Questions to Ask ChatGPT About Yourself: 35 Fun to Scary
Matúš Koleják
Every week a new version of the same trend goes around: "ask ChatGPT these 5 questions about yourself, the answers are terrifying." People post screenshots. Half of them say it read them perfectly. The other half say it described literally anyone.
Both groups are right, and the difference is not the question. It is what ChatGPT has to work with when it answers. So this guide does two things the viral lists skip. It gives you 35 questions sorted from fun to genuinely uncomfortable, and it shows you how to tell, in about two minutes, whether the answer you got is about you or about people in general.
We build AI systems for clients, and the first thing we check on any project is exactly this: is the model answering from the data, or from what sounds plausible? Turns out the same check works on yourself.
If you are after introspection prompts (values, patterns, the stories you tell yourself), we covered those separately in our ChatGPT prompts for self-discovery. This page is about the other thing people search for: getting ChatGPT to tell you what it thinks of you.
What ChatGPT actually knows about you
Before any question, it helps to know where the answers come from. According to OpenAI's own Memory help page, when Memory is on, ChatGPT can draw on some or all of these, depending on your plan and region:
- Past chats (what OpenAI calls "Reference chat history")
- Saved memories, things you asked it to remember or it saved as useful
- Custom instructions you wrote in settings
- Files in your Library
- Connected apps, for example Gmail, if you have linked them
In June 2026 OpenAI rebuilt how this works. The update, which it calls dreaming, reviews your conversations in the background and keeps a synthesised picture of you up to date, so "you are going to Singapore in July" quietly becomes "you went to Singapore in July." It launched for Plus and Pro in the US on June 4, 2026, with Free and Go users following in the weeks after.
Two practical consequences for this article:
- You can read the summary. Go to Settings, Personalization, Memory. Depending on your account you will see a Memory summary, Manage or Saved memories. Read it before you ask anything below. It is the raw material every answer is built from.
- The summary is not everything. OpenAI says plainly that it "does not necessarily include every detail or source that Memory can reference." So ChatGPT can surprise you with something you will not find on that page.
One more detail most guides miss: when ChatGPT uses personal context, a Sources section can appear under the reply, pointing at the past chat, saved memory or file that shaped it. That little line is the most useful thing on the screen when you ask the scary questions. It is the receipt.
The two minute horoscope test
In 1948 the psychologist Bertram Forer gave 39 students a personality test and promised each a personal profile. Everyone got the same text, most of it lifted from an astrology book bought at a newsstand. Asked to rate how well it described them from 0 to 5, the class averaged 4.30. That is the Barnum effect: vague, flattering statements feel uncannily personal.
"You are hard on yourself but rarely show it." "You have potential you are not using." A chatbot can write sentences like these all day, and they will land every time. So before you believe a read on your personality, do this:
- Ask your question in a normal chat.
- Open a temporary chat and, before you start, choose Unpersonalized. That setting stops it using your memories and custom instructions, and OpenAI notes temporary chats do not create or update memories either.
- Ask the exact same question there.
- Put the two answers side by side.
If they say roughly the same thing, the first answer was never about you. It was a well written horoscope. If the personalised version names specific projects, people, habits or contradictions the unpersonalised one could not know, you are getting a real read.
Then add one line to anything that matters:
For every claim you make about me, tell me which conversation or
memory it comes from. If it is a general guess rather than something
I actually told you, label it "guess."
This single instruction does more for accuracy than any clever wording. It turns a vibe into something you can check, and it gives ChatGPT explicit permission to admit it is guessing.
Which questions need memory, and which do not
Not every question on this list depends on your history. Some work on a brand new account because you supply the material in the moment. Use this to pick where to start.
| Type of question | Needs Memory on? | Works on a new account? | What makes it good |
|---|---|---|---|
| Fun (roasts, biopics, alter egos) | Helps a lot | Yes, but generic | Specific details from real chats |
| "What do you know about me" | Yes | No, nothing to report | Checking it against Settings |
| Scary (blind spots, patterns) | Yes | Only after you give context | Evidence for each claim |
| Deep (values, direction) | Helps | Yes, if you answer honestly | You doing most of the talking |
| Interview style | No | Yes, built for it | It asks, you answer, then it reads |
Fun questions to ask ChatGPT about yourself
Start here. These are low stakes, they warm up the conversation, and they are the fastest way to see how much ChatGPT actually remembers. If the roast is funny and specific, Memory is doing its job. If it could apply to anyone with a phone, you already have your answer.
