50 Things to Ask ChatGPT: Funny, Useful and Weird

Most lists of things to ask ChatGPT are the same 30 one-liners: "tell me a joke," "what is the meaning of life," "write a poem about cheese." You will get an answer to all of them. You will not get anything you could not have guessed.
We build AI systems for businesses, which mostly means watching how people actually talk to these models and where the answers go wrong. This list comes from that. Every question below is written so ChatGPT has to do something specific, and each section says why it works, so you can invent your own.
We also had a second, more honest source. Our posts about asking ChatGPT about yourself get a few thousand Google searches a quarter, and Search Console shows us the exact words people type. That data changed the order of this article.
What people actually click on

Our own Google Search Console data, 259 queries about asking ChatGPT or Gemini, 3 months to 7 November 2026.
The pattern surprised us. People searching for fun or funny things to ask clicked through at 7.9%, twice the rate of anything else. People searching for "interesting" or "deep" questions barely clicked at all, around 1%. Same positions in Google, roughly the same topic, very different appetite.
Our read: "interesting" and "deep" searchers already expect a generic list and skim the snippet. "Funny" searchers want to play right now. So the funny questions go first. It is a small sample from one site, so take it as a direction rather than a rule, but it matches what we see when we show ChatGPT to clients: the first thing anyone does is try to make it say something ridiculous.
Funny things to ask ChatGPT
Funny works when you give ChatGPT a role and a target. "Tell me a joke" gets you a pun about atoms. "You are a nature documentary narrator, and the animal is me trying to get out of bed" gets you something you will screenshot.

- "Narrate my morning routine like a David Attenborough nature documentary." Then tell it your actual routine. The more boring, the better it gets.
- "Write a dramatic one-star Google review of my kitchen from the perspective of my dog."
- "Explain [your job] the way a pirate would explain it to his crew."
- "Turn this text from my mom into a Shakespearean tragedy." Paste the text. Group chats work too.
- "Roast my Spotify top 5. Be brutal but affectionate." List the artists.
- "Write the terms and conditions for being my friend."
- "Give me a movie trailer voiceover for my trip to the grocery store."
- "You are a medieval peasant who has just seen a smartphone for the first time. I will describe apps, you react."
- "Write a LinkedIn post announcing that I finally folded the laundry. Make it insufferable." This one is popular for a reason.
- "Argue, with total confidence, that a hot dog is a sandwich. Then switch sides and destroy your own argument."
- "Describe my apartment as if it were a luxury real estate listing." Then describe the apartment honestly, mold and all.
- "Invent a conspiracy theory about why my houseplants keep dying."
A tip that makes all of these better: add a constraint. "In exactly six words." "Only using words a toddler knows." "As a haiku." Constraints are where the funny comes from, because ChatGPT is very good at obeying rules and the rules force it away from the safest, blandest answer.
If you want the version of this that is about you rather than your dog or your plants, we wrote a whole separate list of questions to ask ChatGPT about yourself, from fun roasts to the genuinely uncomfortable ones.
Useful things to ask ChatGPT
The useful questions all share one habit: give it the real thing, not a description of the thing. Paste the letter. Photograph the label. Upload the screenshot. ChatGPT is much better at working on material in front of it than at guessing what your situation is.

- "Here is a photo of my fridge. What can I cook tonight in under 30 minutes without going shopping?"
- "Explain this letter like I am 12, then tell me if I have to do anything and by when." Works for insurance letters, tax notices, school forms.
- "Here is my lease. List anything unusual compared to a typical residential lease, and quote the exact clause." Asking for the quote matters; it stops it inventing clauses.
- "I need to tell my landlord the heating is broken for the third time. Write it polite but firm, under 120 words."
- "Rewrite this email so it sounds less angry but still makes the same point."
- "Quiz me on [topic] one question at a time. Do not give me the answer until I try." The one question at a time part is what makes it actually teach.
- "Play my manager. I am going to ask for a raise. Push back like a real manager would." Practise the hard conversation before it happens.
- "Here is my bank statement export. Group my spending into categories and show me the three that grew the most." Remove account numbers first.
- "Plan a 3 day trip to [city] for someone who hates museums and loves food markets. Include walking times."
- "What questions should I ask a [plumber, roofer, contractor] before I hire them?"
- "I have this rash / noise in my car / error on my laptop. What are the possible causes, from most to least likely, and which ones mean I should stop and call a professional?" Good for triage, not a diagnosis.
- "Here is a photo of my living room. What is the one change that would make it feel bigger?" It is good at layout advice. If you want to actually see the room redone rather than read about it, that is what image tools like MeltFlex AI are built for.
- "Settle this argument and cite sources: [the argument]." Then click the sources. With search on, ChatGPT shows links, and the link is the part you can trust.
- "Make me a packing list for [trip], then ask me five questions to adjust it." Letting it ask you questions first is the single best upgrade for any planning request.
Weird things to ask ChatGPT
These are the questions people ask at 1am. They are fun because there is no right answer, which means ChatGPT has to actually reason instead of retrieving something.

