ChatGPT does not keep a tidy dossier you can read in one screen. OpenAI's Memory FAQ states plainly that the memory summary "will not include everything that ChatGPT remembers based on your chats," and that removing something completely means deleting every source it came from: past chats, archived chats, files, the memory summary, and any connected apps.
So the honest answer to "what is in my ChatGPT profile" is not a list of fields. It is a set of sources that feed a running synthesis, plus a summary page that shows you the highlights of that synthesis. This piece walks through what OpenAI has actually published, what it has not, and the exact steps to audit your own account today.
Where ChatGPT's memory actually comes from
OpenAI describes memory by input, not by content type. The Memory FAQ says that when enabled, memory "helps ChatGPT automatically remember useful context from your chats, files, and connected apps to personalize your experience." That sentence is the closest thing to an official inventory that exists.
- Your chats, including old ones still sitting in your history and archived ones you forgot about.
- Files you have uploaded into conversations.
- Connected apps you have authorised, which stay in scope until you disconnect them.
- Anything you explicitly asked it to remember, under the legacy saved memories system.
- Custom instructions, which OpenAI treats as separate from memory but which still shape responses.
The architecture behind this changed in 2026. On 4 June 2026 OpenAI launched a new memory architecture built on a background process it calls dreaming, first introduced in April 2025 and described as "a background process that allows ChatGPT to learn from many conversations and synthesize ChatGPT's memory state." The company says it makes memory able to include context that "occurs naturally in conversation, without relying on explicit requests to remember something." The rollout started with Plus and Pro users in the US, with Free and Go users following.
That is the meaningful shift. The older system leaned on strong cues, mostly an explicit instruction to remember something. The current one runs in the background and keeps rewriting a picture of you. OpenAI gives an example of the rewriting itself: a memory can change from "You're going to Singapore in July" to "You went to Singapore in July 2026" once the trip ends.
What OpenAI documents, and what is rumour
Here is the counter-narrative most articles get wrong. There is no published OpenAI taxonomy of memory categories. No help article, no policy page, and no announcement enumerates the buckets ChatGPT keeps on you. Every list you see online with headings like "health," "finances," "relationships" is somebody's reconstruction, not documentation.
What OpenAI does give is illustrative examples. Saved memories are described as "details you have explicitly asked ChatGPT to remember, like your name, favorite color, or dietary preferences." The Dreaming announcement frames memory as learning "your preferences, projects, and constraints." It also names three shapes a preference can take: instructions on how to respond, stated personal constraints such as being vegetarian, and implicit preferences, its example being that saying you live near San Francisco should steer local recommendations.
The word "admission" has been attached to this topic in a lot of recent coverage. Treat it carefully. There is no separate confession document. What exists is the plain text of OpenAI's own help pages, which is candid enough on its own and which you can read in five minutes. If a claim about ChatGPT's memory cannot be traced to a help article, a policy page, or an OpenAI announcement, it is somebody's inference about an inference engine.
The line that surprises people most
OpenAI's Memory FAQ answers the question "Does the memory summary include all my memories?" with "Not necessarily." It explains that memory is based on "a continually updated synthesis of context from your past chats, which may be broader than what can be shown as individual items in a summary." Then it adds the sentence worth reading twice: some details may not appear "including when ChatGPT determines they are less relevant or not appropriate to show in this view."
The memory summary is a view, not the store. OpenAI says outright that it will not include everything ChatGPT remembers, and that the system itself decides which details are not appropriate to display there. Auditing the summary tells you a lot. It does not tell you everything.
OpenAI's suggested workaround is conversational rather than technical: "If you want to know if ChatGPT has remembered something, just ask in chat." The same page notes the sources panel is also incomplete, saying sources "may not show every factor or source that shaped a response." Two honest disclosures, both in the official FAQ, both routinely missed.
How to open and read your memory summary
The controls live in Settings, then Personalization, then Memory. That path is current as of OpenAI's July 2026 update to the Memory FAQ, and OpenAI renames these surfaces often enough that it is worth checking the article rather than trusting a screenshot from last year.
- The summary shows a timestamp at the top telling you when it was last updated, such as "2 hours ago."
- If it is empty, OpenAI says that usually means a new account, recently enabled memory, or not enough chat history.
- You can force a rebuild: Settings, Personalization, Memory Summary, Manage, then the three-dot menu and Refresh.
- The legacy saved memories list still exists. Settings, Memory, Saved memories switches you back to it, and "Try improved memory" switches forward again.
Read the summary once with a specific question in mind: which sentences here did you never actually type? Stated facts are easy to scan past. The valuable finds are the conclusions, the ones drawn from context rather than dictated, because those are the ones that will shape answers you did not expect to be personalised.
How to see which memories shaped a single answer
This is the least known control and the most useful for an audit. Tap the book icon below a response and ChatGPT shows the sources used to personalise it, including custom instructions, past chats, files, and memories. Tapping a memory in that panel opens an explanation of why it was applied to that specific answer, and the three-dot menu there lets you make a correction on the spot.
One detail matters for anyone who shares transcripts: OpenAI states that memory sources are not included in chats you share. The person opening your shared link sees the answer, not the profile fragments that produced it.
How to correct or delete something
Editing is now done in plain language rather than through a list of checkboxes. OpenAI documents three routes: type what you want changed into the text box at the bottom of the memory summary, highlight any passage in the summary to make a specific correction, or use the three-dot menu on a memory shown in sources.
Full removal is harder, and OpenAI says so. The FAQ states: "To fully delete something ChatGPT may know about you, you'll need to delete every source where it appears, including past chats, archived chats, files, the memory summary, and disconnect any connected apps that may contain that information." That is a five-place scavenger hunt for a single fact, and the archived-chats step is the one almost everyone forgets.
