ChatGPT Memory is the profile ChatGPT keeps about you across chats. Since OpenAI's Dreaming update, which began rolling out on June 4, 2026, that profile is synthesised in the background from your past conversations and revised over time, not only built from facts you asked ChatGPT to save. Memory lives in Settings, Personalization, Memory. Which controls you see there depends on your account: OpenAI lists Memory, Reference saved memories, Reference chat history, Memory summary, Manage and Saved memories as the possible set, and says any of them may be missing if it has not reached your plan, region, platform or workspace yet. You can read the summary, correct it, mark something with Don't mention this again, delete everything and switch memory off, and check the Sources shown under a personalized reply. A temporary chat never creates or updates memories, but it can still use the ones you have unless you choose Unpersonalized before the first message. Memory and model training are separate settings. OpenAI states that the memory summary does not necessarily include every detail or source that Memory can reference. Verified against OpenAI's Help Center on September 27, 2026.
How to Manage ChatGPT Memory: View, Edit, Delete (2026)
Memory is one of the most-asked-about ChatGPT features in AI engine queries and one of the least visible inside the UI. The panel is buried two clicks deep in Settings, the per-row controls only appear on hover, and ChatGPT rarely confirms when it has saved something. This guide walks through every Memory control, with screenshots taken on September 9, 2026 and every rule re-checked against OpenAI's Help Center on September 27, 2026, explains how Memory differs from Custom instructions and Training data, and shows how AI Toolbox (formerly ChatGPT Toolbox) complements Memory by letting you export and back up the conversations that built your context and search across all of them, since native memory is account-level and cannot be exported on its own. With 40,000+ active users and a 4.7/5 Chrome Web Store rating, AI Toolbox is the most installed productivity layer for ChatGPT.
What Is ChatGPT Memory?
Memory is the layer that lets ChatGPT carry what it knows about you from one chat to the next. Until June 2026 it was a list of discrete saved memories, written when you told ChatGPT something durably useful or asked it to remember. OpenAI's Dreaming update, announced on June 4, 2026, changed the mechanism: a background process now reads your past conversations, synthesises a profile of you, and keeps revising that profile as time passes. OpenAI now labels the older saved-memories system legacy, although saved memories still exist alongside the synthesised profile. Whatever ChatGPT holds persists across sessions, devices, and the apps that share your OpenAI account. OpenAI documents the current behaviour in its Memory in ChatGPT help article and in the Dreaming announcement.
Memory operates separately from Custom instructions and from the data-training toggle. Custom instructions is a single static text block that applies to every new chat. Memory is dynamic: it is updated as ChatGPT learns about you, and you can correct or remove what it holds. Whether your data trains future OpenAI models is controlled by a separate switch in Settings, Data controls.
Memory is not built from your chats alone. OpenAI lists five sources it can draw on, depending on your plan and region: past chats, saved memories, custom instructions, files in Library, and content from connected apps such as Gmail. That matters when you try to remove something, because deleting the conversation only clears one of the five. It also means a personalized answer can quote a document or an email you never mentioned in the chat you are reading.
Control
What it does
Where it lives
Memory (Enable memory)
One toggle that lets ChatGPT personalize from your chats, files, and connected apps
Settings, Personalization, Memory
Reference chat history
Where the control exists, governs whether past conversations feed personalization. Turning it off schedules what was remembered from past chats for deletion within 30 days
Settings, Personalization, Memory
Memory summary, Manage
Opens the profile ChatGPT has written about you, with an Ask or update box for corrections
Settings, Personalization, Memory summary
Don't mention this again
Stops ChatGPT referring to a detail without deleting the chat, file or email it came from
On the memory summary, where the option appears
Sources
Lists the past chat, saved memory, custom instruction, file or connected email behind a personalized reply, with actions to correct or delete each one
Under the response in the chat
Delete and turn off memory
Deletes the memories shown on the summary and switches memory off. It does not delete past chats
Three-dot menu on the Memory summary page
Saved memories (legacy)
The explicit list of individual memory items, still reachable, and switchable back and forth with improved memory where the option exists
The saved memories link under Memory summary
Custom instructions
Static text applied to every new chat, including the More about you field
Settings, Personalization
Improve the model for everyone
Controls whether your data trains future models (independent of Memory)
Settings, Data controls
One Toggle, or Three: It Depends on Your Account
From April 10, 2025 until the Dreaming rollout, memory had two switches: saved memories, the explicit facts you asked ChatGPT to keep, and reference chat history, which let ChatGPT draw on past conversations even when nothing was saved. OpenAI extended reference chat history to Free users on June 3, 2025. On the account photographed for this guide, the Personalization panel showed a single Enable memory toggle, described as personalizing your experience from your chats, files, and connected apps, and that toggle governs the synthesised profile Dreaming maintains: a background process runs after conversations end and writes inferences drawn from them, not only facts you stated outright.
