How ChatGPT and Claude Memory Works and How to Control It

ChatGPT memory works in two ways: saved memories (facts you ask it to remember, or that it saves as useful) and reference chat history (it draws on relevant past conversations). You control both under Settings > Personalization > Memory, and a Temporary Chat never creates or updates memories. Claude’s memory is similar but organized as editable “topics” under Settings > Memory, with a separate memory for each project and incognito chats that leave no trace.

This guide explains how ChatGPT memory and Claude memory actually work in October 2026, what each plan gets, how to view, edit, pause or delete what they know about you, and how to use memory so it saves time instead of causing awkward or private mix-ups.

ChatGPT memory vs Claude memory: quick comparison

Here is how the two assistants handle memory side by side, based on OpenAI’s and Anthropic’s help centers as of October 2026.

What AI memory is (and what it is not)

Every AI model is stateless by default: each new chat starts blank. Memory is an app feature layered on top that stores a short set of notes about you, such as “runs a digital marketing agency in Ahmedabad” or “prefers US spelling and no em dashes”, and quietly adds the relevant ones to new conversations.

Memory is different from the context window, which is everything the model can read in a single chat, including your uploaded files and the full conversation so far. Memory is small and selective; the context window is large but resets with every new chat. Our context window explainer covers the difference in detail.

Memory is also different from custom instructions or a project’s instructions. Instructions are rules you write once and the app always applies. Memory is learned and can change over time, which is both its strength and its risk.

How ChatGPT memory works

Saved memories

Saved memories are details you ask ChatGPT to remember (“Remember that my store ships only within India”) or details it saves on its own as useful context, where that behavior is available. They are stored separately from your chat history, so deleting the original chat does not remove them.

Reference chat history

Reference chat history lets ChatGPT draw on relevant past conversations even if nothing was formally saved. Unlike a saved memory, this information can change as ChatGPT updates what seems most useful. OpenAI says “There is no separate storage limit for what ChatGPT can reference through chat history,” but also that the memory summary is high level and may omit details.

What each plan gets

OpenAI says memory features “can vary by plan, region, platform, and workspace settings.” The pricing page describes memory per plan in broad terms:

Depending on plan and platform, you may also see memory search, sorting, “automatic prioritization” and version history. Our ChatGPT pricing guide explains everything else each plan includes.

How to view, edit and turn off ChatGPT memory

1. Open memory settings. Go to Settings > Personalization > Memory. Here you can see the memory summary and the individual saved memories.
2. Delete specific memories. Remove any saved memory that is wrong, outdated or too personal. Remember that deleting a chat alone does not delete a memory made from it.
3. Turn off reference chat history only. If you want ChatGPT to keep your explicit saved memories but stop pulling from past chats, switch off Reference chat history. Remembered chat information is then scheduled for deletion within 30 days.
4. Turn off memory entirely. Switch off Memory. Turning off Reference saved memories also turns off Reference chat history. Your past chats stay, and ChatGPT may create new memories from them if you turn memory on again later.
5. Wipe it. Use the more options menu on the memory summary and choose “Delete and turn off memory”. To fully remove a detail, also delete the original chat and any copies in files or connected apps. OpenAI may keep logs of deleted saved memories for up to 30 days for safety and debugging.

How Claude memory works

Anthropic redesigned Claude’s memory so it saves “a set of individual topics as you chat” rather than writing one summary at the end of each conversation. You can also simply say “remember that I write for a B2B audience” and Claude will store it.

Availability by plan

• Free, Pro and Max: memory is on by default on the web, desktop and mobile apps.
• Team and Enterprise: an owner must enable memory for the organization, and it stays off for each member until they turn it on.
• Chat search: searching and referencing past chats needs a paid plan (Pro, Max, Team or Enterprise).

Some accounts still use the legacy memory experience, which writes a synthesized summary of your chats refreshed every 24 hours and lives under Settings > Capabilities. See our Claude pricing breakdown for what each plan costs.

