Tell one real disagreement about a work decision, not a personality clash, and spend most of the ninety seconds on what you did to understand the other person's case. The interviewer is not scoring who was right. They are scoring whether a team can disagree with you and still function afterwards, which is why the line that decides the answer is almost always the one about where the relationship ended up, not the one about who won.
The five things being scored
Every one of these is checkable from the transcript, which is why weak answers fail in predictable ways rather than randomly.
| Signal | Lands when it sounds like | Fails when it sounds like |
|---|---|---|
| The conflict has an object | "We disagreed about whether to put a cache in front of a model with a known consistency bug." | "He was very difficult to work with." |
| You can state their case fairly | "His position was that a wrong number in a customer dashboard costs more than a slow one, which is fair." | "He just did not want to do the work." |
| You moved first | "I stopped replying in the thread and asked for twenty minutes with a whiteboard." | "I took it to my manager." |
| The resolution has a mechanism | "We wrote a rule into the review template: block twice and we book fifteen minutes." | "We found a middle ground in the end." |
| The relationship carried on | "He is the person I now ask to review anything I am unsure about." | Nothing. The answer stops at the outcome. |
Before you write anything, pick the story against four filters. The disagreement has to be about a decision with a consequence you can name. The other person has to be a peer or someone more senior, because a conflict with your own report is a management story and gets scored as one. It has to have been resolved by something you did, not by a reorganisation, someone leaving, or a manager overruling. And you must be able to describe their position in a sentence they would sign. If you cannot do that last one, choose a different story, because the answer will leak contempt however carefully you word it.
Four answers to the same question, scored
The first and the fourth are the same underlying event told two ways, so you can see that the difference is framing rather than luck. Annotations in grey are what the interviewer is thinking, not part of the answer.
1. The blame answer
Yeah, I had a staff engineer who sat on my pull request for about a week. [Opens with a verdict on a person instead of a description of a disagreement. The first note taken is about your tone.] Honestly he just did not like that the idea had not come from him. [A motive you cannot evidence. This is now the thing the interviewer imagines you saying about their team.] I explained the design three or four times and he kept raising the same objections. [Repeating your case is not resolution work. There is still no action here aimed at understanding him.] In the end I went to my manager and he told him to approve it. [Escalation as the punchline. The conflict was not resolved, it was outsourced.] We shipped a week late because of him. [The result is a grievance with a named culprit attached.] We are fine now, we just do not really work on the same things any more. [The relationship was fixed by distance, which is the exact opposite of the signal.]
Fails all five. None of it is necessarily untrue, and that is not the problem. The problem is that a hiring manager has just watched you rehearse how you talk about a colleague who disagreed with you.
2. The no-conflict dodge
To be honest I have been quite lucky, I have not really had conflict at work. I get on with pretty much everyone. [A claim no interviewer believes about anyone with three years of experience.] If there is a disagreement I usually just talk it through and it gets sorted. [Hypothetical and in the present tense. There is no event, so there is nothing to score.] I am not really a confrontational person. [Recasts conflict as a character defect you happen to lack, which reads as avoidance rather than harmony.]
Scored as no data, and in a rated loop no data lands in the same band as a bad answer. Interviewers usually offer a second chance here, which is the worst outcome, because you then improvise and lose the structure. If you genuinely cannot find a row, look for a disagreement instead: a design review you lost, a deadline you pushed back on, a reviewer who blocked your work, feedback you did not accept. All of those count, and all of them can be told well.
3. The trivial-conflict answer
There was a disagreement about stand-up. Another engineer wanted it at 9:30 and I wanted 10 because of my commute. [Nothing is at stake, so even a flawless resolution proves nothing about how you behave when a decision has a cost.] We talked it through and agreed to alternate weeks. [The mechanism is genuinely fine. It resolves in one sentence because there was nothing to resolve.] It was fine after that, no bad feeling at all.
Passes on process, fails on stakes. This is the most common near miss, and it comes from picking a safe story on purpose. The safety is the failure. An interviewer cannot tell the difference between someone who handles pressure well and someone who has never been under any, so a conflict with no cost gives them nothing to write down. If your best example is a rota, say out loud that the stakes were low, then tell the second one.
4. The strong answer
Four months into a project, a staff engineer blocked my pull request twice, three days before a launch date sales had already given to customers. [Object, stakes and a clock in one sentence, with no adjective attached to him.] I thought he was holding us up over a hypothetical. He thought I was putting a cache in front of a data model with a known consistency bug, and put like that it was a fair thing to block. [The steel-man. His case in a sentence he would sign, plus an early concession. This one line carries the whole answer.] I stopped replying in the review thread and asked for twenty minutes with a whiteboard, because we had been arguing in comments for three days and the comments were hardening both of us. [A concrete first move, with the reason for it.] I asked him to describe the failure he was picturing, and I wrote it up as a test case rather than as an opinion. It reproduced. [You changed your mind on evidence, in front of him. Most answers never reach this moment.] So we shipped with a five minute TTL and read-through invalidation on that one table instead of the hour I wanted, which cost some of the benefit and none of the correctness. [A trade-off with a price on it, not a "middle ground".] Then we put two lines in the review template: if a reviewer blocks twice, stop commenting and book fifteen minutes. [A mechanism that outlives the incident, which is what separates a resolution from a truce.] We shipped on the date, and he is the person I now ask to review anything I am unsure about. [Relationship evidence given as behaviour rather than as a claim about feelings.]
