The Self-Discovery Profile: A Mirror, Not a Horoscope

This one asks the model to build a profile of you from everything it's seen in your conversations: how you communicate, what you seem to value, how you go at problems, and what you appear to be working toward. It's the first prompt in this thread, and the one the others build on.

You've probably seen the original going around. When chat memory became a common feature, "tell me what you know about me" prompts spread fast, and most of them produced something that read like a very flattering horoscope. This is the version I've kept, rewritten so it separates what you actually said from what it's inferring.

The prompt

Using what you actually know from our conversations, build me a profile of who I am. Cover my communication style, what I seem to value, how I approach problems, and what I appear to be working toward.

Ground every point in something real: quote or point to the specific thing I said or did that led you there. When you're inferring rather than observing, say so plainly, and rate how confident you are. Don't invent traits to round out the picture, and don't flatter me. If the evidence is thin somewhere, say the evidence is thin. I want an accurate read, not a satisfying one.

How to use it

Run it in a chat with real history, or in a tool where memory or past-chat access is turned on. The more it has to go on, the better. In a brand-new chat with nothing in it, there's nothing to profile, and a model asked to describe you anyway will fill the space with generalities.

When the answer comes back, read the evidence it cites before you read the conclusions. If a point rests on one offhand comment you made months ago, weigh it that way.

Heads up

Memory features store things. Before you lean on them for this kind of prompt, it's worth knowing what you've handed over, and keeping secrets out of prompts covers the habits that make that less of a worry.

What it's good at, and where it isn't

The ungrounded version of this prompt fails in a predictable way. Vague, flattering statements like "you value growth but sometimes doubt yourself" feel personally accurate to almost anyone. Psychologists call that the Barnum effect, and it's exactly the kind of output a model drifts toward when you ask it to describe you and don't ask for receipts.

That's what the second paragraph of the prompt is for. Making it quote or point to the specific thing you said turns a vibe into a claim you can check. Asking it to say when it's inferring, and to say when the evidence is thin, gives you a map of where the read is solid and where it's guesswork. The line "I want an accurate read, not a satisfying one" matters too. Models lean toward telling people what they'd like to hear, and naming that up front pushes against it.

The limits are real, though:

• It only sees what you've typed. If you mostly use a model for work questions, the profile is a profile of you at work. That's a slice, not the whole person.
• The confidence ratings are rough. "High confidence" from a model is a gesture, not a calibrated number.
• It can still be wrong with receipts. A quote pulled out of context is still evidence of something, just not always of what the model thinks.

Tip

Push back on the points that feel off and ask what would change its mind. The back-and-forth is often more useful than the first draft.

When you've got a profile you mostly agree with, the natural next step is the Contradiction Scanner, which looks for the places where what you say and what you do don't line up.