How to Tell Whether Your Interview Was Any Good
Mercedes Gordo runs delivery and customer success at Nova and spent two years at LinkedIn in Dublin. Her point about debriefs: the candidate gets evaluated by five people and the interview gets evaluated by nobody.
Updated 7 min read
In this article
Five people interview a candidate and four of them write down what they thought of her. Nobody writes down what they thought of the interview.
Which is odd, because the interview is the instrument. If it’s badly calibrated, every reading it produces is wrong, and no amount of careful debrief fixes a conversation that never got to anything.
Reading your own interview back is uncomfortable and it’s also the fastest way to get better at the part of the job that compounds. Here’s how to do it without it turning into either self-flagellation or a pat on the back.
The rule that makes it honest: no claim without a quote
One rule holds the whole thing up. Every observation has to carry the line from the transcript that supports it. If you can’t quote it, you drop it.
That sounds like bureaucracy. It isn’t. It’s the guardrail that prevents the two failure modes at once.
Without it you invent weaknesses, because reviewing anything puts you in a frame of mind where you find problems to justify the exercise. And without it you over-accuse, because “that question was leading” is easy to assert about a conversation you half remember and hard to assert about a line you have to paste.
So “I didn’t explore her international experience” becomes a verifiable claim rather than a feeling, and “that was a bit of a leading question” either has a line under it or it goes away.
Inventory first, judgment second
Before deciding what was missing, list what you actually asked, with the lines.
This is boring and it’s the step that makes the rest work. Memory is generous about things you meant to do, so the gap you’re about to flag may be a question you did ask in slightly different words at minute forty. The inventory settles it in ten seconds.
It also makes every absence claim checkable. If “you didn’t ask about X” can’t be verified against the list of what you did ask, it doesn’t go in.
And if there’s barely anything to inventory, if the candidate talked for fifty minutes and you spoke four times, that’s the finding. There’s nothing else to review.
Seven things to look at, two or three to work on
The dimensions worth examining:
- Question quality. Open and behavioural, or closed and leading? Did your questions invite evidence, or invite the candidate to tell you what you wanted to hear?
- Depth of probing. Did you ask for the specific project, the client, the result? Did you separate “I did this” from “I was on the team that did this”?
- Coverage against the role. Which must-haves got explored, which never came up.
- Signal drawn out. Did you run the conversation so that there was enough evidence to decide anything, or did it stay on the surface? This is about your conduct, not about whether the candidate turned out to be good.
- How you assessed fit. Did you probe the ways of working the brief actually asked for, through observable behaviour and examples?
- Missed follow-ups. The moments where what the candidate said called for one more question and didn’t get it.
- Phrasing that could introduce bias.
Don’t work on all seven. Pick the two or three with the most to gain. A review that produces fourteen improvement points produces zero changed behaviours, and you’ll know which two matter after reading the inventory.
The format that actually changes anything
For each thing you look at, three parts:
- What worked, with the quote. This isn’t padding. Knowing which of your questions produced the best answer in the interview is more useful than knowing which one was weakest.
- What to improve, with the quote, and this part is optional. If the interview was good, say it was good. Manufacturing a flaw to fill the section is how these reviews lose credibility.
- The better version. Quote the original question first, then write the rewrite or the follow-up underneath, ready to reuse.
That last part is where the value sits. “Your questions were a bit closed” is a judgment. Seeing your actual question with a better version written below it is a thing you can take into tomorrow’s interview.
Finish with two or three quick wins. Not a list per dimension, a synthesis: the handful of changes worth making next time.
What isn’t a flaw
This matters, because a badly done review of an interview mostly criticises someone for not being someone else.
Pace, warmth, humour, accent, small talk, silences, introversion and formality are not flaws. They’re how a person conducts a conversation. They only count against an interview when a specific line shows they cost you signal: so much small talk that half the must-haves never came up, say. That’s a cited consequence, not a style note.