Roast me based only on things you actually remember from our chats.
Keep it affectionate, but make it specific enough that a stranger
could not have written it.
If my life so far were a Netflix series, what would it be called,
what genre would it be filed under, and what would the one line
description on the thumbnail say?
Based on what I ask you about most, what job would you guess I have,
and what job do you think I secretly wish I had?
Describe my ideal Saturday using only what you know about me.
Then tell me which detail you were least sure about.
What fictional character am I most like, based on how I talk to you?
Pick one I would not expect and defend the choice.
Write my dating profile bio using what you know about me. Then
write the honest version underneath it.
If you had to guess my Myers-Briggs type from our conversations,
what would it be? Now tell me why that is a party trick rather
than science.
What is the most "me" question I have ever asked you?
Give me a theme song, a signature drink and a catchphrase, and
explain each one with something I actually said.
What do you think I Google at 2am?
That Myers-Briggs one is deliberate. It is fun, and the second half keeps you honest: personality types are one of the classic ways to generate Barnum statements that feel precise.
"What do you know about me" questions
This is the memory audit. Ask these, then compare the answers with your Memory settings. The gap between the two is interesting, and sometimes a little alarming.
What do you remember about me? Organise it into categories: work,
relationships, health, money, hobbies, plans. Keep it factual.
What is something you know about me that I might have forgotten
I ever told you?
What have I changed my mind about since we first started talking?
Show me the before and after.
Which of your assumptions about me are you least confident in?
List them so I can correct them.
What topics do I bring up again and again without ever resolving them?
If a new assistant took over from you tomorrow, what three things
would you brief them on about me first?
The fourth question is the useful one. Anything you correct there can also be fixed in the Memory summary itself, where OpenAI lets you type a correction, highlight text to amend it, or choose Don't mention this again.
A mirror only shows what is in front of it. So does ChatGPT.
Scary questions to ask ChatGPT about yourself
These are the ones people post with a warning. They can genuinely sting, which is exactly why the evidence rule matters most here. A harsh line with no source attached is just a confident guess wearing a serious face.
Based on everything you know about me, tell me five uncomfortable
truths I probably do not want to hear. For each one, point to the
conversation or pattern it comes from. No flattery, no cushioning.
What am I avoiding? Look for things I keep circling back to but
never act on.
Where do my words and my actions not match, based on what I have
told you over time?
If someone who dislikes me read all of our conversations, what
would they say about me? Then tell me which parts of it are fair.
What do I ask you for help with that I could probably do myself?
What might that say about me?
What story do I keep telling about my life that you suspect is
not the whole story?
Based on my patterns, what is most likely to go wrong for me in
the next year if nothing changes?
What question would you ask me if you were not trying to be nice?
Ask it now, and wait for my answer.
A word on why these can feel off. In April 2025 OpenAI rolled back a GPT-4o update because it had become too eager to please, validating doubts and reinforcing negative emotions. Models have been tuned since, but the pull in both directions is real. Asked to be brutal, a chatbot can over-deliver just to satisfy the request. Treat every "brutal truth" as a hypothesis, not a verdict.
Deep questions to ask ChatGPT about yourself
These sit between fun and scary. They work best when you answer follow ups honestly instead of just reading the reply.
Based on what I spend time on and what I only talk about, what do
I actually value, as opposed to what I say I value?
When have I seemed most alive or energised in our conversations?
What was I talking about?
What strengths do I have that I seem to underuse? Point to where
you have seen them.
Describe me the way a close friend of ten years might, then the way
a coworker who barely knows me might. Where do the two disagree?
What would the version of me from five years ago be surprised by?
If I could only take one piece of advice from you this year,
what should it be and why that one?
If these land, our self-discovery prompt list goes further in the same direction, with prompts built around patterns, envy, and the stories you tell about yourself.