About 200 words, roughly ninety seconds spoken. Notice what is missing: no adjectives about him, no account of who said what on which day, and no moment where you win. The only opinion in it is one you abandoned.
| Signal | Blame | Dodge | Trivial | Strong |
|---|---|---|---|---|
| Conflict has an object | No, a person | No event | Yes, but weightless | Yes, named and costed |
| Their case stated fairly | Motive invented | None | Not needed | Yes, and conceded |
| You moved first | Escalated | None | Yes | Yes, and changed the medium |
| Resolution mechanism | A manager's order | None | Alternating weeks | Written into the process |
| Relationship after | Avoidance | None | Untested | Evidenced by a later action |
The three follow-ups, and the close for each
These arrive in almost every loop, and they are where prepared answers come apart, because people rehearse the story and not the aftermath. Each close below runs about twenty seconds.
"How did the relationship end up?"
They are checking whether the resolution was real or whether the other person simply complied and withdrew. Two answers fail: "fine, no issues at all", and any version of "we do not really overlap now". The rule for your close is to cite an event that happened after the conflict, never a feeling.
He is one of two people I ask for reviews on anything I am unsure about. Six weeks later I sent him the design for the next caching change before I opened the pull request, and that was a direct result of the argument rather than politeness.
"What would you do differently?"
They are checking whether you can name your own contribution without collapsing and disowning the outcome. "I would communicate earlier" is too generic to score. "Nothing, it worked out" is a refusal. Over-apologising is worse than both, because it retracts the good result you just described. Name one process error, put a cost on it, then give the fix.
I would have moved it off the review thread on day one instead of day three. Those three days of comments pushed both of us into positions we then had to climb down from in public, and the whiteboard conversation that actually settled it took twenty minutes.
"Has anything like that happened again?"
They are checking whether you took away a story or a rule. There is a trap here: "no, never since" reads as luck or as no data, and it quietly undoes the mechanism you just described. Yes is the stronger answer, provided the second time was shorter.
Twice, and both were quicker. The last one was with a data engineer about backfill ordering. We hit the two-comment limit, booked fifteen minutes, and that time I was the one who was wrong. The rule is doing the work, which is the point of writing it into the template rather than just resolving to be more careful.
Three neighbouring prompts take the same story with one adjustment each. For "a time you disagreed with your manager", it only works if the other person outranked you, and you need one line confirming that you committed to the decision visibly once it was made. For "a time you had to work with someone difficult", do not accept the word difficult: describe the behaviour and its effect on the work instead. For "feedback you disagreed with", the resolution has to include what you actually did with the feedback, not only how you received it. Our interview questions bank covers the wider set, and the company Q&A generator builds a bank for a specific company and role, which is how you find out whether this question is even likely in that loop.
Where live help is the wrong tool for this one
Worth being straight about, given a vendor is writing this. The conflict question is the single worst one to read off a screen. It is the question where the interviewer listens hardest to tone, and a drafted paragraph read aloud sounds like someone being careful about a person, which is exactly the impression you are trying to avoid. OfferPilot AI's Interview Copilot can put the beats of your own story in front of you in about a second when the prompt lands and your mind goes blank, but the sentence that decides this answer is the one where you state the other person's case fairly, and no model knows which of your colleagues you are able to describe fairly. That judgement is yours, and it has to be made before the call.
Two further limits. It runs on Windows 10 and 11 only, so on a Mac this page is a preparation exercise and nothing else. And there is one category of story to leave out entirely: a conflict touching a protected characteristic, a live grievance or legal matter, or anything where the honest version requires you to disparage a named individual. Not because it is unfair to you, but because an interviewer has no way to verify it, cannot safely ask about it, and will default to caution.
Write one story. Run it against the five signals and cut any sentence that scores on none of them. Then write the three closes and say them out loud, because the part you rehearsed is rarely the part that decides it. If you want the copilot itself, the Windows download is free for five minutes a day with no card, which is enough to work out whether a drafted answer helps you or gets in your way.
Related guides
- How Many Behavioral Stories to Prepare? 8 Cover 30 Questions
- Prepare Behavioral Interview Answers in One Day: 4-Hour Plan
- HireVue Interview Format: What to Expect, Question by Question
- How Long Should a STAR Answer Be? 90-Second Timing Budget
- Threat Modeling Interview Question: STRIDE on a Real System