And a moment is not a pattern. “You tend to interrupt” is a character assessment built from a sample of one. “At this point the candidate was mid-example and the next question came in” is something a person can do something about.
The same restraint applies to bias, with a higher bar still. The useful way to raise it is behaviour plus risk, anchored to the line: asked this way, at this point, this could disadvantage someone, and here’s a better phrasing. Never as a label on the person. And when the case is ambiguous, leave it out. An unfounded accusation costs far more than an unflagged borderline question, and a review that cries wolf gets ignored entirely.
The one finding people get backwards
If the only reason an interviewer liked a candidate is that they liked them, that’s the finding.
A culture-fit read with no behavioural anchor is affinity bias arriving in a structured process wearing a lanyard. It’s the thing to flag, not the thing to reward, and it shows up as a clean, confident, entirely unevidenced positive. Easy to miss precisely because it sounds decisive.
The fix isn’t to stop assessing fit. It’s to assess it through what people do: how they handle an ambiguous brief, whether they give and ask for direct feedback, how they work across functions. Those you can probe, quote and compare, which is the same move the interview scorecard makes on the evaluation side.
Keep it as coaching
One last thing, and it’s what determines whether this is worth doing at all.
The output goes to the person who ran the interview. It’s coaching, not a record, and the moment it becomes something filed about somebody it stops producing honest transcripts and starts producing defensive ones.
So: no grade, no ranking, no assessment of the interviewer as a professional. Specific moments, quotes, better versions, two or three things to try.
For teams where several people interview for the same role, reading back one transcript per interviewer per quarter is the cheapest calibration exercise there is, and it’s the only one that shows you what’s happening in the room rather than what people report afterwards. If you interview alone, do it on the one that went worst and on the one that went best, because the second teaches more than you’d expect.
If you’d rather not do the reading by hand, this method is one of eleven free AI skills our customer success team packaged up. You give it the transcript, it inventories what you asked, picks the two or three areas with the most to gain, and returns each one with the quote and a rewritten version of the question. The Interviews pack has three more alongside it.
And the questions for next time are a separate exercise with a separate structure, which we wrote up in the screening questions guide and in the interview guide.
Frequently asked questions
How do you review your own interview technique?
Work from the transcript, not from memory. First list what you actually asked, quoting the lines. Then pick the two or three areas with the most to gain and look at them against that list: whether your questions were open or leading, whether you probed for specifics, whether you covered the role's must-haves, and the moments where the candidate said something that called for a follow-up you didn't make. Every observation needs the line that supports it, or it doesn't count.
What should you look for when reviewing an interview?
Seven things, of which you work on two or three at a time: question quality, how deep you probed, coverage against the job description, whether you drew out enough evidence to decide anything, whether you assessed fit through observable behaviour, the follow-ups you missed, and any phrasing that could introduce bias. Trying to improve all seven at once produces a list nobody acts on.
Can you review an interview from notes instead of a transcript?
Partly, and it's worth knowing what you lose. Without the literal questions you can't judge phrasing, you can't assess how you probed, and you can't say anything about bias, because all three live in the exact words. What survives is coverage, which must-haves got touched and which didn't, and the obvious gaps. That's still useful, but don't let it masquerade as the full review.
Is it fair to give an interviewer feedback on their technique?
It is when it's coaching rather than a record. The distinction is practical: the output goes to the person who ran the interview, to help them, not into a file about them. It never produces a grade or a ranking, it talks about specific moments rather than the person, and it treats one interview as one interview rather than as evidence of a pattern.
What counts as a biased question in an interview?
Phrasing that plausibly and materially disadvantages people in a protected group, which is a higher bar than phrasing you'd write differently today. The useful way to raise it is behaviour plus risk, anchored to the line: at this point, asked this way, this could disadvantage X, and here's a better version. Never as a label on the person. If the case is ambiguous, say nothing: an unfounded accusation costs more than an unflagged borderline question.
Sources and method
The method described here is the one behind Nova's Interviewer Feedback skill, built and published by the customer success team. It's free to download as part of the Interviews pack.