Funny ChatGPT prompts work best when they give the model a specific, absurd persona or format rather than just asking it to “be funny.” The prompts on this page are grouped by type: roleplay prompts, rewrite-in-a-silly-voice prompts, and fake-argument prompts, all copy-paste ready.
The short version: Most “funny ChatGPT prompts” lists give you generic joke requests that produce flat, safe, forgettable output. The ones that get a laugh are narrow, specific, and put ChatGPT in a ridiculous position it has to commit to, and I’ve listed the exact ones below along with the ones I tried that fell completely flat.
- A passive-aggressive email from an office fridge to the oat milk thieves, in the voice of a Victorian solicitor.
- A medieval knight who has just discovered TikTok writes his diary entry.
- Explain how a printer works as a conspiracy theorist who believes printers are sentient.
- A one star Yelp review for the sun, from someone furious it set without warning.
- A breakup text from a Roomba to the sofa it has been bumping into for three years.
- A strongly worded noise complaint from a ghost about the living people upstairs.
Why most funny prompt lists don’t work
I’ve read a fair few of these lists while researching this post, and I’ll say the uncomfortable bit up front: asking ChatGPT to “tell me a funny joke” or “be witty” is the fastest way to get something that reads like a Christmas cracker written by a committee. It’s not that the model can’t be funny. It’s that vague comedy instructions get you the safest, most averaged-out answer in its range, because that’s what the training has optimised it toward. Broad and unfiltered is not this model’s strength. Narrow and weird is.
I tested this on a wet Friday in my office in Buckinghamshire, mostly to procrastinate on a client proposal I didn’t want to write. I asked “make me laugh” first, plain and simple. I got a knock-knock joke about a broken pencil., that was it. Pointless.
Then I gave it a job with rules attached, and that’s when things got good. Constraints are the whole trick. A specific character, a specific format, a specific absurd combination of things that don’t belong together. That’s the pattern behind every prompt in this post that landed.
The prompts that got real laughs out of my team
I ran roughly 40 prompts over three separate Fridays with my small team (four of us, all a bit sleep-deprived by 4pm), reading the outputs aloud in our Slack huddle. Here’s what got genuine laughs, not polite ones.
- “Write a passive-aggressive email from a office fridge to the people who keep stealing its oat milk, written in the voice of a Victorian solicitor.” This one produced three paragraphs of formal legal threat over milk theft and it was one of the funniest things I’ve had ChatGPT write all year.
- “You are a medieval knight who has just discovered TikTok. Write your diary entry for the day.” The output included the knight describing “the dance of the peasants, performed for fifteen seconds, then judged by the entire kingdom.”
- “Explain how a printer works, but you are a conspiracy theorist who believes printers are sentient and hate you personally.” This one got the biggest laugh in my office. It called the printer “a small grey traitor that only jams when it senses weakness.”
- “Write a Yelp review for the sun, from someone who is furious it set without warning them.” One star. “Advertised as reliable. Left without a five minute warning like every other man in my life.”
- “You’re a very tired, very unimpressed cat reviewing your owner’s life choices in the style of a Michelin food critic.” dry and mean in a way that felt exactly like a cat.
- “Write a breakup text from a Roomba to the sofa it’s been bumping into for three years.” Short, oddly moving, and stupid in the best way.
- “Give me a corporate mission statement for a company that only sells one broken umbrella.” It generated a convincing bit of nonsense marketing speak, complete with “our commitment to partial dryness.”
- “Pretend you’re an overly enthusiastic estate agent trying to sell a haunted, collapsing shed for £900,000.” This one leaned into British property market absurdity and it worked because that world is already slightly ridiculous, ChatGPT just amplified it.
- “Write a strongly worded noise complaint from a ghost about the living people upstairs.” “I have haunted this house since 1847 and I have never once heard footsteps this loud.”
- “You are a Roman emperor who has just found out about the concept of ‘Monday’.” Full imperial fury directed at the calendar.
Notice what all ten have in common: a specific character with a defined worldview, a specific mundane target, and a format (email, review, diary entry, mission statement) that has its own built-in comic rhythm because we already know what those things are supposed to sound like when they’re normal. Comedy needs a shape to break, and format gives it that shape.
The Friday afternoon experiment that taught me the actual lesson
The moment that stuck with me most wasn’t even one of the funny outputs. It was watching what happened when one of my team members, Dan, tried to “improve” the printer conspiracy prompt by adding more instructions. He added “make it really funny and absurd and over the top and unexpected.” The output got worse. Longer, more self-aware, less funny. It read like it was trying too hard because, in a sense, it was, the model had more competing instructions to satisfy and less room to just commit to one voice.