Do not assume one switch is all you have. OpenAI's help centre says the available controls "may include Memory, Reference saved memories, Reference chat history, Memory summary, Manage, or Saved memories", depending on your memory experience, and that a control described there may simply not have reached your account yet. Where Reference chat history exists as its own control, three rules apply, all of them OpenAI's: turning it off schedules information remembered from past chats for deletion from OpenAI systems within 30 days, while the chats themselves stay in your history unless you delete them; turning it off does not delete your saved memories, which are stored separately; and there is no separate storage limit for what ChatGPT can reference through chat history. Where Reference saved memories exists and is on, Reference chat history can be managed on its own.
The two experiences can also be swapped where OpenAI offers the choice. Improved memory "continually updates a broader summary of relevant context", while legacy saved memories "use a more explicit list of individual memory items". In Settings, Personalization, Memory you select Saved memories to use the legacy list or Try improved memory to go back, and OpenAI notes that changing the experience does not delete your chat history. So check the panel itself rather than trusting any guide, including this one, about how many switches you have.
The profile also rewrites itself as time passes. OpenAI's own example from the Dreaming announcement: a memory reading "You're going to Singapore in July" revises itself to "You went to Singapore in July 2026" once the trip is over, without you editing anything. That is useful for a travel plan and worth knowing for anything sensitive, because ChatGPT can hold a conclusion about you that you never typed.
For a single conversation you want kept out of all of this, use Temporary Chat, and pick the right mode before you start. A temporary chat stays out of your history and is not used to improve OpenAI's models while it remains temporary, and neither mode creates or updates memories. The difference is what it reads: Personalized "can use your existing memories, custom instructions, and plugins", while Unpersonalized "does not use memory, custom instructions, or plugins". You choose before sending the first message and you cannot change it afterwards. Saving the chat converts it into a regular one, after which your normal personalization and model-improvement settings apply. OpenAI may keep a copy of a temporary chat for up to 30 days for safety purposes. So temporary is the cleanest answer to "do not remember this", and Unpersonalized is the answer to "do not use what you already know".
How to Open ChatGPT Settings
Click your profile avatar in the bottom-left of ChatGPT, then select Settings from the dropdown menu. The avatar dropdown also holds the My GPTs, Customize ChatGPT, Help, and Log out entries. Settings is the third row in the standard menu. On the web app, the Settings dialog opens in a centered modal. On the desktop app, Settings opens as a native sidebar panel; the contents are identical.
The Settings dialog has a left sidebar of categories, starting General, Notifications, Personalization, with Data controls further down. Memory lives under Personalization, the third entry from the top, below the custom-instructions fields such as More about you.
Toggle Memory On or Off
In Settings, Personalization, the Memory section has one switch, Enable memory, with the description "Let ChatGPT personalize your experience based on your chats, files, and connected apps." It was on by default on the account photographed for this guide. Flip it off and ChatGPT stops personalizing replies from what it has learned. The toggle does not delete the underlying memory; deletion is a separate command on the summary page, covered below. Flip it back on and the same profile is used again.
The toggle is the right control if you want to test how ChatGPT responds without personalization, or if you want to pause memory during a privacy-sensitive task. For a single conversation, Temporary Chat is simpler. To remove the data, use Delete and turn off memory (see the delete-everything section below).
View Your Memory Summary
Click Manage next to Memory summary to open the page titled Memory summary. It is written as prose, not as a list: an Updated timestamp sits beside the title, the body is organized under headings such as Overview and Products, and an Ask or update box runs along the bottom. On the account photographed for this guide the Overview read like a short biography, naming role, products, languages, and audience, several of which had never been typed into ChatGPT as a "remember this" request. There is no filter, sort, or search; the only navigation is scrolling.
