How to delete ChatGPT memory
Delete, clear, reset, erase — in OpenAI's interface they are the same few controls. Here they are in order of how much they destroy, checked against OpenAI's own documentation on 8 September 2026; then how to see what's actually in there, why delete sometimes looks like it did nothing, and the question underneath the one you searched for: why you're pruning a memory at all.
The short version. OpenAI puts the memory controls at Settings > Personalization > Memory. What you do next depends on which of two memory systems your account is running, and that screen tells you: if you see a memory summary — a block of prose with a "last updated" stamp — you're on the current system, and the three-dot menu there holds Delete and turn off memory (added 12 June 2026, per OpenAI's release notes). If you see a list of individual entries, you're on the legacy saved-memories system, where in OpenAI's own words you "can delete individual memories, clear all of them, or turn memory off completely in settings." Either way, deleting a chat is a separate action from deleting memory. On the legacy system OpenAI is explicit that the two are stored apart, so saved memories survive a deleted conversation; on the current system it says only that the memory summary is one of several sources you have to clear in its own right. Full detail below, and then the part worth five more minutes: the reason you had to choose what to lose.
First, find out which memory you're deleting
This is the step every thin walkthrough skips, and it's why so many of them give instructions that don't match your screen. ChatGPT currently has two memory systems, and OpenAI lets accounts move between them.
- The current system — a memory summary. OpenAI describes ChatGPT's memory as "a continually updated synthesis of context from your past chats," surfaced as a summary page that "is automatically updated with new context as you chat" and stamped at the top with when it last changed ("2 hours ago"). You edit this one in place rather than delete from a list.
- The legacy system — saved memories. A set of discrete entries — "Memory first launched in April 2024 (also known as saved memories)," in OpenAI's telling. Its own verdict on that system appears in two places, and they are worth keeping straight: the 4 June 2026 announcement says saved memories "tend to go stale over time and eventually become incorrect or irrelevant," while the Memory FAQ says the previous system "often became stale and relied on users to manually manage updates."
You can move between them deliberately. OpenAI's instructions: "Go to Settings > Memory > Saved memories to revert to the legacy saved memories system. From saved memories, you can select Try improved memory to return to the updated memory experience." So if the steps below don't match what you're looking at, you're on the other one — check that before you assume the feature moved. One more reason the screens differ: OpenAI's 12 June 2026 release note says "memory updates, including memory summary, are still rolling out across ChatGPT plans and countries," and that you'll be notified in ChatGPT when they reach you.
Option 1 — correct one thing without deleting anything
Worth trying first, because most people arrive here wanting one wrong fact gone, not a clean slate. On the current system, OpenAI documents two ways to edit the memory summary in place:
- Type the correction. "Type what you want changed into the text box at the bottom of the memory summary and it will update accordingly."
- Highlight the offending line. "You can highlight any text in the memory summary to make a specific correction."
There's a third route from inside a conversation. Tap the book icon below a response and you get the sources that personalised it — custom instructions, past chats, files, memories. Tapping a memory there opens an explanation of why it was used, and OpenAI says you "can also make a correction by selecting the 3 dot menu." When it introduced that view (release note dated 5 May 2026), OpenAI described the choices as: "If something is outdated or no longer relevant, you can correct it, delete it, or mark it as not relevant." That's the fastest fix when you've just watched ChatGPT repeat something about a client that stopped being true two quarters ago: correct it where you caught it, rather than going hunting for it in Settings.
If your memory summary is empty and you expected content, OpenAI's documented refresh is Settings > Personalization > Memory Summary > Manage, then the triple-dot menu, then "Refresh."
Option 2 — delete individual memories (legacy saved memories)
On the legacy system the controls are exactly what you'd hope for: OpenAI states you "can delete individual memories, clear all of them, or turn memory off completely in settings." You open the saved-memories list, read the entries, and remove the ones that are wrong, stale, or no longer yours to carry. Finding the right one got easier in October 2025, when OpenAI added the ability to "search your saved memories more easily and sort them by newest or oldest" (release note dated 15 October 2025, rolled out first to Plus and Pro on the web).
One consequence people don't expect. Deleting an entry stops it shaping future answers, but it doesn't retroactively clean your history: "You can delete that memory anytime, but it won't remove mentions from past chats. To clear that, you will need to delete the specific chat where it was shared." So if the reason you're deleting is that something sensitive got written down, the memory entry is one of at least two places to visit.
