ChatGPT memory is full: what's actually documented, and what to do

If details have stopped sticking, you want two things: how to clear it, and why it keeps happening. Here are the current steps — plus the part most articles get wrong, which is what OpenAI has actually published about the ceiling: one relative figure, covering two plans, and no absolute number anywhere.

First, the thing most articles get wrong

Search this question and you'll find confident numbers — a maximum entry count, a capacity that doubled, a figure that differs by plan. One of those is sourced and the rest aren't, so it's worth separating them carefully. OpenAI's Memory FAQ, read on 28 August 2026, documents the memory controls thoroughly and says nothing about an entry limit, a capacity figure, or a "memory is full" state. Its 4 June 2026 announcement contains exactly one capacity sentence — that efficiency work made it possible "to increase memory capacity for Plus and Pro users" — with no before, no after and no multiplier.

But the announcement isn't the only thing OpenAI published that day. Its ChatGPT release notes carry a separate entry dated 4 June 2026, headed "Memory that stays more up to date," and the rollout paragraph ends: "For Plus and Pro users, ChatGPT can also remember more useful context, with twice as much memory capacity." That is where the "doubled" claim legitimately comes from. Read it precisely, though, because it is narrower than the way it gets repeated: it is relative — twice their previous capacity, whose size OpenAI has never published — and it covers Plus and Pro only; nothing is said about Free or Go.

So the honest position is narrower than the internet's, and it splits in two. On size: a bound clearly exists — you don't double a capacity that was never bounded — but the only figure OpenAI publishes is that multiplier. No absolute figure is published, on any plan, so "twice as much" tells you the direction of a change and nothing about where the ceiling sits. On what happens when you reach it: that depends on which system you're on. The legacy saved-memories list has a documented "memory full" state, quoted in the FAQ below. For the current memory summary, OpenAI documents nothing of the kind. If new details have stopped sticking on the summary, that's a real thing to act on — it just isn't a documented state with a documented fix, and we'd rather tell you that than invent a number.

The two systems, and which one you're on

ChatGPT's memory has changed shape, and most of the advice online still describes the old version. The current system is a memory summary: a continually updated synthesis of context drawn from your chats, files, and connected apps, which refreshes on its own as you use it. The top of the summary shows when it last updated.

The older system was saved memories — a list of discrete entries you managed by hand. It still exists, reachable from Settings → Memory via the "saved memories" link beneath the memory summary. OpenAI's own account of why it moved on is worth reading, because it explains the design: the old list "often became stale and relied on users to manually manage updates," and entries could contradict one another.

One caveat that matters more than the capacity question: OpenAI says the summary "should capture the most important details" but "will not include everything that ChatGPT remembers based on your chats." What you can see is a view of the memory, not the memory itself.

How to view, correct, and clear it

Controls live in Settings → Personalization → Memory. On the current system:

• Correct one detail: type the change into the text box at the bottom of the memory summary, or highlight the passage you want fixed and correct it in place.
• Check whether something was remembered: ask in chat. OpenAI suggests this precisely because the summary doesn't show everything.
• See what shaped an answer: tap the book icon beneath a response to view the sources used — custom instructions, past chats, files, memories. OpenAI notes this "may not show every factor or source."
• Wipe it: the summary's three-dot menu has Delete and turn off memory. This does not delete your past chats, and turning memory back on later can rebuild memories from the chats still in your history.
• Keep a conversation out of memory entirely: use a Temporary Chat. Those neither use existing memories nor create new ones.

The step most people miss: removing something for good is a sweep, not a switch. OpenAI's instruction is that to fully delete something ChatGPT may know about you, you must delete every source it appears in — past chats, archived chats, files, the memory summary, and any connected apps that hold it. Clearing the summary alone leaves the sources it was synthesized from intact.

The limit underneath the limit

Suppose capacity were unlimited tomorrow. You'd still hit the constraint that actually costs you something: the reach of what it remembers. OpenAI says memory draws on your chats, your files, and any apps you connect — so it can go wider than the chat window, but only by you wiring up a whole account. Short of that, it never saw the email a client sent on Tuesday, the voice memo you dictated in the car, or this morning's meeting notes.

And that's the trade worth noticing. To get at the four emails that matter before a client call, you're asked to connect the entire mailbox — a standing, account-wide connection, to solve a specific and occasional problem. Pruning feels like a chore that never ends for the same reason: you're maintaining a store whose reach you can only widen in one large step.

What cross-source memory looks like instead

Equerry is a personal assistant built around the opposite shape. You forward or CC the emails that matter, drop in a PDF or a screenshot, save a note, or dictate a voice memo — and it answers across all of them together. Before a client call it briefs you from what you've shared, and it shows where an answer came from when it draws on your notes, emails or the web, so you can open the original rather than trust a summary.

It also inverts the visibility problem above. Every fact it has drawn out is listed in one place and removable on a tap, each note shows what was taken from it, and deleting a note removes what came from it. Memory you can audit in a minute is a different proposition from memory you're told is mostly accurate.

ChatGPT memory vs. cross-source memory

So: clear it, or move past it

If details have stopped sticking, the immediate move takes two minutes: open Settings → Personalization → Memory, correct the summary in place, or use Delete and turn off memory and start clean. That unblocks you today.

But if you keep hitting the wall — and if the things you most need remembered are the emails, documents, and meeting notes ChatGPT never sees — the ceiling isn't really the problem. The memory you want isn't a bigger chat-history store; it's durable, spans the sources you chose, and can be inspected and corrected. That's the gap cross-tool memory is meant to close.

ChatGPT memory behaviour and controls above were verified on 28 August 2026, and every capacity statement re-verified on 15 September 2026, against three OpenAI sources: the Memory FAQ help-center article; OpenAI's 4 June 2026 announcement of the updated memory system; and the ChatGPT — Release Notes help-center article (article 6825453), for the entries dated 15 October 2025 and 4 June 2026. help.openai.com refuses automated fetches with a 403, so the release notes were read on 15 September 2026 through a text-extraction reader proxy and confirmed separately in a browser session the same day. Quoted wording is OpenAI's. Where OpenAI does not document something — an absolute capacity figure on any plan, and what a full memory summary does — this page says so rather than filling the gap.

Last updated 2026-09-15.

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.