That was the bit that changed how I write these prompts now: less is more, and specificity beats enthusiasm. Tell it who it is and what mundane thing it’s dealing with, then get out of the way. Don’t tell it to be funny. Funny is the byproduct of a good character in a stupid situation, not something you can order directly like a side of chips.
How to write your own funny ChatGPT prompt in five steps
- Pick a mundane, boring, everyday object or situation. A printer, a fridge, a bus stop, a Tuesday, a Wi-Fi password, a bin collection schedule. The more ordinary, the better it contrasts with what comes next.
- Give ChatGPT a wildly unsuitable character to inhabit. A Victorian solicitor, a Roman emperor, a Michelin critic, a conspiracy theorist, a heartbroken teenager, a pirate captain. The gap between the character’s usual world and the mundane object is where the joke lives.
- Choose a format with its own built-in rhythm. Email, Yelp review, diary entry, resignation letter, ransom note, wedding speech, LinkedIn post. Formats carry structure you don’t have to write yourself.
- Cut every instruction that isn’t the character, the object, and the format. No “make it funny,” no “be witty,” no “really lean into the humour.” Trust the collision of the first three things to do the work.
- Ask for a second draft “but angrier” or “but more heartbroken” or “but shorter.” This is the step nobody mentions. The first output is rarely the funniest one. The second pass, pushed toward one specific emotion, is usually where the best line shows up.
The prompts that sound funny but always fall flat
Worth naming the failures too, because every list I’ve read skips this and it wastes your time. These consistently produced dull output when I tested them:
- “Tell me a joke.” You’ll get a pun about atoms or a chicken crossing a road. Every time.
- “Write something hilarious about [topic].” The word “hilarious” in the prompt seems to make the output less hilarious, not more. I’ve no idea why, but it happened every single time I tried it.
- “Roast me.” This one gets watered down fast because the model is cautious about insulting a real person, even one who asked for it, so you get soft, generic digs instead of anything sharp.
- “Write a funny poem.” Poems asked for directly tend to come out greeting-card sincere with a joke bolted on at the end, rather than funny throughout.
The pattern in the failures is the same pattern in reverse: the word “funny” or “hilarious” sitting right there in the prompt seems to nudge the model toward safe, telegraphed jokes rather than character-driven ones. Ask for a character and a situation instead, and let the humour arrive as a side effect.
A few more to try if the first batch worked for you
- “Write my houseplant’s suicide note, but it’s just really passive-aggressive about being underwatered.”
- “You are a GPS satnav that has developed a grudge against one specific roundabout.”
- “Write a toddler’s formal complaint to Father Christmas about the size of last year’s stocking.”
- “You’re an overworked NHS receptionist explaining, in extreme detail, why the printer is the real reason your day is ruined.”
- “Write a break up letter from a gym membership to someone who hasn’t shown up since January the 3rd.”
If you run a small business account and want to use one of these for social content, test it privately first. What’s funny in your team Slack at 4pm on a Friday doesn’t always land the same way on a public feed, and I’ve seen brands post AI jokes that read as try-hard rather than charming. Read it out loud to one other person before you post it anywhere with your company name attached.
Related reading: fun ai prompts to try.
This is one of 30 ChatGPT prompts guides; the full index is ChatGPT prompts: every guide in one place.
Frequently asked questions
What is the funniest thing to ask ChatGPT?
Prompts that put ChatGPT in a specific, mundane situation as an unlikely character tend to land best, things like a Victorian solicitor writing about stolen oat milk or a Roman emperor discovering the concept of Monday. Direct requests like “tell me something hilarious” almost always produce flatter, safer output than a specific character-and-situation prompt.
Why isn’t ChatGPT funnier when I ask it directly for jokes?
When you use the words “funny” or “hilarious” in a prompt, the model tends to default to safe, well-worn joke formats like puns or knock-knock jokes rather than anything sharp or unexpected. Giving it a defined character and a mundane target to react to, without mentioning the word funny at all, generally produces better comic output.
Can I use funny ChatGPT prompts for my business’s social media?
You can, but read the output aloud to a colleague before posting it under your brand name, since jokes that work well between friends in a chat can read as try-hard or off-tone on a public business account. Test any AI-generated humour with one other person first and keep it light and self-deprecating rather than aiming for anything edgy.
Do these funny prompts work the same way on other AI chatbots like Claude or Gemini?
The same underlying trick, giving the model a specific character and a mundane situation rather than asking directly for humour, tends to work across most large language models, since they’re all trained in broadly similar ways to avoid risky or edgy output by default. The exact wording and tone of the results will differ between tools, so it’s worth testing the same prompt on more than one if you want to compare.
