The round is still running
Almost everyone treats the last ten minutes as the cool-down, and interviewers do not. Your questions go into the written feedback, often on a line of their own, because they are the only part of the hour you controlled completely. What you choose to be curious about is read as evidence of what you attend to at work, and it is unusually hard to fake — you cannot ask a good question about operating a system if you have never operated one.
There is a blunter reason too: a candidate with no questions has said they are not evaluating the job, which reads as indifference or as an assumption that any offer gets accepted. "No, I think you have covered everything" is the most common answer given here and it produces nothing for the file.
Aim at the person, not at the company
The mistake that wastes these ten minutes is asking a company-wide question of someone who cannot
answer it. A recruiter cannot tell you how code review works; a staff engineer cannot speak with
authority about calibration; a hiring manager can do both but will filter anything about team health,
because they are recruiting. Match the question to the seat.
- Recruiter or talent partner. The mechanics: remaining stages, who you will meet and in what
role, how long decisions take, what the level is called internally.
- A future peer. The truth about the work: what they did last week, which part of the codebase
they dread, how a change reaches production, how much of their time goes on unplanned work. Peers
answer honestly more often than anyone else in the loop, because they are not being judged on the
answer.
- The hiring manager. Ownership and direction: how priorities are set and by whom, what they
would want this seat to have changed in six months, where they disagree with their own manager
about the roadmap.
- A skip-level or director. Why the role exists in this shape: what would make the team's next
year a failure, what they are deliberately not doing, how the team is funded.
What separates a strong question from a filler one
Two properties. It could not be answered by the careers page, and it asks about a mechanism rather
than a value. "Do you have a good engineering culture?" asks for a value and gets a yes. "When
someone wants to change a service another team owns, what happens?" asks for a mechanism, and the
answer is informative whichever way it goes.
The mechanisms worth asking about are how decisions travel, how work arrives, and how failure is
handled — the three things that determine whether a job is pleasant and the three no job description
covers. Some that reliably produce real answers:
| Question | Why the answer tells you something |
|---|
| "How does something get from merged to in front of a user?" | Length and manualness of that path predicts your week better than the stack does |
| "What proportion of a sprint is typically unplanned work?" | A hesitation before answering is itself the answer |
| "What did the last incident turn out to be, and what changed afterwards?" | Separates a team that learns from one that assigns blame |
| "What would you want whoever takes this role to have fixed in six months?" | Reveals whether the role has a purpose or is headcount |
| "Where do you and your manager currently disagree about the roadmap?" | Tests whether disagreement is survivable here |
| "What happened to the person who had this job before?" | Occasionally the most important fact in the process |
Two questions can be worth more than five. Volume is not the signal, and a rapid list reads as
prepared rather than curious.
The follow-up is the part candidates skip
Asking is half of it. The answer is data, and the scored behaviour is whether you do anything with
it. Ask about unplanned work, hear "it varies", and the natural next move is "what did it look like
last sprint?". That single follow-up gets you the real answer and demonstrates in front of the panel
that you probe vague claims — which is a competency they have been assessing all hour. One probe per
question is plenty; you are asking, not cross-examining. But nodding at every non-answer wastes the
ten minutes in both directions.
What not to ask, and why in each case
Compensation, holiday, notice periods and remote-working policy belong with the recruiter or in the
offer conversation — not because they are shameful, but because the person in front of you usually
cannot answer and it costs you a question. Anything answered on the first page of the company website
is worse than silence, because it shows you did not look.
Two subtler ones. Avoid the question that is a boast in disguise — "how would you use someone with my
depth in event-driven architecture?" — because everyone can hear it. And be wary of "what are the
challenges here?", which is not a bad question but is so common it produces a rehearsed answer; the
specific version, "what about this codebase does a new joiner always underestimate?", gets you
something real.
Prepare, too, for your questions to be pre-empted. In a well-run loop the first two will have been
answered during the hour, so have four, and being able to say "you already covered the release
process, so instead — " reads as attentiveness rather than script.
Sample phrasing when you want the unfiltered version
Interviewers filter by default, and framing changes what you get. Asking for the most recent concrete
instance rather than a general characterisation works better than any other technique:
Weak: "How is the work-life balance?"
Better: "When was the last time someone on the team worked a weekend, and what was going on?"
Weak: "Is the team collaborative?"
Better: "Walk me through the last technical decision the team disagreed about. How did it get
settled?"
A request for the most recent case is hard to answer evasively, and the answer is nearly always the
truth about the team.
The night before
Write six questions the night before, tagged with who you will ask each of, and include one that only
applies to this company — derived from a change in their pricing, a job ad implying a reorganisation,
a talk one of their engineers gave. That question does more for you than the other five together,
because it is the only one you could not have asked anybody else.
Then decide what answers would make you decline. If nothing you could hear in those ten minutes would
change your mind, you are not really asking questions, and the panel will hear the difference.
© 2026 Preptima. Originally published at preptima.com (opens in a new tab).