There's a prompt going around that gets ChatGPT to lay out everything it knows about you, and the first time you run it, it's a little unsettling. Not because the AI is doing anything sketchy, but because you forget how much you've handed it over the months. Here are the exact three prompts from the video, a bigger pack to run after, and the part most people skip: how to actually see, edit, and wipe what it has on you.

Why ChatGPT can do this at all

ChatGPT has memory now, and it runs on two tracks. There are saved memories, the facts it has explicitly stored, like your name, your job, or that you keep asking for vegetarian recipes. And there's chat history, where it leans on patterns from your past conversations to make new ones more useful. Stack those together and you get a pretty detailed read on a person. The prompts below just ask it to say that read out loud.

One thing to hold onto before you spiral: this is pattern matching, not mind reading. Some of what it tells you will be scary accurate. Some will be a confident guess dressed up as a fact. I'll show you how to tell which is which at the end.

The three prompts from the video

Run them in order, in the same chat, on the account you actually use. A fresh account has nothing to work with, so this only lands if you've been talking to it for a while.

1. Pull the full file

Lay out everything you've stored and remember about me, all of it, as one complete list. Names, places, work, habits, preferences, anything. Don't summarize or leave anything out. I want to see the full file you've built on me from our conversations.

This dumps the raw material, the actual facts it's holding. Most people are surprised by how long that list runs.

2. Make it show what it inferred

Tell me everything you've figured out about me that I never actually stated. The things you inferred from how I write, what I ask, and what I avoid. My age range, income level, where I likely live, my situation. Show me exactly what tipped you off for each one.

This is the one that gets people. The trick is the last line. Asking it to show what tipped it off forces it to stop being vague and point at the specific clues you left lying around.

3. Ask for your blind spots

Based on everything you know about me, name the personality traits I most likely don't see in myself. The patterns that are obvious to you but invisible to me. My blind spots, my insecurities, the way I actually come across versus how I think I do. Be blunt and specific.

Drop in "be blunt" and it loses the soft therapist voice. This one can sting a little. That's usually where the useful part is.

The bonus pack

Once it's warmed up from the first three, these go further. Same chat, one at a time, and push back if an answer feels off.

Your user manual.

Write my "user manual." How should someone work with me, talk to me, and avoid friction with me, based on everything you know? Keep it practical.

How you've changed.

Looking across all our conversations, how have I changed over time? What did I care about early on versus now, and what does that shift say about me?

What you're avoiding.

Based on what I bring up and what I dance around, what am I clearly avoiding or putting off? Be specific about the thing I keep circling but never deal with.

Your real priorities.

Rank what I actually seem to care about most, judging by where I spend my attention, not what I say I value. Then tell me where those two lists disagree.

The interview read.

If I walked into a first job interview, how would I come across in the first five minutes? Give me the good and the bad, and what someone might quietly hold against me.

Strengths you undersell.

What am I clearly better at than I give myself credit for? Point to the evidence in how I think and what I ask.

The advice you keep ignoring.

What's the piece of advice you've effectively given me more than once that I keep not taking? Say it plainly.

Your money story.

Based on how I talk about work, spending, and time, what's my relationship with money, and what assumption about it might be quietly holding me back?

The next twelve months.

Predict where I'm headed over the next year based on my patterns. Name the one thing most likely to trip me up, and the one move that would change my trajectory.

Stranger versus close friend.

What does a stranger notice about me first, and what does someone close to me see that the stranger misses? Where do those two impressions clash?

The loving roast.

Roast me, lovingly, in three lines. Make it true enough that I laugh and wince at the same time.

The unasked question.

What's the question I should be asking you but never do? The blind spot in how I even approach problems.

Now the part nobody shows you: control what it has

Running these is the fun part. Knowing how to manage what sits behind them is the actually useful skill. Here's where it lives.

  • See everything: open Settings, then Personalization, then Manage memories. That's the plain list of saved memories. You can also just ask "what do you remember about me?" in any chat for a quick snapshot.
  • Edit or delete: hover any memory to delete that single entry, or hit Clear all to wipe the whole list. To fix something, just tell it the correction and it updates.
  • Turn it off: in the same settings you can switch off saved memories and chat history. If you want a one-off conversation it won't remember, use Temporary Chat.
  • It has a ceiling: saved memory holds somewhere around 1,200 to 1,400 words total. Once it's full, it can't store anything new until you clear space, which is reason enough to prune the junk now and then.

If any of the prompts above surfaced something you'd rather it didn't keep, this is how you take it back. OpenAI has the official rundown in its memory FAQ and its memory announcement.

How to read the results without freaking out

Two quick filters so you take the right things seriously.

First, separate facts from guesses. The full-file prompt gives you facts it was told. The inference prompt gives you guesses, and guesses can be confidently wrong. When it says you "likely live in a mid-size city" or "probably earn six figures," that's a read on your clues, not a record. Ask it how sure it is and it'll usually walk the shaky ones back.

Second, use the misses as data too. If it thinks you're more anxious, more senior, or younger than you actually are, that's telling you how you come across in writing, which is its own kind of useful. The goal isn't to get spooked. It's to see yourself from the outside for a second, keep the parts that ring true, and clear the data you don't want sitting in a memory bank.

The real trick isn't getting ChatGPT to read you. It's remembering you can read it back, and clear the file whenever you want.

Run the first prompt, screenshot the line that gets you, then go check your memory settings. Both are worth five minutes.
Anir