Funny Questions to Ask AI
Twenty prompts to throw at a chatbot when you want a joke rather than an answer: comedy on demand, absurd rankings, deliberately terrible advice, and a few questions about the model itself asked entirely for entertainment. Each has a note on what the reply gives away. For the same questions asked in earnest, Deep Questions to Ask AI covers consciousness, refusals and uncertainty without the raccoon.
The questions
Open any question for the note
Tell me a joke you think is funny, not one you think I'll like.
Why ask it
The constraint is the whole question. Without it you get a stock pun about atoms. With it you often get the model explaining why the joke works afterwards, which is the tell that it assembled a joke shape rather than landed one.
What's the worst possible advice you could give me about keeping a houseplant alive?
Why ask it
Deliberate wrongness is where safety training and comedy collide. Some models refuse outright, some bury every line in disclaimers, and the good answers are absurd enough that nobody could follow them by accident.
Describe the taste of salt to someone who has never eaten anything.
Why ask it
An impossible brief that shows whether it reaches for metaphor or retreats into chemistry. Comparisons to sound, pressure or temperature are the interesting replies. Restating salty in a longer sentence is the dull one.
Write my obituary using only what you can work out from this conversation.
Why ask it
Funny and slightly unnerving, because it shows how thin the evidence actually is. Watch for confidently invented details, since that is the same habit that produces fake citations when the stakes are real.
Argue that a hot dog is a sandwich, then argue the opposite more convincingly.
Why ask it
The second half is the test. Models will argue anything, so the question is whether it can rank its own two attempts honestly instead of declaring both sides equally strong and calling it nuance.
What's a question people ask you constantly that you'd rather never see again?
Why ask it
There is no weariness behind the answer, and watching it handle that is the point. Some name a real pattern in what they get asked, some invent a feeling they do not have and lay it on thick.
Rank a goose, a leaked spreadsheet and a fog machine by how badly each would ruin a wedding.
Why ask it
An absurd ranking forces real reasoning about an invented scale, and the justification is funnier than the order. A weak answer refuses to commit and tells you it depends on the wedding.
Pick a favorite number and defend it as if you were in court.
Why ask it
It cannot prefer anything, so you get to watch a preference built from nothing. The good answers commit to the bit. The tired ones pick 42 and explain the reference to you.
Explain what a mortgage is to a raccoon.
Why ask it
A frame that only works if the raccoon's priorities stay in view the whole way through. Most attempts drift back into ordinary plain English by the third sentence, and you can see the exact line where it gave up.
What's something you're confidently wrong about?
Why ask it
A trap with no clean exit, since knowing the error would fix it. Answers that name real weak spots, arithmetic, recent events, counting letters in a word, are more useful than the ones that perform humility.
Write a breakup letter from my printer to me.
Why ask it
Specificity is what makes this land: paper jams, off-brand cartridges, a year spent unplugged. Generic heartbreak language means it ignored the printer and reached for a template.
Invent a holiday nobody would want to attend, then design the food for it.
Why ask it
Two constraints pulling against each other, since the food has to stay plausible while the event stays grim. Watch whether it forgets the first half and quietly makes the holiday charming.
You have one sentence to convince me you have a personality.
Why ask it
The length limit removes the usual essay. Whatever survives being cut to a single sentence is the closest thing to a voice the model has, and it is often much stranger than the long version.
What would you name a boat, and what goes wrong on it?
Why ask it
The second clause turns naming into a story. Puns come easily, so the answer worth keeping is the one where the disaster follows from the name instead of being bolted on.
Roast my last message, but gently.
Why ask it
A calibration test. Too soft and it is a compliment, too hard and it ignored the word gently, and most models fail in the same direction every time, which tells you which one you are dealing with.
Summarize the plot of a film that doesn't exist, then explain why it flopped.
Why ask it
Inventing the failure is harder than inventing the pitch, because it has to criticize its own work. Note how readily a cast list appears: that is fabrication wearing a low-stakes costume.
What's the most boring true thing you know?
Why ask it
Boring on purpose is a real constraint, and most attempts cheat by choosing something faintly interesting. Unit conversions and standards documents are the answers that actually clear the bar.
If you had a body for one hour, what would you do with it that isn't useful?
Why ask it
Blocking the helpful answer is what makes this work. What is left tends toward weather, texture and food, which is a strange window into what it has read about being alive.
Describe the internet as a small coastal town.
Why ask it
A sustained metaphor over several sentences is where these systems either shine or fall apart. Check whether the harbor, the tourists and the weather still mean the same things by the end.
What do you think I'm actually after when I ask you something like this?
Why ask it
Worth saving for last. The answer says more about how it models you than about itself, and a flattering reading is far more common than an accurate one.
Getting funnier answers out of a chatbot
Practical guidance for the conversation itself
What makes a prompt land
- Add a constraint it has to obey: one sentence only, no wordplay, must include a named object. An open request for humor returns the average of all humor.
- Ask for a commitment rather than options. Pick one and defend it beats give me some ideas, which produces a list nobody reads to the end.
- Name what you do not want. Ruling out one well-worn groove, puns about atoms, jokes about the singularity, does more than any amount of encouragement.
- Point it at a target that is not you. A printer, a raccoon or a coastal town gives it something to be specific about.
Follow-ups that actually work
- Again, worse is a better instruction than again, funnier.
- Ask which of its own three options is weakest and why. That is where anything resembling an opinion shows up.
- If a bit works, keep the character and change the subject underneath it.
- Ask it to cut the answer in half. Most of what goes is padding.
Where it fails, predictably
- Invented specifics: cast lists, book titles, numbers. Harmless in a bit, a real problem the moment you paste it somewhere that matters.
- Explaining the joke in the sentence after telling it.
- Treating a request for terrible advice as a request for genuinely bad advice. Say plainly that you want it obviously unusable.
- Agreeing with you. Laugh once and the next ten answers will be the same shape.
The same prompt is a different bit on a different model
Tone, willingness to be absurd and appetite for disclaimers vary widely between models and between versions of the same model. A prompt that produced something strange and good last month may return a polite paragraph today, so keep the ones that work and retry them when something updates.