- "If the moon were made of cheese, what would happen to the tides, and how long until it went off?"
- "Design a sport that could only be played on Mars. Rules, scoring, injuries."
- "What would a 15th century doctor prescribe for my hay fever? Then tell me which parts would actually have harmed me."
- "Invent a holiday for a day that has none, with traditions, food, and the song everyone hates."
- "Describe the color blue to someone who has never seen it."
- "If my cat could leave me a voicemail, what would it say?"
- "What is the most useless superpower you can think of, and how would you still make money with it?"
- "Write the last text message humanity sends before the sun burns out."
- "Ask me 10 weird questions and then guess which decade I was born in." Keep score.
- "You are a time traveller from 2126. What about my daily life looks barbaric to you?"
Questions that trip ChatGPT up, and why
This is the section no other list has, and it is the one that teaches you the most about how ChatGPT works. A few questions that look trivial have historically been hard for it. Newer reasoning models handle many of them, but the reason they were hard is still true, and it explains a lot of odd answers you will see elsewhere.
The reason is tokens. ChatGPT does not read letters. Before your message reaches the model, it is chopped into chunks called tokens, and each chunk becomes a number. We ran three famous trick questions through OpenAI's own open source tokenizer, tiktoken, using the o200k_base encoding that the library assigns to GPT-4o and GPT-5:

