ChatGPT Now Remembers You. Have You Checked What It Knows?

On June 4, OpenAI changed how ChatGPT remembers you. Not with a new chatbot. With a feature most people will never open.

For two years, ChatGPT only kept what you told it to keep. You would say "remember I'm vegetarian," and it wrote that down like a coworker taking notes, then forgot everything you didn't dictate. The new system, which OpenAI calls dreaming, runs in the background. It reads across your past conversations and decides on its own what is worth remembering.

That is a real shift. The tool is no longer waiting for instructions. It is forming a picture of you whether you ask it to or not.

Let me slow down here.

For most people this is a quiet upgrade. You stop re-explaining yourself. Past context carries forward. You ask for something compatible with "my setup" and it already knows what your setup is. Less friction, better answers.

But there is a part worth thinking about if you run a business.

When a tool decides for itself what to remember, you lose track of what it thinks it knows. And what it thinks it knows shapes every answer it gives you. A wrong assumption it picked up six months ago, in a conversation you have long forgotten, can quietly steer advice you are treating as fresh.

OpenAI built a fix for exactly this, and almost nobody will use it. There is now a memory summary page. It shows you, in plain language, what ChatGPT has decided about you. You can read it, correct it, delete what is wrong, and tell it which topics to raise or leave alone.

So do this one thing this week. Open it. In ChatGPT, that is Settings, then Personalization, then Memory. Read what your most-used AI tool believes about you and your work. You will probably find a few things that are out of date, and maybe one that was never true.

That last one matters more than it sounds. An assumption you never made, baked into answers you trust.

Now the harder question for business owners.

Most people run one ChatGPT account for everything. Personal questions, work questions, half-formed thoughts about clients and pricing and strategy. That was fine when memory was a short list you controlled. It is different now that the tool is folding all of it into a standing profile.

Think about what that profile holds if your business runs through your personal account. Client names. Deal terms you were working out. The competitor that keeps you up at night. None of it secret, exactly, but all of it now part of a persistent picture the tool carries into every session, including the ones where you are just asking what to make for dinner.

I am not telling you to panic. The controls are real and the feature is genuinely useful. I am telling you to be deliberate about something that used to happen on its own.

The practical version is simple. Decide what you want your AI tool to remember about your business, and what you would rather keep out of it. If you have been running everything through one login, this is a fair moment to separate the work that carries sensitive context from the work that does not. A business account, or at least a clear rule about what you discuss where.

There is a bigger pattern here, and it is the part I would watch.

AI tools are moving from things you operate to things that know you. A tool that remembers is more useful and harder to leave. The context you have built, the profile it has assembled, the answers that finally feel tailored to you. That is real value. It is also what makes walking away cost more than it did a year ago.

That is the trade to hold in your head. Every piece of memory that makes the tool fit you better also ties you to it a little tighter. Neither half of that is bad. But you want to be the one choosing how far to lean in, not drifting into it because the default chose for you.

The old ChatGPT forgot you the moment you closed the tab. The new one is quietly taking notes on who you are. The only question is whether you have ever looked at what it wrote down.