What ‘memory’ means in ChatGPT, Claude, Gemini and Grok
All four assistants can carry information between conversations, but they draw different lines around what is remembered, where it comes from and how users control it. The most useful comparison is not which has memory, but what each feature actually does.
2026-11-06 · The Konshus Team · Sources checked before publishing
Want more news like this?
- One short email a week with the biggest changes in AI memory — ChatGPT, Claude, Gemini, new companies and research.
- Every story is checked against its sources before it goes out.
- No spam: just the weekly news, no sales emails, and we never sell or share your address.

What changed: ‘memory’ no longer means one thing
An assistant that remembers your preferences sounds straightforward. In practice, the word can describe a saved note, context selected from earlier chats, information found by searching an old conversation, or personalization drawn from another app. Those are different capabilities with different privacy implications.
The four products in this comparison illustrate the problem. ChatGPT distinguishes saved memories from personalization based on chat history. Claude distinguishes editable memory from searching previous chats, and offers memory for individual projects. Gemini can personalize from past Gemini chats and, for eligible users, connected Google apps. xAI describes Grok as remembering information across chats, but provides less detail about its consumer memory in the official material reviewed here. OpenAI · Claude · Gemini · Grok
This is a state-of-play comparison, not a ranking or an announcement that every account has the same features. The practical question is what information an assistant might carry into your next conversation—and what you have to do if you want it to stop.
ChatGPT: saved details and history-based personalization
OpenAI describes two related features. Saved memories are details ChatGPT keeps for future use, whether you ask it to remember them or it saves something it considers useful. Reference chat history lets ChatGPT use relevant information from earlier conversations; that information can change as the system updates what it considers worth remembering. A saved memory and an old chat are therefore not the same record. OpenAI’s memory FAQ
That distinction matters when cleaning up. OpenAI says saved memories are stored separately from chats: deleting a conversation does not automatically delete a memory saved from it. Users can ask ChatGPT to forget a detail or remove saved memories through settings. Turning off Reference chat history schedules information remembered from past chats for deletion within 30 days, but does not itself delete saved memories or the original conversations. OpenAI’s memory FAQ
For a conversation that should neither use nor update memory, OpenAI offers Temporary Chat. That is a different choice from having an ordinary chat and later deleting it. If a remembered detail worries you, the useful question is not merely, ‘Did I delete the chat?’ It is also, ‘Is there a saved memory, and is chat-history reference on?’ OpenAI’s memory FAQ
Claude: editable entries, project boundaries and separate search
Anthropic’s help material describes Claude’s memory as individual entries organized by topic, which Claude can read and update during conversations. Its release notes say this replaced an earlier daily-summary approach. Users can inspect, edit and delete remembered topics in Settings > Memory. Claude also supports separate project memory, rather than treating every project as one undifferentiated memory space. Claude memory guide · Release notes
There is an important distinction here: Anthropic documents searching previous chats separately from memory, and describes chat search for paid plans. An assistant carrying forward a remembered preference is not necessarily retrieving the old exchange where you said it. Conversely, finding an old exchange does not mean its contents have become a standing memory. Claude memory guide
Claude also offers different ways to set a boundary. Anthropic says incognito chats are excluded from both chat history and memory. It documents a separate per-chat memory control that can prevent memory use or updates while leaving the conversation in chat history. That difference is worth checking before a sensitive discussion: keeping a record of a conversation and letting it inform future answers are separate decisions. Claude incognito guide · Claude memory guide
Gemini: past chats, plus an optional route into Google apps
Google documents memory of past Gemini chats as one way to personalize responses. Separately, its Personal Intelligence features can, for eligible users who connect apps, draw on data from selected Google apps. That makes the source of personalization an especially important question for Gemini: was the answer informed by something you told Gemini, or by information available through a connected app? Google on past-chat memory · Google on connected apps
Google says users can manage memory in Personal Intelligence settings, find and delete past chats in Gemini Apps Activity, and choose which supported apps are connected for personalization. It also documents a chat-level way to turn off personalization from past chats, instructions and connected apps. These controls address different sources of context; deleting a chat is not a substitute for reviewing an app connection if that connection is the concern. Google on past-chat memory · Google on connected apps
