Claude, ChatGPT, and Gemini all let you view, edit, or delete what they remember about you, but the three companies’ own help pages describe three different systems. Claude saves memory as separate topics you can open and edit individually, and a deleted chat does not erase what it already taught Claude. ChatGPT’s own FAQ tells you to just ask the assistant what it remembers, without describing a management screen. Gemini splits the feature in two: things you explicitly asked it to save, managed under a setting called Personal context, and a separate, still-rolling-out ability to reference past chats.
None of this is obvious from using the products day to day, because the memory toggle in each app looks almost identical: a switch, a list of items, a delete icon. The differences only show up when something goes wrong — you delete a conversation and expect the memory from it to disappear, or you want to know exactly what an enterprise admin can see about your account. This post goes through each provider’s own documentation line by line and lines up what actually happens.
What each company says memory actually stores
Anthropic’s support article for Claude describes memory as a running set of notes, not a transcript. Anthropic’s own help article states that “Claude saves memory as a set of individual topics as you chat, rather than summarizing conversations after they end.” Those topics cover ordinary context — your role, your projects, recurring preferences — and the article says Claude will not store sensitive topics unless you opt in.
OpenAI’s help center describes ChatGPT’s memory in looser terms. The OpenAI memory FAQ says the model “can now carry what it learns between chats” and gives an example of teaching it a dietary preference so future recipe suggestions respect it. It does not define a unit of storage the way Claude’s “topic” does, and it does not separate this from the related, older “reference chat history” setting that lets ChatGPT pull context from past conversations directly.
Google’s support page for Gemini Apps splits the feature into two named settings. Google’s support article describes “saved info” — things you explicitly tell Gemini to remember about your life, work, or preferences — which lives under a page called Personal context, and a separate, still-expanding ability for Gemini to look at your past chats before answering. Google is explicit that the second half is a partial rollout: the article says the company is “gradually releasing the ability to reference past chats in the Gemini mobile app.”
How you actually view, edit, or delete it
This is where the three products diverge the most, and it is the part most comparison posts skip because it requires reading the support pages instead of the marketing pages.
Claude gives memory its own settings screen. Go to Settings, then Your Claude, then “Manage” next to Your memory, and “everything Claude remembers is listed under Topics,” according to Anthropic’s article. You can open any topic and edit or delete it individually, or tell Claude in a normal chat message to remember, change, or forget something specific.
ChatGPT has no equivalent screen described in OpenAI’s own FAQ. The article’s actual guidance for finding out what it remembers is to ask it directly in a chat — “to understand what ChatGPT remembers just ask it.” The same article says you can “delete specific or all memories” and reset the feature, but it does not walk through where that control lives in the interface, and it does not mention an edit option at all, only deletion.
Gemini’s path is a named settings page: gemini.google.com, then Settings & help, then Personal context (or the direct URL gemini.google.com/saved-info). Google’s support article says that from there you find the item, click More, and choose Edit or Delete. The same page also carries the toggle for turning “Info you asked Gemini to Save” on or off entirely.
| Question | Claude | ChatGPT | Gemini |
|---|---|---|---|
| Storage unit, per the provider’s own docs | Individual “topics,” written as you chat | Not defined; described only as “what it learns” | “Saved info” entries plus (rolling out) past-chat references |
| Where to view everything stored | Settings → Your Claude → Manage → Topics | Not documented; FAQ says to ask the model directly | Personal context page (gemini.google.com/saved-info) |
| Can you edit one entry without deleting it? | Yes, open and edit a topic | Not mentioned in the FAQ | Yes, via the item’s “More” menu |
| Full account-wide off switch | Yes, toggle off “Memory from chats”; a separate “Reset memory” permanently deletes everything | Yes, a setting the FAQ says exists but doesn’t name precisely | Yes, toggle next to “Info you asked Gemini to Save” |
| Does deleting a chat delete the memory it created? | No — explicitly stated | Not addressed in the FAQ | Not addressed for saved info in this article |
The gotcha: deleting a chat does not delete what it taught the model
Of the three, only Anthropic’s documentation directly answers a question most users assume works the obvious way: if you delete the conversation that created a memory, does the memory go with it? Claude’s support article is unambiguous: “When a conversation is deleted, including under a retention period, memory entries generated from it aren’t removed.” The memory and the chat that produced it are stored separately, so removing one does not touch the other — you have to go back into Topics and delete the memory entry on its own.
