rigid and it feels like it wastes the rapport. It is also the only thing that makes the answers comparable. Follow-ups can and should vary, since they depend on the answer. The core questions should not.
A useful test: if you cannot put two candidates' answers to the same question side by side, you did not run an interview you can decide from.
Talk less than you think
The common failure is an interviewer who talks for half the session. It happens through enthusiasm, through nerves, or through selling the role too early, and it produces an interview where the candidate said very little and the interviewer left with a warm impression of their own explanation.
Aim for the candidate speaking around eighty percent of the time. Sell the role properly at the end, once you know whether you want to.
Silence is a tool. When someone finishes an answer that felt incomplete, wait. Most people fill the gap, and what they add is usually more candid than what they prepared.
Take notes you can actually use
Write what they said, not what you concluded. "Led the migration, three engineers, slipped two weeks on a vendor dependency, escalated at day four" is usable evidence. "Good technical leadership" is your conclusion with the evidence deleted, and it cannot be compared with anything.
Score straight after the interview, before you speak to the other interviewers. Discussing first is how panels converge on the loudest voice rather than on the evidence. Score independently, then compare, and treat the disagreements as the interesting part. When two people rate the same answer very differently, one of them heard something the other missed.
The habits that make interviews unreliable
Deciding in the first few minutes. First impressions form fast and then quietly recruit the rest of the interview as supporting evidence. You cannot switch this off, but knowing it happens and scoring against written criteria limits the damage.
Hiring the person you liked. Rapport is a real signal for some roles and a contaminant in most. The question is whether they can do the job, not whether the conversation was enjoyable.
Letting one strong answer carry everything. The halo effect. One impressive story raises every other score. Scoring each criterion separately, before you form an overall view, keeps them independent.
Stress techniques. Deliberately hostile or trick questions tell you how someone handles a contrived situation invented by a stranger. They do not predict performance, they damage your employer brand, and the best candidates simply leave the process.
It runs in both directions
The candidate is deciding too, and in a tight market for the role you are filling, they usually have options.
Leave real time for their questions and answer honestly, including about the difficult parts of the role. Candidates who join on an accurate picture stay. Candidates who join on a sales pitch work out the difference in month two.
Be straight about what happens next and then do it. You do not need to promise a timeframe. Saying what the next step is and then actually taking it is enough, and it is rarer than it should be.
Common questions
How long should a job interview be?
Forty-five minutes to an hour for most roles. Longer than ninety minutes and quality drops for everyone. If you need more coverage, add a stage rather than extending the session.
How many interview stages should there be?
Two or three for most roles. Every additional stage loses candidates, and beyond three the marginal information is usually small. If a fourth stage exists because nobody will decide, the problem is the decision rule, not the information.
Are structured interviews really better?
Yes, and the research on this is unusually consistent. Structured interviews predict job performance substantially better than unstructured ones, mainly because they make candidates comparable.
Should you send questions in advance?
For some roles, yes. It reduces the advantage held by candidates who are practised at interviewing rather than better at the job, and it tends to improve answer quality. You still get to probe with follow-ups.
Assess four or five written criteria and nothing else, ask every candidate the same core questions about things they have actually done, follow up until you understand what they personally did, let them talk for most of the session, and score independently before the panel discusses anything.