Option 3 — delete everything and switch memory off
On the current system these are the same action, which is the most important thing to know before you tap it. OpenAI: "You can delete the memories shown on your memory summary page and turn memory off by selecting Delete and turn off memory from the three-dot menu." The control arrived on 12 June 2026, according to OpenAI's release notes, eight days after the memory summary itself.
Three qualifications OpenAI attaches to it, all worth reading before you commit:
- It doesn't touch your chats. "This does not delete your past chats." The conversations stay; the summary built from them goes.
- It can come back. "If you turn memory back on later, ChatGPT may create new memories from chats that remain in your chat history, including older chats." Off is not the same as erased — the raw material for a new summary is still sitting there.
- It may not be on your phone yet. OpenAI notes this control "is available on the web and rolling out on mobile soon" — wording unchanged between our 28 August and 8 September readings. If you can't find it in the iPhone app, that's why; open ChatGPT in a browser rather than assuming you've missed a menu.
On the legacy system the two jobs stay separate — clear the entries out of the saved-memories list, and turn memory off. OpenAI's wording covers both: you "can delete individual memories, clear all of them, or turn memory off completely in settings," with the legacy controls under Settings > Memory, where the switch is labelled "Enable memory." Labels on that screen change often enough that it's worth reading what's actually in front of you rather than hunting for a specific button someone described in a video.
And if you only want one conversation to leave no residue, you don't need any of this. OpenAI's answer is Temporary Chats, which by default "do not use existing memories or create new memories" — a way to keep a single client conversation out of memory entirely without giving up everything memory has already learned. Two changes from 27 August 2026 (OpenAI release notes) are worth knowing: you can now choose to personalise a temporary chat when you start it, in which case it "can use your memories" but still does "not create new memories," and you can save a temporary chat, which "converts it into a regular chat" that memory can then draw on. For the no-trace use, leave both switches alone.
How to reset or erase ChatGPT memory (it's the same button)
"Reset" and "erase" are searched almost as often as "delete," so it's worth saying plainly: OpenAI's documentation has no separate reset or erase control. Both words map onto the options above, and the only question is how far back you want to go.
- To erase memory on the current system: Delete and turn off memory (Option 3). On the legacy list: "clear all of them."
- To reset memory — wipe it and let it start again: the same action, then turn memory back on. OpenAI tells you what happens next: "ChatGPT may create new memories from chats that remain in your chat history, including older chats." A reset that keeps your chats is a reset back towards the same memory, rebuilt from the same material. For a genuinely blank start, the FAQ's own sentence is the checklist — 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."
- To reset one thing: Option 1. Correct it in place; nothing else needs to move.
How to see ChatGPT's memory before you delete it
Before you delete anything, look. OpenAI gives you three views, none of them complete, and knowing which is which saves you deleting the wrong thing.
- The memory summary — Settings > Personalization > Memory. "You can see the information that ChatGPT remembers in the memory summary," with a "last updated" stamp at the top. In the 4 June 2026 announcement's words, from there "you can quickly glean the highlights of what ChatGPT knows about you, add or update information about yourself, and provide instructions on what topics ChatGPT should bring up and when."
- The sources behind one answer — the book icon under a response. OpenAI's 5 May 2026 release note lists what it can show: "relevant saved memories, past chats, and custom instructions," and for Plus and Pro accounts also "files in their library and referenced emails from a connected Gmail account." One detail worth knowing if you share transcripts with clients, from the Memory FAQ: "Memory sources are not included in chats you share."
- Asking — OpenAI's fallback for anything not on the page: "If you want to know if ChatGPT has remembered something, just ask in chat," or, per the announcement, "if you want to drill down into a particular area to learn more, just chat with the model."
The caveat sits in OpenAI's own text. The summary "will not include everything that ChatGPT remembers based on your chats"; the memory underneath it "may be broader than what can be shown as individual items in a summary"; and "some details may not appear in the summary, including when ChatGPT determines they are less relevant or not appropriate to show in this view." The sources view, likewise, "may not show every factor or source that shaped a response." On the legacy system you get the more literal answer — a list of entries — which is one reason some people revert to it.
What deleting memory covers, and what it doesn't
Deleting memory deletes memory, and nothing else. Your conversations stay where they are — OpenAI is explicit that Delete and turn off memory "does not delete your past chats" — and, as in Option 2, removing an entry doesn't retract the times that fact was already spoken aloud in your history.