Availability is another qualifier. Google describes Personal Intelligence as rolling out gradually, and says past-chat memory does not apply in some experiences, including Gems and Live chats. A description of what Gemini can use should not be read as a promise that it will use that context in every conversation. Google on connected apps · Google on past-chat memory
Grok: cross-chat memory, with a thinner public account
xAI’s Grok product page describes memory across chats, including preferences and past conversations. In its memory announcement, xAI described controls for viewing referenced memories and choosing what to forget, along with a ‘Personalize with Memories’ setting under Data Controls. Those are meaningful controls to look for, even though the available official descriptions give less detail about the underlying consumer memory than the help pages for the other three products. xAI’s Grok page · Grok’s announcement
The announcement called the feature a beta and described limits on where it was initially available. Those launch-era limits are not reliable evidence of exactly where Grok memory is available now. Nor do the sources reviewed here establish a detailed, current account of precisely how Grok forms or retains each memory. The fair comparison is narrower: xAI says Grok offers cross-chat personalization and ways to inspect or forget memories; questions about an individual account require checking its current controls. Grok’s announcement · xAI’s Grok page
Why the distinctions matter
These products do not merely put different names on an identical feature. ChatGPT’s saved detail can outlive deletion of the chat it came from. Claude makes memory and old-chat search distinct, and provides project-specific boundaries. Gemini can extend personalization beyond the chat archive when supported Google apps are connected. For Grok, the available official material establishes the broad capability and some controls, but leaves more questions about its precise scope. OpenAI · Claude · Gemini · Grok
A useful check before relying on any of them is to ask three questions. What is the source? It could be a saved entry, earlier chats or a connected app. What is the boundary? A setting might apply to one chat, a project, a feature or an account. What does deletion change? Removing a conversation and removing a memory may require different actions.
The documentation does not settle which assistant remembers most accurately, selects the most useful details, or makes the fewest mistakes. It explains features and controls, not a shared test of memory quality. For now, ‘has memory’ is the start of the comparison—not its conclusion.
Our take (opinion)
Memory is becoming less like a single notebook and more like a set of permissions for reusing context. That can make an assistant substantially more helpful: OpenAI’s separation of saved details and chat-history reference gives users two different levers; Anthropic’s distinction between memory and search clarifies what Claude is doing; and Google’s app controls matter precisely because personalization can reach beyond conversations.
But good controls have to match users’ expectations. Someone who deletes a chat may reasonably think they have removed what an assistant learned from it, even when a separate saved memory remains. Someone who turns off chat memory may not immediately think to check connected apps. Those are not reasons to reject personalization. They are reasons to judge it by how clearly a product explains the source, scope and removal of remembered information. Grok deserves the same standard—and more detailed current consumer documentation would make that judgment easier. None of these sources establishes a winner on memory quality.
Questions
Does deleting a ChatGPT conversation delete a saved memory from it?
No. OpenAI says saved memories are stored separately from chats. Check and remove the saved memory separately if that is what you want to delete. https://help.openai.com/en/articles/8590148-memory-faq
Is Claude’s memory the same as searching my old chats?
No. Anthropic documents memory and previous-chat search as distinct capabilities; it describes chat search for paid plans. https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-context
Can Gemini personalize without relying only on Gemini conversations?
Yes, for eligible users who enable supported connections. Google says Personal Intelligence can use information from connected Google apps as well as personalization from past Gemini chats. https://support.google.com/gemini/answer/16598406
Can I see and remove what Grok remembers?
xAI has described controls to inspect referenced memories and choose what to forget. Check your current Grok account’s Data Controls for the options available to you. https://x.com/grok/status/1912670183472476653
Sources
- Memory FAQ — OpenAI Help Center
- Use Claude’s chat search and memory to build on previous context — Claude Help Center
- Release notes — Claude Help Center
- Use incognito chats — Claude Help Center
- Get personalization with memory of your past Gemini chats — Google
- Connect your Google apps to personalize your Gemini experience — Google
- Grok — xAI
- Grok memory announcement — Grok on X
The AI memory news, once a week.
If you're interested in AI memory, want your own AI to remember you better, or just want to stay up to date, this is the newsletter for you. Every week we comb through the news for the big shifts in AI memory — new companies, new features from ChatGPT, Claude and Gemini, new research and new ideas — and sum up what matters in one short email.
For people who use AI every day, builders and founders, and anyone curious where AI memory is going.
About Konshus: a private memory Vault that works across ChatGPT, Claude, Gemini and more. Learn more.