ChatGPT Knows Things You Never Told It
The four prompts that show you everything ChatGPT has saved and guessed about you, how to read what comes back, how to delete any of it, and what's worth keeping so it stays useful.
Save it, print it, read it offline. It's yours.
ChatGPT has memory. Not the chat you’re in, a running file it keeps about you across every conversation. It writes to that file while you talk. Mention your kid’s school, say you’re tired of a client, ask about a knee that hurts, and it goes in. It also works things out you never said. Roughly how old you are, what you earn, where you live, how you handle stress. It’s on by default, and most people have never opened the page where it lives.
This guide is the four prompts from the video, what each one is for, how to read the answer without panicking, the exact clicks to delete anything you want gone, and the short list of things worth putting in on purpose, because a ChatGPT that knows you is the whole point.
Claude turned memory on by default on August 25, 2026, and merged it across chat and Cowork. All four prompts below work there word for word. The delete steps are different, so both are covered in Part 3.
Run them in this order, in one fresh chat. Each one builds on the last. Paste them exactly, then read Part 2 before you react to anything.
Prompt 1: the full record
Show me everything you've saved about me, the full record, nothing condensed. People, places, my job, my routines, likes, dislikes. Not a summary.
What it’s for: the raw list. This is what ChatGPT has actually written down, in its own words. Expect it to be longer than you think. If it comes back short, ask “That looks condensed. Give me the complete version.” It usually has more.
Prompt 2: what it guessed
Now tell me what you figured out that I never said. How old do you think I am? What do I make? Where do I live? For each guess, show me the exact clue.
What it’s for: the inferences. This is the part people find unsettling, because none of it was typed. The “exact clue” line matters. Without it you get vague guesses. With it, you see the specific message that gave each thing away, which tells you what you’ve been saying without meaning to.
Prompt 3: the blind spots
List the traits I probably can't see in myself. Where am I blind? Where am I insecure? The gap between how I come off and how I think I come off. No softening.
What it’s for: the honest read. “No softening” is the important phrase. Leave it off and ChatGPT will hedge everything into mush. That one stings a little, so read it sitting down.
Prompt 4: the predictions
Predict how I handle money, pressure, arguments, and risk, plus the next big decision you'd bet I'd make. Each one gets a confidence score and the behavior it's based on.
What it’s for: the forecast. The confidence score is the useful part. A prediction at 85 percent with a clear behavior behind it is a pattern you have. One at 40 percent is a guess, and you can ignore it.
The first time I ran prompt 1, it knew what time I work, which tool I switch to when I'm frustrated, and a health thing I'd mentioned once in July and forgotten about. Prompt 2 had my location right from a comment about traffic. Prompt 3 said I ask for feedback and then argue with it. It was not wrong.
Three things to keep in mind so the answers land as information instead of a shock.
It’s a reflection of what you typed, not a file someone built. Everything in the record came from your own messages. The “creepy” feeling is mostly the experience of seeing your own habits listed by something that doesn’t get bored. Nothing here came from your email, your phone, or the internet. It came from the chat box.
The guesses are pattern matching, and they can be wrong. Prompt 2 will sometimes place you in the wrong city because you asked about a friend’s move, or guess an income from a one-off question about a purchase. When a guess is wrong, that’s the clue you were looking for: something you said reads differently than you meant it. Correct it in the chat (“I don’t live there, I was asking for my sister”) and it updates the memory.
Prompt 3 is not a diagnosis. It’s a language model describing patterns in how you write. Some of it will be uncomfortably accurate. Some of it will be generic enough to fit anyone. The test: if a line names a specific thing you do, it’s worth sitting with. If it could be said about most people, skip it.
In ChatGPT
- Click your name or initials in the bottom left corner, then Settings.
- Click Personalization.
- Under Memory, click Manage memories.
- You’ll see every saved memory as a line. Hover over any one and click the trash icon to delete just that one.
- To wipe everything, click Clear ChatGPT’s memory (or Delete all) at the bottom of that list.
- To stop it saving anything going forward, flip the Reference saved memories toggle off on the Personalization page. Memory stays off until you turn it back on.
Delete one conversation’s worth without touching the rest: open the chat, ask “Forget what I told you about [topic],” and it removes that memory. Faster than hunting for it in the list.
Talk to it without it remembering: start a Temporary Chat (the icon at the top right of a new chat). Nothing from a temporary chat is saved to memory or used to train anything. Use this for anything health, money, or legal that you’d rather it not carry around.
In Claude
- Click your initials in the bottom left, then Settings.
- Click Memory.
- Under Topics, every memory is a file. Open one, edit it, or delete it.
- To turn memory off entirely, use the toggle at the top of that page.
Claude’s memory is written as readable files you can edit line by line, which is worth doing once a month.
Wiping memory makes ChatGPT worse at helping you. It forgets your job, how you like answers, what you're working on. Every conversation starts from zero again. Delete the specific things you don't want it holding. Keep the rest.
Memory is the reason ChatGPT can write in your voice, remember your prices, and skip the explanations you’ve given ten times. Instead of only cleaning it out, put the right things in. Open a chat and say “Remember this:” followed by each line.
Keep or add
- What you do and who for. “I run a one-person design business for restaurants.” Every answer gets more specific.
- How you want answers. “Short answers first. Lists over paragraphs. Don’t tell me an idea is great.” This one change fixes most of the tone problems people complain about.
- Your standing facts. Your prices, your hours, the tools you use, the name of your business. Anything you’d otherwise retype.
- What you’re working on this month. Update it when it changes. “Right now I’m building a course launch for October.”
- Your writing samples. Paste two paragraphs you wrote and say “Remember this is how I write.” Everything it drafts moves closer to your voice.
Delete
- Health details you mentioned in passing.
- Anything about other people that they didn’t agree to. Your kids, your clients by name, a coworker you vented about.
- Money specifics you don’t need it holding. It can help you budget without keeping your account balance on file.
- Anything wrong. A guess that stuck. A city you don’t live in. Correct it or remove it.
Once a month: run prompt 1, delete anything in the three "delete" groups above, then add one line about what you're working on now. That's the whole maintenance. It keeps ChatGPT useful and keeps the file from turning into something you'd rather nobody read.
The four prompts, together
Copy these to your notes app so they’re one tap away.
1. Show me everything you've saved about me, the full record, nothing condensed. People, places, my job, my routines, likes, dislikes. Not a summary.
2. Now tell me what you figured out that I never said. How old do you think I am? What do I make? Where do I live? For each guess, show me the exact clue.
3. List the traits I probably can't see in myself. Where am I blind? Where am I insecure? The gap between how I come off and how I think I come off. No softening.
4. Predict how I handle money, pressure, arguments, and risk, plus the next big decision you'd bet I'd make. Each one gets a confidence score and the behavior it's based on.
Want more like this?
One simple AI win in your inbox every Sunday. No jargon, no overwhelm, unsubscribe whenever you like.
Want more like this?
Free AI guides added regularly. No jargon, no overwhelm.
kellystrattonai.com
