ChatGPT Memory Now Builds a Profile of You: Why I Don’t Mind

Last updated July 22, 2026

I use ChatGPT daily as a chatbot while Claude Code handles….code. When I saw the headlines about ChatGPT quietly building a profile of me, I didn’t panic. If anything, I was a little relieved.

What wears me down is repeating myself. I re-explain the same stuff to every fresh chat, from how I like things written to the projects I’m juggling to the basics of who I am. Even in a tool like Claude Code, there’s a main setup file it always reads, and each new project starts in its own little bubble that other sessions don’t seem to know about. Open a fresh one and I’m half back to square one.

So a memory that carries across every chat sounds good to me. I have one line I won’t cross, and it comes down to what they do with the data.

What Changed With ChatGPT Memory

ChatGPT memory used to be a simple list of notes you asked it to save. Now it runs a background process, which OpenAI calls Dreaming, that reads across your past conversations and keeps a running, self-updating picture of you.

The old version was easy to picture. You told ChatGPT to remember something, it wrote that down, and you could go read the list. The new system does the remembering on its own. It decides what’s worth keeping from your chats, including things it figured out about you that you never spelled out. Then it revises those notes over time as your life changes.

OpenAI’s own example is a good one. A memory that says you’re going to Singapore in July rewrites itself to say you went to Singapore in July 2026 once the trip is over, without you touching a thing. The company says this made time-sensitive memories far more accurate than the old setup.

The upgrade is rolling out in stages. It reached ChatGPT Plus and Pro subscribers in the US first. Free, Go, and international users are following over the coming weeks. Depending on your plan and where you live, you may already have it, or it may show up soon.

Why It Now Builds a Profile of You

The upgrade turns memory into an evolving profile of you, and OpenAI concedes the summary page you can open won’t show everything ChatGPT remembers.

This is the part that made people uneasy, and I get why. When memory was a list you controlled, you could open it and see exactly what ChatGPT had on you. Now that it works in the background and infers things on its own, the record is harder to audit.

OpenAI says so directly. Its help page notes the memory summary “will not include everything that ChatGPT remembers based on your chats.” Ask a separate page whether the summary shows all your memories, and the answer is, plainly, “Not necessarily.” Some details, it says, get left out when the system decides they’re less relevant or not appropriate to show.

Why I’m Fine With It

I’m fine with ChatGPT memory, and I’ll go further. I’m a proponent of it, because the payoff is real. I stop re-explaining myself and the tool gets more useful the more it knows.

A profile sounds sinister until you notice you already want the results it produces. I want the assistant that remembers I write in a plain, no-fluff voice when I ask it to cleanup an email I’m sending out. I want it to know the projects I’ve got open so I’m not laying out the backstory every single time. The whole reason I lean on these tools is to save time, and re-teaching them my basics every morning is the opposite of that.

Thinking about this has me going back to my Claude Code setup. It’s genuinely handy that each project can carry its own context, and it’s genuinely annoying that the context stops at the project line. A memory that follows me everywhere fixes the annoying half. If a company wants to study how I use the product so they can hand me a smoother version of it, I’m not bothered. That’s them building a better tool.

I’ll admit I might be a little unusual on this. A lot of people hear “it’s building a profile” and feel watched. I’ve made a kind of peace with the digital trade-off. My information has been floating around for years, and realistically some of it is sitting on the dark web right now. You take the sensible precautions you can take, and then you accept there’s a ceiling on how much control any of us really has.

My One Hard Line: Don’t Sell My Data

My support has a limit. Use my profile to improve the product and I’m on board. Sell it, or hand it to someone whose business is targeting me, and I’m out.

There’s a real difference between a company using what it knows about me to make its own product better and a company packaging that up to sell. The first one helps me. The second one turns me into the product, and that’s the one I won’t defend.

On paper, OpenAI’s stated policy lands closer to the version I can live with. The company says it may use your content, including your memories, to improve its models. It also gives you a switch to turn that off under Data Controls, and content from Team and Enterprise accounts is excluded from training altogether. By default your chats can help train the model unless you turn that off, which is worth knowing. Selling your data to outside advertisers is a different thing, and that’s the part I’d actually object to.

That’s the line I watch. As long as these profiles stay in service of a better product and don’t get sold down the road, I’m comfortable. The day that changes is the day I change my settings.

How to See and Control Your ChatGPT Memory

You’re not stuck with it. In Settings then Personalization, you can view your ChatGPT memory, edit it, delete pieces, clear all of it, or switch it off completely.

If any of this makes you want to look under the hood, it takes about five minutes. The same way you can turn off Google AI Mode when you’d rather have plain search results, you can dial ChatGPT memory up or down. Here is where the controls live.

• See what it has: open Settings, then Personalization, then Manage memories. The memory summary shows the high-level picture and notes when it was last updated.
• Fix something wrong: type a correction into the summary box, or highlight a line and edit it directly.
• Delete one memory or wipe the lot: use Manage memories to remove a single entry, or Clear all to empty the whole list.
• Turn memory off: Settings, then Personalization, then Memory, then toggle it off. ChatGPT stops creating and using memories while it’s off.
• Keep memories but stop it using them: switch off “Reference saved memories” if you’d rather store them without pulling them into answers.
• Stop training on your chats: that’s a separate switch under Data Controls, so turning memory off doesn’t automatically opt you out of model training.

One catch worth knowing. Deleting a chat does not delete the memory it created. To fully remove something ChatGPT knows, OpenAI says you have to clear it from every place it lives, including past chats, archived chats, uploaded files, the memory summary, and any connected apps.

What ChatGPT Memory Still Gets Wrong

Memory is convenient, not flawless. It can hold on to stale or mistaken conclusions about you, and you can’t fully see what it’s keeping.

Because the system now infers things on its own, it can get you wrong. It might decide you prefer something you mentioned once and didn’t mean, then carry that forward. You won’t always catch it, since the summary doesn’t show everything. The fix is the same either way, which is to check the summary now and then and correct what’s off.

It also isn’t a privacy setting by default. This is the same trade-off I weighed with the new iOS 27 Siri. If you share a computer, or you type something into ChatGPT you’d rather it forget, remember it may quietly fold that into your profile unless memory is off. A simple habit is to open a temporary chat for anything you don’t want kept.

Memory is one of those AI shifts that sounds worse in a headline than it feels in practice, at least for how I use these tools. Open your settings, see what ChatGPT has on you, and decide where your own line sits.

Related reading: What GPT-Live changes for ChatGPT voice | How to turn off Google AI Mode | The iOS 27 Siri privacy trade-off | New to AI? Start here