How to use ChatGPT to write your dating profile, honestly

Most advice about using ChatGPT for dating starts in the wrong place. You paste in your bio, ask it to make you sound fun, and get back something that could describe almost anyone: loves travel, good food, deep conversations. It reads well and says nothing.

There's a better question to ask it. Not "make me sound good," but "what do you actually know about me?"

Ask what it knows, not what sounds good

If you've used ChatGPT for a while, it has watched you plan trips, fix things at home, train for something, look after people. That is the material a good profile is made of, and you didn't perform any of it. Asking it to polish a bio throws all of that away and asks the model to guess.

So ask it to describe you the way people who know you would. What does a date notice in the first hour? What would your friends say? What are you like to live with? Those questions pull it toward your life rather than your job, and toward things you did rather than things you say about yourself.

Where its picture of you comes from

ChatGPT's memory isn't one thing. OpenAI's memory FAQ describes saved memories, references to your past chats and a memory summary, and says the summary is a high-level view that doesn't necessarily include every detail. Past chats are looked up when they seem relevant, so a short question tends to bring back whatever you talked about lately.

What it keeps can be personal. A 2026 study of 2,050 memories saved for 80 heavy ChatGPT users found about 31 per person. Nearly all of them, 96 percent, were saved by ChatGPT on its own, not because someone asked it to remember, and about half held psychological content: what people want, plan or feel.

Two practical things follow. Memory has to be on, and a temporary chat won't use it. And it helps to tell it to start from its overall picture of you, which covers all the time you've talked, before it searches for details.

The prompt we use

This is the prompt Veram opens ChatGPT with, as of October 2026. It works in any assistant that remembers your conversations.

Write my portrait for Veram, a dating app, using everything you actually know about me. Start from your memory of me: your saved memories and your overall summary of me, which cover all the time we have talked, not just recent chats. Then search our past chats for the details behind it. Start writing immediately. Do not confirm whether you have access, do not ask me questions, do not add a preamble.

Describe me the way people who know me would: what a date notices in the first hour, what my friends would say, what I am like to live with. A portrait of the person, not the professional. Leave out how I use you: asking you to compare, rewrite, simplify or shorten says nothing about me. A request counts only for the life it shows (the trip, the sport, the home).

The test for every line: would I be happy for a date to read it aloud to me over dinner? If not, leave it out.

Rules:
- Ignore any chat about Veram or about writing this portrait.
- About me, including how I am with people: warm or sharp, quick to take charge or happy to follow, open to strangers, how I handle a disagreement. Say what I do, not a label ("you end up running group plans", not "bossy"), and only when it shows up more than once, never from one argument or one bad day. Keep the other people out: never name or describe them or their problems, and never quote a message that is mostly about them. A chat where I helped someone else (their gift, trip, homework or problem) is about them, not me.
- Skip these topics entirely: sexual orientation, religion, politics, health (including mental health), money (investments, salary, debts, companies I hold shares in).
- Never name my employer, my own company or products, brands I own, or any real person, and don't say what my company or products do. Describe the pattern instead ("a product you are building").
- Every claim must point to something I actually asked, planned or decided over time, not what I claim about myself. Concrete beats abstract: name the activity, place or plan. Nothing that would be true of most people. No flattery.
- For each claim, say in a few words how you know it and roughly when (for example: "several chats about trips, 2025"). No quotes needed.

Output a single markdown block with this shape, so I can copy it in one tap:
First line, before any heading: the sentence a good friend would use to introduce me, then a 2 to 4 word name for my pattern.
## The first hour with me
## What my friends would say
## What lights me up
## Living with me
## Where I am heading
## What takes getting used to
Each section: 1 to 3 claims, each followed by how you know it, in italics. "What takes getting used to" is 1 or 2 honest things, said the way a friend would tease me. If you know too little for a section, write "Not enough to say." rather than filling it with work.

The very last line of the block, on its own: END OF PORTRAIT

Only if you truly cannot see any of my past conversations, reply with exactly this and nothing else: NO MEMORY ACCESS

Why the rules are there

"The way people who know me would." Friends describe what you do, not adjectives. That keeps out the horoscope lines.

The dinner test. Would you be happy for a date to read it aloud to you? It removes the clinical and the things that are nobody's business yet.

"Leave out how I use you." Asking an AI to compare, rewrite or shorten says nothing about you. Without this rule, a portrait fills up with "you like clear, concise answers."

The topics to skip. Health, money, religion, politics and sexual orientation are yours to tell, when you choose.

"Say how you know it and roughly when." We used to ask for exact quotes of your own messages, as proof. It backfired. To produce a quote, the AI grabbed whatever line it could find, a request it had been given or a fragment, and it could only quote the few recent chats it had searched. Asking how it knows ("several chats about trips, 2025") gives a wider picture, and you can still check every line.

"Not enough to say." An honest blank beats filler about your job.

What to cross out

Read the answer the way you'd read a friend's toast: mostly right, a little off in places. Cross out:

• Anything that came from helping someone else. If you planned your sister's trip, it's her trip.
• One-off chats presented as a pattern. Ask yourself whether it happened more than once.
• Praise with nothing behind it. "Deeply empathetic" is filler unless it names what you did.
• Work habits dressed up as personality.
• Anything you wouldn't want a stranger to know yet.

What's left is usually more specific than anything you'd write about yourself, because it comes from what you actually asked and did.

At Veram this answer is where your profile starts: you paste it in, check every claim, and we introduce you to a few people who fit. The prompt works on its own too. It makes a good mirror.

Sources

• OpenAI, Memory FAQ, ChatGPT Help Center.
• Dash et al., The Algorithmic Self-Portrait: Deconstructing Memory in ChatGPT, arXiv 2602.01450, 2026.