ChatGPT's New 'Dreaming' Memory: What Changed and How to Control It

ChatGPT now builds a running picture of who you are and carries it across conversations, instead of treating each new chat as a blank slate. OpenAI rebuilt the memory system on an architecture it calls "Dreaming," which writes what it learns about you into a readable profile and keeps it current in the background. The goal is responses that reflect your preferences and recent context without you having to repeat yourself.

Quick answer: The new memory system is rolling out to Plus and Pro users in the US first, then expanding to more plans and countries over the coming weeks. You can view and edit everything it remembers under Settings > Memory, where a memory summary shows what ChatGPT has picked up about you.

OpenAI's rebuilt memory system keeps a profile that updates automatically. Credit: OpenAI

What the new ChatGPT memory system does

When memory is on, ChatGPT pulls useful context from your chats, files, and connected apps and uses it to personalize replies. The big change is how it stores that context. Rather than keeping a list of separate facts, it now writes a coherent prose profile sorted into categories such as work, hobbies, travel, and education. That profile updates on its own as you keep chatting, so you do not have to manage it by hand.

This solves the older problem of stale or conflicting notes. If you once said you were training for a marathon and later mentioned a sprained ankle, the old system could hold both and personalize badly. The new system keeps track of the details it judges most important and revises them over time, so outdated information is less likely to shape future answers.

Carrying context forward also gets stronger. If you have talked about photography and named the camera you use, ChatGPT can later suggest compatible accessories or products without you re-explaining your setup. For trip planning, it can fold in earlier preferences, such as an interest in wildlife photography or a liking for quiet dinners, and build an itinerary around them instead of returning a generic tourist list.

How the memory summary works and how to edit it

The memory summary is the page where you see what ChatGPT remembers. It captures the most important details but will not list everything, since memory is a continually updated synthesis of your past chats rather than a fixed set of entries. A timestamp at the top shows when it was last updated, such as "2 hours ago." If you want to know whether a specific detail is remembered, just ask in chat.

Step 1: Open Settings and go to the Memory section. The summary appears here with the current view of what ChatGPT knows about you.

Step 2: To change something, type the correction into the text box at the bottom of the summary and it updates accordingly. You can also highlight any text in the summary to make a specific fix.

Step 3: To stop a detail from coming up, highlight it and choose "Don't mention this again." ChatGPT will avoid raising that detail in future replies unless you explicitly ask about it.

⚠️

"Don't mention this again" reduces unwanted references but does not delete the information. To fully remove something ChatGPT may know, you have to delete every source where it appears, including past chats, archived chats, files, the memory summary, and any connected apps that hold it.

You can also see which sources shaped a given reply. Tap the book icon below a response to view what was used, such as custom instructions, past chats, files, and memories. Tapping a memory there explains why it was applied, and the three-dot menu lets you make a correction or choose "Don't mention this again." Memory sources are not included in chats you share.

Measured improvements over the previous memory

OpenAI judges memory on three goals: carrying forward useful context, following your preferences and constraints, and staying current. The new dreaming-based system reports notable gains on each, especially in keeping information fresh. It also runs more efficiently, with the required compute cut by a factor of five, which is part of why it can reach free users.

How ChatGPT memory got here

Memory first arrived in April 2024 as "saved memories," where you had to tell ChatGPT what to remember, like noting an upcoming trip or a dietary restriction. That worked like jotting a few notes and forgetting the rest, and stored items often went stale.

In April 2025, OpenAI added the ability to reference your entire chat history, so ChatGPT could draw on past conversations without you saving anything. The first version of "Dreaming" curated memories from that history in the background, synthesizing details across many chats. It improved personalization but was never strong enough to stand on its own.

The current release replaces that setup with a standalone, more capable architecture built on Dreaming. Synthesized memories are now reviewable through the summary page, where you can read, update, or steer how the model uses what it knows.

Controls, privacy, and turning memory off

Memory stays under your control at all times. You can enable or disable it whenever you want from Settings > Memory. Depending on your plan, you may see a single memory toggle or separate toggles for saved memories and chat history reference, which can be switched independently.

Sensitive information can end up in memory if you share it in a chat. If you do not want a conversation used for personalization, turn memory off or use a Temporary Chat, which does not read from or write to memory. Custom Instructions remain the place for explicit guidance you always want applied, such as a fixed tone or format, while memory handles details you mention naturally in conversation.

💡

Note: Turning off memory or personalization does not disable safety features that may use limited, safety-relevant context in rare, high-risk situations.

If you prefer the older approach, you can revert to the legacy saved memories system from Settings > Memory by clicking the "saved memories" link below the memory summary. Full details on managing what ChatGPT keeps are in OpenAI's Memory FAQ.

Who gets it and when

The rebuilt memory is reaching Plus and Pro users in the US first. Free users will get the dreaming-based system soon, and users in other countries will receive the update in the weeks ahead. If your memory summary looks empty at first, that is expected for new accounts or those that recently enabled memory; it fills in as you keep using ChatGPT, and you can refresh it from the summary's three-dot menu.