Treat what you see as a summary, not a complete record. OpenAI's help centre states that the memory summary "does not necessarily include every detail or source that Memory can reference", and answers the question directly in its own FAQ: the summary "is a high-level view and may omit details that are less relevant, unsuitable for the summary, or derived from sources that are not displayed individually". Its suggested alternative is to ask: "You can also ask ChatGPT what it remembers about you." If a reply surprised you with a personal detail you do not remember sharing, the source may be a saved memory, the synthesised profile, reference chat history, a file in Library or a connected app, and it may not be listed on the summary page. Re-read the summary every few weeks if you use ChatGPT for sensitive work, and correct or delete anything that should not carry over between projects.
What the Memory View Does Not Show
This is OpenAI's published caveat, not a third-party accusation: the memory summary is a curated view of a larger synthesis, and some of what ChatGPT remembers is not displayed. Its help centre says the summary "does not necessarily include every detail or source that Memory can reference", and that details can be omitted when they are "less relevant, unsuitable for the summary, or derived from sources that are not displayed individually". Tom's Guide reported the same acknowledgement on June 18, 2026. OpenAI's stated mitigation is the Sources panel described in the next section, which names the specific memory or chat behind a given answer. Even there the hedge repeats: "Sources may not show every factor that shaped a response." The practical rule: if a detail must not influence ChatGPT, delete it at the source and turn memory off rather than assuming its absence from the summary means it is gone.
Sources: Which Memory Shaped a Reply
When ChatGPT uses personal context, a Sources control can appear below the response, and it is the only place that names the specific item responsible. OpenAI says Sources "can help you review information such as a past chat, saved memory, custom instruction, file, or connected email that contributed to personalization". Depending on the source, you may be able to open or review it, correct or delete a saved memory, delete a referenced chat, mark a source as relevant or not relevant, or open a referenced file or email.
Two limits are worth knowing. Sources "may not show every factor that shaped a response", so it is a window rather than an audit log. And Memory Sources are not included when you share a conversation through a shared link, so anyone opening your shared chat sees the answer without the personal context that produced it. If a reply pulls in something you want gone, Sources is the fastest route to the item itself, which is usually faster than hunting for it in the summary.
Correct or Remove a Single Detail
Type the change into the Ask or update box at the bottom of the Memory summary, for example "Remove the note about my audience size" or "Update my role to senior content lead", and send it. The summary is rewritten and its Updated timestamp changes. OpenAI also describes highlighting text in the summary to make a specific correction. There is no per-row trash icon any more, because there are no rows. The original conversation transcripts in your chat history are unchanged either way.
There is a softer control next to it, and it is easy to misread. Where the option appears, Don't mention this again tells ChatGPT to stop bringing something up. OpenAI is explicit about what it does not do: it "reduces future references to the information" and "does not delete the original source". The chat, file or email that carries the detail is untouched, which means the information is still there to be referenced by a different route later. Use it for things you are tired of hearing, and use deletion for things that should not exist.
For memories written before June 2026, open the saved memories link in the Memory section; that legacy list is managed separately and is described in OpenAI's help centre. Use corrections to prune details that drifted out of date, are wrong, or belong to a finished project: old job titles, abandoned product names, a tone you no longer want, and personal details you would rather not have ChatGPT reference. For a clean slate, jump to the next section.
Save the prompts behind your best memories so you never lose them. AI Toolbox adds folders, full-text search across your active (unarchived) ChatGPT conversations, a prompt library with the // shortcut, and prompt chaining for multi-step workflows. Free plan covers 2 folders, 2 pins, and 2 saved prompts. Unlock unlimited everything on Premium ($9.99 per month) or pay once for Premium Lifetime ($99). Working across ChatGPT, Gemini, Claude, and Grok? All Access Lifetime ($199) covers all four modules and saves $197 versus four single-platform Lifetimes ($99 × 4 = $396). Add AI Toolbox to Chrome free →
Delete Everything and Turn Memory Off
Open the three-dot menu beside the Memory summary title and choose Delete and turn off memory. The menu has only two items, About memory and that red delete command. It removes the memories shown on the summary and switches memory off in one step, which is why there is no separate Clear all button any more. To start collecting memory again afterwards, switch Enable memory back on in Settings, Personalization.
Delete and turn off memory is the right control when you want a clean slate, for example when handing the account over to a new role, switching from personal to work use, or recovering from a chat session where ChatGPT saved entries you did not want. The action cannot be undone from your side. OpenAI notes that it may retain logs of deleted saved memories for up to 30 days for safety and debugging purposes, after which they are gone. It also states that the command "does not delete past chats", so your history survives the wipe.