Project memory

Each Claude project has its own memory and project summary, kept separate from other projects and from chats outside projects. That makes projects the cleanest way to keep client A’s details away from client B. Our guide on how to use Claude Projects shows how to set them up.

How to view, pause and reset Claude memory

1. Open Settings > Memory. Your memories appear under “Topics”. Select a topic to read exactly what Claude has stored.
2. Edit or delete a topic. Correct anything wrong or remove it. You can also tell Claude in chat what to remember, change or forget.
3. Pause memory. Claude keeps existing memories but stops using them and stops creating new ones. Useful during a short sensitive project.
4. Reset memory. This permanently deletes all memories, including project memories, and cannot be undone.
5. Use incognito for one-offs. Click the ghost icon in a new chat. Incognito chats are not saved to memory or chat history, and Claude will not search them later.

What about Gemini?

Google’s Gemini also remembers you. Google’s Gemini Live page says it “can remember key details across sessions, such as preferences and important dates.” We could not confirm the exact Gemini memory settings and plan limits on an official help page at the time of writing, so check the Gemini app’s settings for personal context and past chats controls. Our Gemini review covers the app more broadly.

Memory, privacy and training: what to know

Memory controls decide what the assistant remembers about you. Training controls decide whether the vendor may use your chats to improve its models. They are separate switches.

• ChatGPT personal accounts: OpenAI may use chats and remembered information to improve models when “Improve the model for everyone” is on. Turn it off under Data controls.
• ChatGPT business workspaces: by default OpenAI does not train on content from Business, Enterprise, Edu or Healthcare workspaces.
• Safety exception: OpenAI notes that limited safety-related context may still be used in rare, high-risk situations, even with memory off.
• Regulated work: improved memory is off by default in ChatGPT for Healthcare and Regulated Enterprise workspaces, and OpenAI says it is not covered under the BAA, so do not enter protected health information.

If you run a team, write these rules down. Our business guide to AI data privacy and AI usage policy template help you decide what staff may store in memory.

How to use AI memory well: 7 practical tips

1. Save stable preferences, not projects. Good memories: your role, audience, spelling, tone and formatting rules. Bad memories: this week’s campaign numbers.
2. Tell it explicitly. “Remember: my agency’s clients are Indian D2C brands and I report in INR.” Explicit memories are more reliable than ones the AI infers.
3. Use projects for client work. In Claude, project memory stays inside the project. In ChatGPT, keep client files in separate projects too.
4. Audit monthly. Open the memory page and delete anything outdated. Old memories can quietly steer answers in the wrong direction.
5. Go private for sensitive topics. Use Temporary Chat or incognito for health, legal, salary or personal questions.
6. Do not rely on memory for long documents. Upload the file again or use a project knowledge base. Memory stores summaries, not full text.
7. Combine with good prompts. Memory sets the background, but a clear prompt still decides quality. Our prompt engineering guide shows what to include each time.

Memory can also save usage. When the assistant already knows your context, you can write shorter prompts, which helps if you often hit caps. See our tips to reduce token usage in ChatGPT and Claude.

Common memory mistakes

• Assuming that deleting a chat deletes the memory it created. In ChatGPT it does not.
• Mixing personal and client work in one account without projects, so details leak between conversations.
• Forgetting that Claude team members must opt in, then wondering why memory is not working.
• Expecting memory to recall every detail. Both vendors say it keeps highlights, not everything.

Frequently asked questions

Our verdict: turn memory on, but manage it

For most people, memory is worth keeping on. It makes ChatGPT and Claude feel like assistants that know your business, and it trims the background you retype every day. Claude’s topic-based memory and strict project separation make it the tidier choice for freelancers and agencies juggling clients, while ChatGPT’s reference chat history is stronger at picking up context you never formally saved. Whichever you use, audit memory monthly and use Temporary or incognito chats for anything sensitive. Still deciding between the two? Read our ChatGPT vs Claude comparison. Features and plan details change often, so confirm current settings on the OpenAI memory FAQ and Claude memory help page before you rely on them.