That distinction matters if you came here because something confidential went into a chat, rather than because something stale came out of one. In that case memory is one stop of several, each on a different screen, and OpenAI documents the whole sweep separately. We've walked it through in what deleting a ChatGPT conversation actually removes — do that one properly, then come back here for the memory half.
ChatGPT won't delete memory? Why delete can look like it did nothing
"Delete not working" is one of the most common follow-up searches on this topic. Most of the time nothing is broken — OpenAI's own documentation explains the behaviour. Take these in order before assuming a fault.
- You deleted the visible line, not the source. OpenAI's Memory FAQ: "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." A fact removed from the summary can be re-synthesised from a chat that still mentions it. This is the usual version of "it came back," and it is documented behaviour rather than a bug.
- You turned memory off and back on. "If you turn memory back on later, ChatGPT may create new memories from chats that remain in your chat history, including older chats."
- It was never on the page you edited. The summary "will not include everything that ChatGPT remembers." So "I deleted it and it still knows" can mean the detail lived in the broader synthesis, not on the summary. OpenAI's check is to ask in chat; its fix, if the answer is yes, is the correction box or the sources menu (Option 1).
- You're on the phone. Delete and turn off memory "is available on the web and rolling out on mobile soon," and OpenAI's 4 June 2026 note adds that to get the new memory on iOS or Android you should "update your ChatGPT app to the latest version." Try the web app before concluding the control is gone.
- Your account is on the other system. The rollout is still uneven across "plans and countries" (12 June 2026 note). If you're looking at a list of entries, the controls are the legacy ones — delete individually or "clear all" — and the three-dot Delete and turn off memory won't be there.
- You expected the log to vanish too. For the legacy list, OpenAI says: "We may retain a log of deleted Saved Memories for up to 30 days for safety and debugging purposes." That's a retention log — OpenAI does not say it's used in responses — but it's the honest answer to "is it really gone the second I tap delete."
- It was deprioritised, not deleted (legacy list). The same 15 October 2025 note says you "can see which memories are currently top of mind and choose to prioritize or deprioritize any specific memory yourself in settings," and can "view and restore prior versions of saved memories." An entry that's still listed but out of the top-of-mind set is a documented state, not a failed delete — and a restore is a documented way for an entry to come back.
Then there are the cases OpenAI doesn't document. We can't verify any of these; they're here because they are what people typing "ChatGPT won't delete memory" are describing, and they're labelled as what they are.
- Reported by a user on the OpenAI community, 1 August 2026: after removing personal details from the personalization settings and asking in chat to erase everything, the same details reappeared in new chats when the model was asked to audit its memory. The thread had no replies and no OpenAI response when we read it on 8 September 2026.
- Reported on Reddit and summarised by PiunikaWeb on 1 October 2025 (legacy saved-memories list): the delete button appearing to do nothing when the list was nearly full, across reinstalls and switching between iOS and the web. Some found that asking ChatGPT in chat to forget a specific memory worked; others found the entry still there afterwards.
- Reported by a user on the OpenAI community, April 2025 (legacy list): new memories wouldn't save even after deleting entries; their own fix was finding and deleting one very large entry, which freed the room the smaller deletions hadn't.
If you're in one of those: try the correction route from inside the conversation, try the web app, and if the problem is confidentiality rather than clutter, do the full sweep in the deleted-chats guide rather than retrying the memory page.
Why you're actually here
Almost nobody searches this to tidy a settings screen. Two things send people looking, and they need different answers.
Something wrong got repeated back to you. A CFO who left. A cadence that changed in March. A deal structure from the version that didn't get signed. The fix is Option 1 — correct it where you caught it — and the lesson is that a stale memory doesn't go quiet, it keeps steering answers with exactly the same confidence as a current one.
Or it filled up, and you have to choose what to lose. This is the harder one, and it's worth being precise about what OpenAI actually publishes. For the legacy saved-memories list, OpenAI's release notes (15 October 2025) do name a "memory full" state: automatic management moves less important entries "to the background" and, in OpenAI's words, "helps avoid" it — not that it can't happen — and you can turn that management off at any time, in which case you're back to pruning the list yourself. For the current memory summary, the Memory FAQ (read 8 September 2026) documents no entry limit, no capacity figure, and no "memory is full" state — it is silent on all three, and silence is not a denial. On size, OpenAI has said two things, both dated 4 June 2026: the announcement says the new system made it possible "to increase memory capacity for Plus and Pro users," and the release note for the same day puts it as "twice as much memory capacity." A multiplier, with no absolute figure on either side. Any specific ceiling you find quoted in a listicle is not on any of the three OpenAI pages this guide checked, so your own Memory screen is the only authority for your account.