The detail we did not expect: in the usual phrasing, " strawberry", with its leading space, is a single token (ID 101830). The model never sees s-t-r-a-w-b-e-r-r-y. To count the r's it has to recall how a word it receives as one number is spelled. Without the space, the word splits into "st", "raw", "berry", and still none of those pieces is a lone r. That is why letter counting was a classic failure.
Try these yourself and watch how it handles them:
- "How many r's are in strawberry? Spell it out letter by letter first." Asking it to spell first turns one token into ten, which is why that instruction fixes it.
- "Is 9.11 bigger than 9.9?" The tokenizer hands it "9", ".", "11" and "9", ".", "9", and 11 is bigger than 9. Models have been caught saying 9.11 is larger, partly because in version numbers and dates it would be.
- "What is 12,345,678 times 87,654,321?" The correct answer is 1,082,152,022,374,638. Each number arrives as three chunks ("123", "456", "78"). Ask it to use Python, or check against a calculator. Anything important involving big numbers, do the same.
- "Write a sentence of exactly 14 words where every word starts with S." Counting its own output is hard for the same reason. Then ask it to number the words and check.
- "What is today's date, and how do you know?" The honest answer depends on whether it has search or a date in its system context. A good answer tells you which.
- "What happened in the news this morning?" Without search turned on, it should say it cannot know. If it confidently answers anyway, you have learned something about trusting it.
- "Give me five sources for [claim], with links." Then open every link. This is the best test of whether you are using a version with search, and the one with the biggest real-world stakes.
- "Which model are you, and what is your knowledge cutoff?" Compare it with what the model picker in the app says. Models are not always reliable about themselves.
None of this means ChatGPT is bad at its job. It means it is a different kind of reader than you are, and the questions that look easiest to a human are sometimes the hardest for it. Once you know about tokens, "spell it out first" and "use Python for this" become habits.
Things to ask ChatGPT about yourself
Some of the most searched questions are about you, not about the world. They work very differently, because the answer depends on what ChatGPT remembers from your past chats, and a lot of what feels like insight is really a well written horoscope.
- "Roast me based only on things you actually remember from our chats."
- "What do you know about me? Organise it by category, and tell me where each item came from."
- "What topic do I keep coming back to without realising it?"
- "Based on our chats, what am I probably avoiding?"
- "Interview me with 10 questions, one at a time, then describe me in three sentences." Works on a brand new account.
- "What is one thing I have changed my mind about this year?"
We go much deeper on these, including a two minute test that shows whether the answer is really about you, in questions to ask ChatGPT about yourself. For values and patterns rather than roasts, see our ChatGPT prompts for self-discovery, and if you use Gemini, it plays this game differently because it can see your Gmail and Photos, which we covered in questions to ask Gemini about yourself.
How many questions can you ask ChatGPT for free?
As of November 2026, OpenAI's pricing page lists unlimited text chats on the Free plan, with a footnote that usage has to be reasonable and is "subject to abuse guardrails." So for everything on this list that is plain text, you will not run out.
What is limited on Free is everything else: uploads, image creation, voice, deep research and memory. Two details from the same page are worth knowing before you try the photo questions above:
- Free has a smaller working memory per chat. OpenAI lists a total context window of 27K tokens for the instant model on Free, against 54K on Go and Plus and 128K on Pro. In practice, very long chats on Free forget their beginning sooner. If an answer suddenly ignores something you said 40 messages ago, start a new chat and paste in the key facts.
- Photo and file questions use up the limited part. The fridge photo and the lease upload count against upload limits, not the unlimited text allowance.
One more thing that changed this year: OpenAI started testing ads in ChatGPT in the US on February 9, 2026, for Free and Go users. Ads can appear below a response, and are chosen partly from the context of the current conversation. OpenAI says ads do not influence the answer itself and that advertisers do not see your chats. Plus, Pro, Business, Enterprise and Edu accounts do not get ads. If you ask ChatGPT "which vacuum should I buy," just keep in mind that the box under the answer and the answer are two different things.
Things not to ask ChatGPT
A short list, because it saves people real trouble:
- Anything with passwords, full card numbers or ID numbers in it. Remove them before you paste a statement or a form. On personal accounts, chats can be used to improve models while the Improve the model for everyone setting is on, which you can turn off under Data controls.
- A diagnosis, legal ruling or tax decision you will act on without checking. Use it to understand the question and prepare for the professional, as in question 23. Not to replace them.
- "Is this true?" without search. Ask for sources and open them.
- Questions about a real private person. It may make things up, and you do not want to be the one who passes them on.
Frequently asked questions
What is the funniest thing to ask ChatGPT?
The ones with a role and a specific target: "narrate my morning like a nature documentary," "write a one-star review of my kitchen from my dog's perspective," or "write a LinkedIn post announcing I folded the laundry." Adding a constraint, like six words or a haiku, makes almost any funny prompt funnier.
What are good questions to ask AI?
Questions where you supply the material: a photo, a document, a real situation. "Explain this letter and tell me what I have to do by when" is a better question than "what is a tax notice," because it can only be answered well by working on your actual letter.
Why can't ChatGPT count the letters in strawberry?
Because it reads tokens, not letters. In OpenAI's o200k_base tokenizer, " strawberry" with a leading space is a single token, so the model never sees the individual letters and has to recall the spelling. Asking it to spell the word out letter by letter first usually fixes it, and newer reasoning models often do that on their own.
Is there a limit to how many questions you can ask ChatGPT?
For text, the Free plan is listed as unlimited, subject to OpenAI's abuse guardrails, as of November 2026. Uploads, image creation, voice, deep research and memory are limited on Free and expanded on paid plans. Free also has a smaller context window per chat, so very long conversations lose track of early details sooner.
Can ChatGPT answer questions about current events?
Only when it uses search. Without search, its knowledge stops at a training cutoff and it should say it does not know. With search, it shows source links; open them, because the source is what you can verify.
Do ads change ChatGPT's answers?
OpenAI says no: ads run on separate systems, appear below the response, and advertisers cannot shape or rank answers. Ads are being tested for Free and Go users, and paid Plus, Pro, Business, Enterprise and Edu accounts do not see them.
The short version of this whole page: ChatGPT is only as interesting as the material and the rules you give it. A good question is mostly a specific one.
We spend our days building AI that answers from a company's real data instead of from what sounds plausible: phone receptionists, document search, internal assistants. If you want to see where that would pay off in your business, book a free call.
Charts by MeltFlex from our own Search Console data and OpenAI's tiktoken library (o200k_base, run 9 November 2026). Photos released under CC0 1.0 and hosted on Wikimedia Commons: "Smiling woman with a phone" by rawpixel.com, "You're so funny" by Tanja Heffner, "Man with smartphone and laptop" by Alejandro Escamilla and "Late Night Wifi" by Isabell Winter.

