How Claude, ChatGPT and Gemini Remember You

Short answer: all three assistants build a lasting profile from your chat history, just on different rhythms. Claude synthesizes conversations every 24 hours, ChatGPT edits entries in the background via a feature called Dreaming, and Gemini keeps one account-wide memory under the name Personal Intelligence. All three can be turned off, but none ship off by default.

How does Claude remember you?

Since March 2, 2026, Claude has shipped automatic Chat Memory on by default across every plan, synthesizing your conversations every 24 hours and carrying that context into the next chat. Two independent toggles control this: "search and reference chats" (look up past conversations on demand) and "generate memory from chats" (build persistent memory entries). