Ask the Two Questions Only Someone Who Read the Role Would

What Makes a Closing Question Actually Yours?

A question is yours when you can follow up on the answer without a script. The stock lists (what's the culture like, what does success look like in this role) are recognizable precisely because they work for any job at any company, which means they demonstrate nothing about whether you read this one. An interviewer who has heard the same three questions four times this week is not learning anything new the fifth time either.

The test is simple to apply to a question before you ask it: if the interviewer gave a surprising answer, could you say something intelligent back, or would you just nod and move to your next prepared line? A question you can't follow up on wasn't built from anything you actually know. It was built to fill the silence, and it reads that way.

There's a reason an unscored, freeform stretch of conversation tells an interviewer less than they think it does. In a 2013 study, 76 undergraduates predicted classmates' semester grades, and the predictions they made after conducting an unstructured interview tracked the real result less closely than the classmates' prior grades alone did ². That was a lab task rather than a hiring decision, so the size of the gap doesn't carry over, but the mechanism does: an unscripted exchange can add noise to concrete evidence that was already sitting there. A generic closing question is that kind of exchange, dressed up as a moment that matters.

The stock lists were never secret. Career sites have published the same thirty questions for years, and a model just makes them faster to retrieve. What changed is that an interviewer now hears the identical phrasing from candidate after candidate in the same week, so the list went from mildly generic to immediately placeable, which is a worse position to be in than having nothing prepared.

Build Three Questions From What You Actually Have

Build from three sources that are different for every interview you do: the posting, the conversation you just had, and how the work gets judged early on. There is no clever universal question that impresses every interviewer at every company, and hunting for one is what sends most people back to the recycled list.

• An ambiguity in the posting. Job descriptions are often written by someone summarizing a role they don't fully own, and they leave gaps. If the posting lists a responsibility that could mean two different things, ask which one it actually is in practice.
• Something said in this room. If someone mentioned a recent change, a reorg, a new tool, a shipped feature, ask about it directly. This is the easiest tell that separates a real question from a memorized one, because it could only have been written after this specific conversation started.
• How the work gets judged early on. Ask what the first real project looks like and how it gets evaluated in the opening months. A structured interview process asks every candidate the same questions in the same order and rates them all on one common scale ¹. Whether the closing segment sits inside that scale or outside it varies by company, so assume it counts the way the rest of the conversation did. It's also information you'd want regardless of what it signals about your candidacy: knowing what doing well looks like in month one changes how you'd actually approach the job if you got it.

One good question from each source beats ten generic ones. You will run out of interview time before you run out of a long list anyway.

Write the three down before the interview, even loosely, because a question built live under nerves tends to drift back toward the generic version your brain already has memorized. Having a specific ambiguity or a specific moment from the conversation already noted gives you something concrete to reach for instead of reaching for a stock phrase out of habit.

Use AI to Pressure-Test the Question, Not Write It

The legitimate use of a model here comes after you've drafted the question, not instead of drafting it. Write your two or three questions first, from the posting and the conversation, then ask a model what a good answer and a bad answer to each would sound like. That tells you whether the question is actually sharp enough to distinguish anything, before you're the one finding that out live in the room.

What you're checking for is whether you'd notice a vague non-answer if you got one. What follow-up questions expose whether someone actually understands the answer they just gave? covers the same skill from the interviewer's side, and it transfers directly: specify, invert, then ask what would have to be true for the answer to be wrong. If you can't tell a strong answer from a deflection on your own question, sharpen it before you walk in.

The question about how the work gets judged is worth a specific mention here, because it does double duty. Beyond showing you read the room, it's a fair thing to actually want to know before you accept an offer: how the first stretch of a role gets evaluated. How do you structure an AI-heavy new hire's first 90 days? shows the kind of answer a thoughtful employer might give when this question is asked well, which is worth having read once in advance.

One more use worth naming: if you're unsure whether a question reads as too basic or too aggressive for the room, describe the interview so far and ask a model to guess how the question would land. It won't always be right, and ignoring the suggestion costs nothing, but a second read on tone is cheap.

See how it works >