What OpenAI does document is a difference in upkeep. The new system, it says, "updates memories automatically with ChatGPT keeping track of the details it determines are most important," where the previous one "relied on users to manually manage updates." That changes who does the pruning. It doesn't change the thing that made pruning feel bad.
The part that makes this unpleasant
Deleting a memory is destructive, quiet, and hard to review — three properties that are fine individually and miserable together.
You can't see the whole of it. OpenAI says the memory summary "will not include everything that ChatGPT remembers based on your chats," that some details are withheld "when ChatGPT determines they are less relevant or not appropriate to show in this view," and that the per-answer sources view "may not show every factor or source that shaped a response." The documented way to check whether one specific fact is in there is to ask in chat — which is a fine tool for a question you already know to ask, and no help at all for the ones you don't.
You can't check a memory against where it came from. A saved memory is a distillation of a conversation, not a pointer to a document. When you're deciding whether to delete "the client prefers monthly invoicing," there's nothing to open and verify — no email, no signed schedule, no note from the call. So you're not really reviewing; you're guessing, then deleting more than you meant to because deleting broadly is faster than adjudicating each line.
And underneath both: what's in there came from talking to ChatGPT, from files you uploaded, and from any account you connected to it wholesale. The four emails before Thursday's call, the redline attached three replies deep, the voice memo you dictated walking back to the car, the PDF a client sent last month — unless you handed over the whole mailbox or drive, none of it is in memory to begin with. You are carefully pruning a record of one conversation stream while the material your week actually runs on sits somewhere else entirely.
A memory you don't have to ration
If the pruning is the symptom, the shape worth wanting is different in three specific ways: bounded by what you deliberately sent rather than by a quota; listed in full and individually deletable; and attached to the original, so deleting is a decision rather than a gamble.
Equerry is a personal assistant built that way. Nothing reaches it unless you send it, which is what makes both halves work — what's in there is what you chose to put there, and it doesn't get evicted to make room for something newer.
- You decide what goes in. Put your own Equerry address on a client thread — CC it going forward, or forward what's already there. Send a PDF, a contract, or a screenshot through the share sheet. Type a note or dictate a voice memo. That's the whole boundary: it knows what you sent it, and nothing else in your inbox travels along.
- You can read the whole list. Everything Equerry has remembered is set out in one place, each item with the source beside it — not a summary that captures "the most important details," but the actual inventory. Any note also expands to show what was remembered from it.
- Deleting is reviewable. Before you remove a fact you can open the email, note, or file it came from and see whether it's still true. Tap to remove it, and it's out of every answer from then on.
- Answers carry their receipts. Ask something and the answer arrives with a source chip. Tap it and the original opens — you're reading the client's actual sentence, not weighing whether to trust a paraphrase of it.
- And it gets ahead of you. Ask what you're walking into before a call and the brief is assembled from everything that's landed since you last looked — including the reply that arrived while you were in the previous meeting.
None of this is a reason to stop using ChatGPT. Delete what's stale using the steps above, and it'll keep doing what it's genuinely good at — drafting, thinking out loud, working a problem. Just stop asking it to be the record of your client work, because it was never in the room for most of it. Equerry is available on the App Store.
Related reading: what ChatGPT does and doesn't keep between chats, what deleting a conversation actually removes, what the memory summary is and why it rewrites itself, and how Equerry holds the working record your day runs on.
Checked on 8 September 2026 against three OpenAI sources: the Memory FAQ help-center article (its "Updated" stamp put its last edit at the end of August 2026 when read on 8 September 2026); OpenAI's 4 June 2026 post Dreaming: Better memory for a more helpful ChatGPT; and the ChatGPT — Release Notes help-center article, for the entries dated 15 October 2025, 5 May 2026, 4 June 2026, 12 June 2026 and 27 August 2026. Every quotation above is OpenAI's own wording, attributed to the page it came from. Where those pages are silent — the size of the memory bound on any plan, and what a full memory summary does — this guide says so rather than filling the gap. User reports are labelled as reports, with their dates; neither we nor OpenAI have verified them. Control labels, rollout states, and which memory system your account is on all change, so treat the paths as a guide and your own Settings screen as the authority. First published 28 August 2026; retitled and re-verified 8 September 2026.
