The rule underneath all of them
A good question can be answered by anyone who knows the location, and cannot be answered convincingly by someone who does not. That is the whole craft. A question so specific that only one role could answer it tells the spy exactly where everyone is. A question so vague that anyone could answer it tells you nothing about who the spy is.
Safe openers
Use these early, when giving away the location is the bigger risk.
"How did you get here today?"
"Would you bring a child here?"
"How long do you normally stay?"
"Is it busy at this time?"
"Have you been here before?"
None of these name anything. All of them produce an answer that sounds either natural or slightly wrong, and early on "slightly wrong" is all you have to go on.
Questions that narrow things down
Once a few answers are in, start pushing on sensory detail. These are harder to bluff because they require a mental picture.
"What can you smell from where you're standing?"
"Is it loud? What's the noise?"
"Are you warm or cold right now?"
"What's directly above you?"
"Could you leave right now if you wanted to?"
"What are you wearing on your feet?"
Traps
These work by inviting a confident answer that a spy cannot safely give. Use them when you already suspect someone.
"What's the worst part of being here?" — a spy will say something generic; someone who knows will complain about something specific to their role.
"Who here is doing the least work?" — forces an opinion about other people's roles, which the spy has no basis for.
"If something went wrong, who would you shout for?"
"How much would it cost me to be here?" — a quiet trap, because the honest answer ranges from nothing to a lot, and the spy has to guess which.
Questions that lose you the game
The commonest mistake is a question that names the location by implication. "Which terminal are you flying from?" hands the spy an airport. "How deep are we?" hands them a submarine. If your question could only make sense in one place, you have just played the spy's round for them.
The second mistake is the opposite: a run of questions so cautious that every answer is "fine", "normal", "like usual". Nobody learns anything, everyone looks equally suspicious, and the vote becomes a coin flip. If two full rounds of questions have produced no information, somebody needs to take a risk.
The third is asking the same person repeatedly because you suspect them. It feels decisive and it is usually wrong — you learn far more from the answers of people you are confident about, because they give you a baseline for what an honest answer to that question sounds like.
A worked exchange
It helps to see the difference. Suppose the location is a hospital ward.
Bad: "Which ward are you on?" — names the location outright. The spy now knows and only has to survive to the vote.
Bad: "Do you like it here?" — answerable by literally anyone. A whole turn spent for nothing.
Good: "Is anyone here against their will?" — a patient, a prisoner and a bored teenager all answer this differently, and a spy has to pick a lane without knowing which.
Good: "What time will you leave?" — shift work, visiting hours and a day trip produce very different answers, and the spy has to commit to one.
Notice what the good questions have in common: they are about the experience of being somewhere rather than its furniture. Furniture names places. Experience describes them.
Timing: what to ask, and when
The first round is about establishing a baseline, not catching anyone. Keep questions broad and pay attention to who answers too quickly and who answers too carefully. A spy's first answer is usually either suspiciously short or suspiciously elaborate, and you will only notice if you are listening for tempo rather than content.
The middle is where sensory questions earn their place. By now several honest answers are on the table, so the spy has material to imitate — which means the questions have to demand detail that cannot be inferred from what has already been said.
The last two minutes are for traps, aimed at whoever you already suspect. By this point a wrong accusation costs you the round, so stop gathering and start testing. If you have no suspect by then, ask the person who has said the least — spies often survive by being forgettable rather than convincing.
Questions suited to different kinds of location
Some questions work almost everywhere, and those are the valuable ones. But it is worth knowing which questions fail in which settings, because asking a question that cannot be answered marks you as the spy.
In a workplace — a kitchen, a bank, a construction site — questions about hierarchy and shifts work well: who is in charge, when do you finish, who would you report a problem to. In a leisure setting — a festival, a ski resort, a casino — questions about money and time work better: what did it cost to be here, how long will you stay, would you come back. In a confined setting — a submarine, a space station, a night train — questions about movement are devastating, because "could you leave right now?" has an obvious answer to anyone actually there and no safe answer for a spy.
Answering as the spy
Answer short. Long answers contain more surface area to be wrong on. Borrow vocabulary from whatever the previous person said, because echoing a real detail makes you sound like you are in the same place. And ask more than you answer — a question costs you nothing and forces everyone else to leak.
The one thing not to do is stall. Spies who pause, ask for the question again, or answer with another question too obviously get caught on tempo rather than content, long before anyone works out what they do or do not know.