There is a nuclear option: the three-dot menu on the memory summary page offers "Delete and turn off memory." OpenAI notes it does not delete your past chats, that turning memory back on later means ChatGPT "may create new memories from chats that remain in your chat history, including older chats," and that the control was on the web with mobile rolling out.
What survives when you delete a chat
Deleting a conversation does not delete what ChatGPT learned from it. Under the legacy system OpenAI is explicit that saved memories are "stored separately from your chat history," so "even if you delete a chat, any saved memories from it can still be used in future conversations." The reverse is also true: deleting a memory does not scrub mentions of it from old transcripts.
| What you do | Effect on memory | Effect on chat history |
|---|---|---|
| Delete one chat | Memories already formed from it can still be used | That conversation is removed |
| Edit or delete an entry in the summary | Removed from the summary view | The original chat still contains it |
| Turn memory off | Nothing already stored is deleted | Chats stay exactly where they are |
| Turn memory back on later | New memories can form from chats still in history, including old ones | No change |
| Use a Temporary Chat | Existing memories are not used and none are created | Never appears in history; kept up to 30 days for safety |
| Delete your account | Removed within 30 days absent a legal or safety hold | Same 30-day window |
Two retention numbers are documented. OpenAI says it "may retain a log of deleted Saved Memories for up to 30 days for safety and debugging purposes." Its US privacy policy says that once you choose to delete personal data, "we will remove it from our systems within 30 days" unless it must be kept longer for legal, safety, fraud or record-keeping reasons, or unless it has already been de-identified and disassociated from your account through model improvement.
Memory and model training are different switches
Turning memory off does not stop your conversations being used to improve OpenAI's models, and turning off training does not stop memory. They sit in different settings panels. Training is controlled by "Improve the model for everyone" under Settings, then Data Controls. OpenAI says switching it off means your conversations "will still appear in your chat history but won't be used to train ChatGPT," that the setting applies account-wide across devices, and that you can change it at any time.
Temporary Chat is the one control that covers both at once. OpenAI states Temporary Chats are deleted from its systems after 30 days, are not used to train its models, do not get saved in your history and do not create memories. Custom instructions still apply if you have them enabled. And there is a documented exception worth knowing: turning off memory and personalization "does not disable safety features that may use limited, safety-relevant context in rare, high-risk situations."
A ten-minute audit of your own memory
Run this in order. It goes source by source, which is how OpenAI itself frames removal, rather than hunting for categories that were never published.
- Open Settings, Personalization, Memory and read the whole summary. Note the last-updated timestamp.
- Mark every sentence describing something you never stated outright. Those are conclusions, and they are where the surprises live.
- Ask directly in chat what it knows about a sensitive topic. OpenAI's own advice, because the summary is incomplete.
- Tap the book icon under two or three recent answers to see which memories are actually being applied in practice.
- Correct wrong entries by highlighting the passage or typing the fix into the summary text box.
- Check your archived chats, not just active ones, before assuming a topic is gone.
- Review connected apps and disconnect anything you no longer use.
- Decide on your two switches: memory in Personalization, training in Data Controls. They are independent.
Before a conversation you would not want feeding personalisation, start a Temporary Chat instead of cleaning up afterwards. OpenAI documents that it neither reads existing memories nor creates new ones, which is a far cleaner outcome than deleting five sources later.
One more honest note. OpenAI also states that "sensitive information may appear in memory if you share it with ChatGPT." In its 2024 memory announcement OpenAI said it steers ChatGPT away from proactively remembering sensitive information such as health details unless you explicitly ask, and it has not restated that commitment for the 2026 architecture. Either way it is a steer, not a guarantee, and it does not apply to anything you ask it to keep. What goes in can come back out in an unrelated conversation months later.
The structural problem underneath all of this
The difficulty is not that OpenAI is hiding things. It is that a synthesised memory has no clean edges. When a system reads across years of conversations and rewrites its own picture of you in the background, there is no single row in a table to point at and delete. That is why removal takes five steps, and why the summary can only ever be a highlights reel.
That trade-off is exactly what MemX takes a different position on. MemX only knows what you deliberately save into it, photos, PDFs, voice notes and chats, and every answer it gives comes back with the source document attached, so there is nothing inferred sitting behind the curtain. It is private by architecture: per-user isolation, customer-managed encryption keys, encryption at rest, capture on your device. Less magic, and a memory you can actually read end to end.
01Can I see everything ChatGPT remembers about me?
No. OpenAI's Memory FAQ says the memory summary "will not include everything that ChatGPT remembers based on your chats," and that some details are withheld when the system judges them less relevant or not appropriate to display. OpenAI's suggested workaround is to ask ChatGPT directly in a conversation.
02Does deleting a chat delete what ChatGPT learned from it?
No. OpenAI states saved memories are stored separately from chat history, so memories from a deleted chat can still be used later. Removing something completely means deleting past chats, archived chats, files, the memory summary, and disconnecting connected apps that hold it.
03What categories of information does ChatGPT store about me?
OpenAI has never published a category list. Its documentation describes sources, meaning your chats, files and connected apps, and gives examples like your name, dietary preferences, projects and constraints. Any tidy list of memory categories circulating online is a reconstruction, not official documentation.
04How do I delete one ChatGPT memory instead of all of them?
Highlight the passage in the memory summary to correct it, type the change into the text box at the bottom of the summary, or use the three-dot menu on a memory shown in the sources panel under a response. The legacy saved memories list also supports deleting individual entries.
05Does turning off memory stop OpenAI training on my chats?
No, they are separate controls. Memory sits in Settings, Personalization. Training sits in Settings, Data Controls, under "Improve the model for everyone." Temporary Chats cover both: OpenAI says they create no memories, are not used for training, and are deleted after 30 days.