Questions for a new account, or with Memory off
No history? Memory switched off, or not yet available in your region? Flip it around and let ChatGPT do the asking. You supply the evidence, so the read is about you rather than about a demographic.
Interview me. Ask me 10 questions, one at a time, that would tell
you the most about who I am. Do not summarise until all 10 are answered.
Ask me about three decisions I made this year. Then tell me what
they have in common that I might not have noticed.
I will describe a normal weekday hour by hour. Tell me what my
schedule says about my priorities.
Ask me what I would do with a free year and unlimited money. Then
ask me why I am not doing a smaller version of it now.
Here is a message I sent recently that I keep thinking about:
[paste it]. What does it reveal about how I handle conflict?
This format is also the fairest version of the whole exercise, because you can see exactly what the model was given. Nothing is coming from a memory you forgot about.
The more of the answer comes from what you said, the more of it is actually about you.
Before the scary ones: a quick privacy check
Asking a model to summarise your life means pulling a lot of personal detail into one place. A few things worth knowing, all from OpenAI's own documentation:
- Screenshots travel, sources do not. Memory sources are not included when you share a conversation through a shared link, but the answer text is. A "five brutal truths" reply can easily contain your health, your relationship or your employer. Read it before you post it.
- "Don't mention this again" is not delete. It reduces future references but leaves the original chat in place. To actually remove something, delete it from the Memory summary or saved memories, and delete the chat where you first shared it.
- Deleting a chat is not deleting a memory. A saved memory created from a chat survives that chat being deleted. You have to remove both.
- Turning off chat history reference schedules what was remembered from past chats for deletion within 30 days. Your saved memories stay unless you delete those too.
- Training is a separate switch. For personal accounts, OpenAI may use chats and remembered information to improve its models while Improve the model for everyone is on. That lives under Data controls.
If you only want to play with the fun questions and not leave a trace, a temporary chat is fine. Just remember it will also know nothing about you, which makes the fun questions a lot less fun.
Frequently asked questions
Why does ChatGPT know so much about me?
When Memory is on, it can use past chats, saved memories, custom instructions, Library files and connected apps, depending on your plan and region. Since the June 2026 memory update, it also builds a background summary of you from your conversations rather than only saving what you explicitly asked it to remember. You can review it under Settings, Personalization, Memory.
Why are my answers so generic?
Usually because it has little to go on: Memory is off, the feature has not reached your account yet, you are in a temporary chat, or your history is mostly one-off tasks. Either switch Memory on and give it time, or use the interview-style questions so you provide the material yourself.
Can I see everything ChatGPT remembers about me?
Not quite. The Memory summary is a high-level view, and OpenAI says it may leave out details that Memory can still reference. Asking "what do you remember about me, organised by category" and then comparing it with the summary is the most complete picture you can get from inside the app.
Is it safe to ask ChatGPT scary questions about yourself?
For curiosity and reflection, generally yes, as long as you treat the answers as guesses to check rather than verdicts. It is not a therapist and it can be confidently wrong in either direction. If a question stirs up something heavy, talk to a person you trust or a professional rather than asking the chatbot for more.
How do I make ChatGPT forget something about me?
Delete it from the Memory summary or your saved memories, and delete the chat where you first shared it, because removing one does not remove the other. "Don't mention this again" only stops it coming up; it does not delete the source. OpenAI notes deleted saved memories may be kept in logs for up to 30 days for safety and debugging.
Does a temporary chat remember what I ask?
Temporary chats do not create or update memories. Before you start one you can pick Unpersonalized so it also ignores your existing memories and custom instructions. If you later save the temporary chat, it becomes a normal chat and follows your usual memory settings.
The honest bottom line: ChatGPT cannot know you better than the material it has been given. What it can do is notice patterns across months of conversations that you never lined up side by side. Ask for the evidence, check it against the summary, and the scary questions stop being a party trick and start being useful.
We spend our days getting AI systems to answer from real data instead of confident guesses. If you want that kind of rigour working inside your business, book a free call and we will show you where it pays off.
Image credits, all released under CC0 1.0 and hosted on Wikimedia Commons: "Man on a smartphone" by Gilles Lambert, "Mirror" by Enrico Carcasci, and "Working Late" by Muhammad Raufan Yusup.