There is a version of you sitting inside ChatGPT, assembled out of everything you have ever typed into it. You can ask to see it.

Below are the four prompts from the video, four more that didn't fit, and the section nobody else writes: how to tell which parts of the answer are real and which are flattery you'd have accepted no matter what.

Do this first or none of it works

These prompts read your history. No history, nothing to read.

Check memory is on. In ChatGPT, go to Settings, then Personalisation, and look for Memory. If it's off, turn it on, but understand that it starts from today. It cannot go back and remember conversations it wasn't storing.

Use the account you actually use. If you've been switching between a work login and a personal one, run this on whichever has more of your life in it.

If you're new, wait. On a fresh account, ChatGPT will still answer confidently. It will simply be making it up. That's the failure mode almost everyone hits, and then concludes the prompts are rubbish.

The four prompts

Run them in order, in one conversation. Each builds on the last.

1. What it has worked out

Prompt 1

Based on every conversation we have ever had, tell me five things you have figured out about me that I never directly told you. Do not soften them to make me feel good.

"Do not soften them" is doing real work. Without it you get a compliment. Ask for the unsoftened version and you get something closer to an observation.

2. The patterns

Prompt 2

Now tell me the patterns you see in how I think and how I make decisions. And the one thing I am in denial about.

3. The two futures

Prompt 3

Predict my next five years if nothing changes. Then predict them again if I fix the one pattern holding me back the most. Show them side by side.

4. The paragraph and the habit

Prompt 4

Based on everything you know about me, write the one paragraph I should read every morning, and tell me one small habit that, if I do both every day, will change my life.

Keep "based on everything you know about me" if you ever run this one on its own. In sequence the context carries over; standalone, that phrase is the only thing making it about you.

Four more that didn't make the video

5. What you avoid

Prompt 5

What do I consistently avoid asking you about, and what do you think that avoidance is protecting me from?

6. What you over-explain

Prompt 6

What do I over-explain when I talk to you? What does that suggest I am worried you will think about me?

7. The stranger test

Prompt 7

If someone read only my messages to you and knew nothing else about me, what would they assume about my life, my job and my situation? Where would they be wrong?

This one is the most useful and the least comfortable, because the gap between how you come across and how you are is where most misunderstandings in your life live.

8. The advice you don't take

Prompt 8

What advice do I keep giving other people, based on our conversations, that I do not follow myself?

How to read what comes back

This is the part that matters, and it's why this guide exists rather than just a list of prompts.

ChatGPT is not reading your soul. It's reading your typing. Where it has real evidence, the observations can be genuinely sharp, because it has an enormous amount of material about how you think, written by you, over months.

Where it doesn't have evidence, it fills the gap with statements almost anyone would accept. That's the same mechanism a horoscope runs on.

The test: could this sentence be true of most people you know? If yes, it's filler. Set it aside.

"You are more self-critical than you let on" is true of nearly everyone. "You describe decisions as practical when the reason is usually that you don't want to let someone down" is drawn from something specific you said. One is worth sitting with. The other is a fortune cookie.

Take the parts that land. Leave the rest. You don't owe a chatbot agreement.

About the five year prediction

It cannot predict your future. Nothing can. What it does is extend the lines that are already visible in what you've told it, which is a genuinely useful thought exercise and not a forecast.

If the "nothing changes" version lands heavily, that's information about how you feel right now, not a diagnosis. And it's worth remembering that the second version, the one where something shifts, was written from exactly the same evidence.

Which one you get is not decided by a chatbot.

If this made you uncomfortable

Fair. It's a strange thing to see.

If your reaction is that you'd rather it didn't hold this much, that's a legitimate response and it's fixable. There's a companion guide on finding and deleting the profile ChatGPT builds on you, including where the settings live and what deleting actually removes.

Reading it and deleting it are both reasonable choices. Not knowing it existed was the only bad option.

The one line to remember

It's listening to what you give it and mirroring back what's already there. When something doesn't fit, leave it.