ChatGPT memory: what it remembers and how to control it

ChatGPT memory lets the assistant remember details from your chats and use them in later answers, so you do not have to repeat your job, your preferences or your projects. On October 23, 2026 it is more than a list of saved facts: depending on your plan and region, it can draw on past chats, saved memories, custom instructions, files in your Library and connected apps such as Gmail. This guide explains what it uses, how to check and turn it off, how to delete something for good and how Claude and Gemini handle the same thing, based on each company's help pages read on October 23, 2026.

What ChatGPT memory can use

OpenAI lists five possible sources, depending on plan and region:

• Past chats, when Reference chat history is on.
• Saved memories, details you ask it to remember or that it saves as useful context.
• Custom instructions, the guidance you write yourself in Personalization.
• Files in Library.
• Content from connected apps, such as Gmail.

It does not keep every detail of every conversation, and it does not search your history for every answer: ChatGPT looks for relevant context when it is likely to improve the response, and memory can also shape the queries it uses for a web search, for example your city or a dietary preference. Memory is not the same as the context window: the context window is what the model reads in the current chat, and memory decides what extra context gets brought into it.

Saved memories, chat history and the summary

Saved memories are explicit and stored separately from your chats. Deleting the chat where something came up does not delete a saved memory created from it.

Reference chat history lets ChatGPT use relevant information from all your past conversations. It has no separate storage limit, and what it derives can change over time. If you turn it off, what was remembered from past chats is scheduled for deletion from OpenAI's systems within 30 days, while the chats stay in your history. Turning off Reference saved memories also turns off Reference chat history.

The memory summary is a high-level view of what ChatGPT may use to personalize answers. OpenAI warns that it does not necessarily include every detail or source, so it is a good place to start, not a complete record. Where available, you can switch between the current improved memory and the older list of individual saved memories.

How to check it and turn it off

Go to Settings, Personalization, Memory. Depending on your account you will see controls such as Memory, Reference saved memories, Reference chat history, Memory summary and Manage. From there you can:

• Correct the summary or a saved memory, or tell ChatGPT in the chat that something is out of date.
• Use Don't mention this again, which reduces future references but does not delete the original chat, file or email.
• Use Delete and turn off memory, which deletes what the summary shows and turns memory off, without deleting your chats.
• Check Sources, which may appear under an answer and show which past chat, memory, file or email contributed. They may not show every factor.

For a single private conversation, start a temporary chat: it never creates or updates memories, and if you choose Unpersonalized before it starts it will not use your existing memories or custom instructions either. You cannot change that choice once the chat has started, and saving a temporary chat turns it into a regular one.

How to delete something for good

Turning memory off does not delete anything, and deleting one thing rarely deletes all its copies. To remove a piece of information completely, OpenAI says to delete it from every place it lives: the memory summary, saved memories, regular and archived chats, files in Library and connected apps. Deleting a saved memory stops it being used, but it stays in the past conversations where you mentioned it. Changes can take time to take effect, and OpenAI may keep logs of deleted saved memories for up to 30 days for safety and debugging.

Memory also matters for training. On personal accounts, OpenAI may use chats and remembered information to improve its models when Improve the model for everyone is on, a setting you can turn off in Data controls. By default it does not train on content from ChatGPT Business, Enterprise, Edu or ChatGPT for Healthcare. Our guide on whether ChatGPT is safe covers the rest of the data settings.

How Claude and Gemini compare

Table: memory in ChatGPT, Claude and Gemini (official help pages, checked October 23, 2026)

Two details stand out. Anthropic says Claude does not store sensitive topics such as health, religion or politics by default, unless you turn on Include sensitive topics in memory, and each project in Claude has its own memory. Google says that to delete something Gemini learned from a connected app you must both delete the chats and disconnect the app, because doing only one leaves a copy it can still use. Our comparisons of Claude and ChatGPT and ChatGPT and Gemini cover the rest of the differences.

Our advice

Leave memory on if you use ChatGPT every day for the same work, and read the memory summary once a month. Keep health, finance and client details out of it, or use a temporary chat for them. And before turning memory on in a shared or work account, check what your workspace admin allows, because they control the available settings.