Frequently asked questions
What happens when ChatGPT memory is full?
It depends which memory you're on, and OpenAI documents the two unevenly. For the legacy saved-memories list, OpenAI's own release notes (entry dated 15 October 2025) acknowledge a "memory full" state: automatic memory management was introduced to keep "the most relevant details prioritized" while "moving less important ones to the background," which "helps prevent saved memories from reaching capacity and helps avoid 'memory full' state in your ChatGPT account." OpenAI says this "helps avoid" the full state — not that it can't happen — and the same entry notes you "can turn off automatic memory management at any time," in which case you are back to managing the list yourself. Users have also reported the full state persisting; see the section on delete not working. For the current memory summary, neither the Memory FAQ (read 8 September 2026) nor the 4 June 2026 announcement describes a full state, an entry limit, or a capacity figure — silence, which is not the same as a denial. What OpenAI has said about size: the announcement says the new system made it possible "to increase memory capacity for Plus and Pro users," and the release note for the same day puts it as "twice as much memory capacity" — a multiplier, with no absolute figure on either side of it. Any specific number quoted elsewhere isn't from OpenAI; treat your own Memory screen as the authority for your account.
If I delete a memory, does it disappear from my past chats too?
No. OpenAI says of a saved memory that "you can delete that memory anytime, but it won't remove mentions from past chats. To clear that, you will need to delete the specific chat where it was shared." Deleting memory and deleting history are two jobs, and doing one never does the other. The reverse direction — whether deleting a chat removes what ChatGPT already took from it — is covered in does ChatGPT remember deleted chats. If your reason for deleting is that something confidential went in rather than that something stale came out, memory is one stop of several: OpenAI's own checklist is 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." (Memory FAQ, read 8 September 2026.)
How do I see what ChatGPT remembers before I delete it?
Not everything, and OpenAI says so in its own words. Of the memory summary: it "should capture the most important details" but "will not include everything that ChatGPT remembers based on your chats," and "some details may not appear in the summary, including when ChatGPT determines they are less relevant or not appropriate to show in this view." The documented way to check for a specific fact is to ask: "If you want to know if ChatGPT has remembered something, just ask in chat." There is also a per-answer view — the book icon under a response lists the sources that personalised it, and tapping a memory explains why it was used — though OpenAI notes those "may not show every factor or source that shaped a response." The legacy saved-memories system is the more literal of the two: a list of discrete entries you can read and delete one by one. That gap between what you can see and what is held is the thing to weigh before you start deleting. (Memory FAQ, read 8 September 2026.)
How do I reset ChatGPT memory completely?
There is no separate reset button in OpenAI's documentation; a reset is a delete followed by turning memory back on. On the current system that means Delete and turn off memory from the three-dot menu on the memory summary page, then re-enabling memory. Read what OpenAI says happens next before you count on a blank slate: "If you turn memory back on later, ChatGPT may create new memories from chats that remain in your chat history, including older chats." A reset that keeps your chats is a reset back towards the same memory, rebuilt from the same material. For a genuinely clean start, OpenAI's own sentence is the checklist — delete "every source where it appears, including past chats, archived chats, files, the memory summary, and disconnect any connected apps." And if what you actually want reset is one wrong fact, you don't need any of this: type the correction into the box at the bottom of the memory summary, or highlight the line and fix it. (Memory FAQ, read 8 September 2026.)
Should I turn ChatGPT memory off?
Usually not as a first move — it's a heavier action than most people intend, because on the current system the same control does both jobs at once. Try the smaller instruments first: correct a specific line by highlighting it in the memory summary, type the correction into the box at the bottom, or delete the individual entry if you're on the legacy list. Turn memory off outright when you want a clean break, or when the account is shared or being handed over. For a single conversation you'd rather leave no trace on, OpenAI's answer is Temporary Chats, which by default "do not use existing memories or create new memories" — that's the surgical option, and it doesn't cost you the memory you've already built. Two things to know: since 27 August 2026 you can opt a temporary chat into personalisation when you create it (it then uses your memories but still creates none), and saving a temporary chat turns it into ordinary history that memory can draw on — so for a no-trace conversation, leave both alone. And before you flip memory back on: OpenAI says "ChatGPT may create new memories from chats that remain in your chat history, including older chats."
Last updated 2026-09-08.
All product names, logos, and brands are property of their respective owners. Equerry is not affiliated with, endorsed by, or sponsored by any other product mentioned; comparisons are provided for informational purposes.