Anthropic’s article also describes two different levels of “turning it off.” Pausing memory means, in the article’s words, Claude “won’t use memory or make new memories” — existing topics stay on file but stop being used. Resetting memory is the destructive option: it “permanently deletes all memories including project memories.” For a single conversation, there’s a lighter switch in the “+” menu before you send the first message; turning it off there means “Claude won’t use what it remembers about you in that chat, and nothing from the chat is saved to memory,” without touching anything you’ve already stored.
Neither OpenAI’s nor Google’s support articles make an equivalent statement about chat deletion and memory independence, one way or the other — which is itself useful information if you’re trying to decide what to trust by default. Absence of a stated guarantee is not the same as a guarantee that deleting a chat also clears its memory; if the detail matters to you on those platforms, the safer assumption is that it behaves like Claude’s documented case until a provider says otherwise.
What this means if you manage a team account
Anthropic’s article adds a detail that matters for Team and Enterprise plans specifically: “Owners can’t view or edit a user’s individual memories.” Memory on those plans is private to the person who created it, not something an administrator can audit from the admin console. The same article notes that all memory data is still included in data exports, and that project memory — memory tied to a specific Claude Project — is deleted automatically when the project itself is deleted, rather than needing a separate cleanup step.
If your organization is choosing between these tools partly on data-handling grounds, that single sentence about admin visibility is worth more than any marketing comparison: it tells you memory is closer to a personal notebook than a shared company asset, at least on Claude’s current implementation.
Which provider’s documentation actually answers your questions
Reading all three support pages side by side, the gap isn’t in what the products can do — it’s in what each company is willing to put in writing. Anthropic’s article reads like a reference manual: it names the exact settings path, defines the two levels of turning memory off, and states the chat-deletion behavior as a direct fact rather than leaving it implied. Google’s article does the same for the parts it covers, but explicitly flags the past-chat-reference half of the feature as incomplete, which is a more honest signal than most companies give about a feature still rolling out.
OpenAI’s FAQ is the odd one out, and not because ChatGPT’s memory necessarily behaves worse — there’s no public documentation saying it does — but because the help article treats “ask the model” as an acceptable substitute for a settings page. That’s a reasonable answer for a casual user who wants a quick summary, but it’s a poor one for anyone who needs to audit what’s stored, prove it was deleted, or explain the behavior to a compliance team. If your use case depends on verifiable memory handling rather than a conversational description of it, that distinction should factor into which assistant you route sensitive, recurring work through.
FAQ
If I turn memory off, do my existing memories disappear?
Not on Claude or Gemini, based on their own documentation: both describe pausing or disabling the feature as separate from deleting what’s already stored. Claude calls the destructive option “Reset memory” specifically so it isn’t confused with the pause toggle. OpenAI’s FAQ doesn’t separate the two actions clearly enough to answer this from the document alone.
Does using a temporary or incognito-style chat avoid memory entirely?
On all three platforms, yes — Claude’s per-chat memory toggle, ChatGPT’s Temporary Chat, and Gemini’s equivalent are each described as keeping that single conversation out of memory, without changing anything already stored from earlier chats.
For more on how these providers handle other kinds of persistent context, see our breakdowns of Claude’s Files API for reusing uploaded documents, how context caching pricing compares across Claude, GPT, and Gemini, and Claude’s Citations API versus manual quote-prompting.
Sources: Anthropic, “How does Claude’s memory work?”; OpenAI, “What is Memory?”; Google, “Save info and reference past chats in Gemini Apps.”
Featured image: “EPROM Microchip SuperMacro” by Zephyris, via Wikimedia Commons, licensed under CC BY-SA 3.0.