Deleting the conversation is not the same as deleting the memory, and this is the single most common mistake. OpenAI puts it plainly: "Deleting a chat alone does not necessarily delete a separate saved memory created from that chat." Saved memories are stored separately from chat history, so the chat can be gone while the fact it produced is still in play. To remove a detail properly, OpenAI lists the places it can live and asks you to clear each one: the memory summary, saved memories, regular and archived chats, files in Library, and connected apps. Disconnecting an app is its own caveat: it "prevents future access to that app" but "does not delete past conversations that already used app content".
Memory vs Training Data: Two Different Toggles
Memory and the model-training switch are independent. Memory controls whether ChatGPT remembers facts for your own replies. The model-training control (Settings, Data controls, Improve the model for everyone) controls whether your conversations and memories can be used to train future OpenAI models. Turning off Memory does not opt you out of training. Opting out of training does not stop ChatGPT from saving memories for your own use.
If you want full privacy: use Delete and turn off memory on the summary page, check the legacy saved memories link, then open Settings, Data controls, and turn off Improve the model for everyone. With both switches off, ChatGPT operates without personalization and your conversations do not feed training. For a deeper walk-through of every data-controls switch, see the ChatGPT limits and controls guide.
Force ChatGPT to Remember Something Specific
Start the message with "Remember that..." or "Save this to memory:..." to force ChatGPT to write a specific entry. ChatGPT replies with a Memory updated indicator under the assistant message when it has saved the fact. Without the explicit instruction, ChatGPT decides on its own what to keep, and since Dreaming that decision is no longer limited to facts you stated outright: a background process runs after a conversation ends and writes inferences drawn from it into your profile. A casual mention may not produce a saved memory row, yet it can still shape the synthesised profile.
The same syntax works for editing or replacing an existing memory. If a saved row is wrong, send a message like "Update the memory about my role to: senior content lead at AI Toolbox". ChatGPT writes a new entry that supersedes the old one. Check the Memory summary afterwards to confirm the old detail is gone; if it lingers, send a correction through the Ask or update box.
How AI Toolbox Extends ChatGPT Memory
AI Toolbox does not change Memory itself; ChatGPT keeps building and applying your profile exactly as described above. Native memory is account-level and has no export of its own: OpenAI describes the data-export ZIP as "your chat history and other relevant account data" and does not document memory as a component of it, the memory summary has no download button, and the documented way to read what ChatGPT holds is to open the summary or to ask ChatGPT in the chat. Saved memories have to be copied out of Settings by hand. AI Toolbox complements memory by letting you export and back up the conversations that built that context and search across all of them. It runs as one unified Chromium install (Chrome, Edge, Brave, Opera, Arc) on top of the native ChatGPT interface.
Export and back up the source conversations as TXT, Markdown, JSON, or PDF, so the chats that taught ChatGPT each memory have a local copy you control.
Full-text search across your active ChatGPT conversations (archived chats are not indexed), including the messages where ChatGPT saved a memory in the first place. Use Cmd+Shift+F (Ctrl+Shift+F on Windows) to search by message content, role, or date range with exact-match toggle.
File memory-heavy chats into folders to group them by project. Folders use search-based adding: you search the conversation list and add the matches to the folder, no drag-and-drop required.
AI Toolbox is also available for Gemini and Claude as part of the same unified install. The Gemini module adds full-text search across every synced Gemini conversation with date, role, and exact-match filters. The Claude module adds full-text search with exact-match toggle and message-level bookmarks. Pricing details are on the AI Toolbox pricing page.
Frequently Asked Questions
What is ChatGPT Memory?
Memory is the profile ChatGPT keeps about you across chats, covering things like your preferences, ongoing projects, and writing style. Since OpenAI's Dreaming update, rolling out from June 4, 2026, a background process synthesises that profile from your past conversations and revises it over time, on top of the saved memories you explicitly ask for. You review it in Settings, Personalization, Memory, and OpenAI states the summary shown there does not include everything ChatGPT remembers.
How do I see what ChatGPT remembers about me?
Click your profile avatar, open Settings, go to Personalization, and click Manage next to Memory summary. A page titled Memory summary opens with an Updated timestamp, prose sections such as Overview and Products, and an Ask or update box at the bottom. OpenAI states the summary does not include everything ChatGPT remembers. Older explicit memories are behind the saved memories link in the same panel.
How do I delete a single ChatGPT memory?
Open the Memory summary (Settings, Personalization, Manage) and type the change into the Ask or update box, for example "Remove the note about my audience size". The summary is rewritten and its Updated timestamp changes; OpenAI also describes highlighting text in the summary to correct it. There are no rows or trash icons any more. Memories from before June 2026 are managed through the saved memories link instead.
How do I wipe all ChatGPT memories at once?
Open the Memory summary, click the three-dot menu beside its title, and choose Delete and turn off memory. That single command removes the memories shown on the summary and switches memory off. OpenAI notes it may keep a log of deleted saved memories for up to 30 days for safety and debugging. To collect memory again later, switch Enable memory back on in Settings, Personalization.
Can I turn ChatGPT Memory off completely?
Yes. In Settings, Personalization, switch Memory off. Depending on your account you may also see Reference saved memories and Reference chat history as separate controls, so check the panel rather than assuming a single switch. Turning memory off does not delete existing data and does not delete your past chats, so also use Delete and turn off memory on the summary page and check the saved memories link for older entries. For a one-off conversation, start a Temporary Chat and choose Unpersonalized before your first message, which stops it using existing memories as well as creating new ones.
Does ChatGPT Memory train OpenAI's models?
Memory entries are stored against your account and used to personalize replies. Whether they train future models depends on your data controls setting in Settings > Data controls > Improve the model for everyone. Turning that toggle off keeps your conversations and memories out of model training. Memory and training are independent settings, so you can keep one on and the other off.
Why does ChatGPT keep forgetting things I told it?
ChatGPT keeps what its own process judges useful, and since the Dreaming update that judgement runs in the background after a conversation ends rather than at the moment you say something. A casual mention may not become a saved memory row, and a detail can still be missing from the profile weeks later. To make sure a fact is kept, start the message with an explicit instruction like "Remember that..." or "Save this to memory: ..." and watch for the Memory updated indicator. If both memory controls are off, nothing is saved at all.
Does deleting a ChatGPT chat delete what it remembered?
Not necessarily. OpenAI states that "deleting a chat alone does not necessarily delete a separate saved memory created from that chat", because saved memories are stored separately from chat history. To remove a detail for good, clear it from the memory summary or saved memories, delete the chat where you shared it, and also check archived chats, files in Library and any connected app that holds the same information.
How do I see which memory ChatGPT used in an answer?
Look for Sources under the response. OpenAI says it can show the past chat, saved memory, custom instruction, file or connected email that contributed to personalization, and lets you open the source, correct or delete a saved memory, delete a referenced chat, or mark a source as not relevant. Two limits: Sources "may not show every factor that shaped a response", and Memory Sources are not included in a shared link.
What is the difference between ChatGPT Memory and Custom instructions?
Custom instructions is a single static text box you write yourself, and it applies to every new chat. Memory is written by ChatGPT: the saved memories you ask for, plus a profile that a background process synthesises from your past chats and revises over time. You change custom instructions by rewriting the box; you change memory by correcting or deleting what ChatGPT holds. Custom instructions cover your stable identity. Memory covers everything ChatGPT has learned about you since.
Conclusion
Memory turns ChatGPT from a stateless chat into a tool that gets better the more you use it, as long as you keep the saved list curated. Open Settings, Personalization, and read the Memory summary every few weeks. Correct details that have drifted out of date through the Ask or update box. Use Delete and turn off memory when you start a new project or hand the account to a new role. Turn the Memory toggle off when you want a privacy-sensitive session, and remember that the data-training toggle is a separate switch. To search across the conversations Memory was built from, save the prompts that produced your best memories, and organize memory-heavy projects into folders, install AI Toolbox for Chrome: free plan covers 2 folders, 2 pins, 2 saved prompts, and full-text search with up to 5 results per query. Premium ($9.99 per month) or Premium Lifetime ($99 one-time) unlocks unlimited everything plus bulk export.
A Full Stack Developer with 7+ years of experience building AI productivity tools. Leads product development and frontend architecture for AI Toolbox, the Chrome extension suite (ChatGPT, Gemini, Grok, and Claude modules) that helps users search, organize, and export their AI conversations.
Liked this guide? Get the next one first.
Join 40,000+ readers getting practical prompts, new features, and subscriber-only deals. No spam, unsubscribe in one click.
New features and product updates before anyone else
Time-saving prompts and workflows for ChatGPT, Gemini, Claude, and Grok
Adi Leviim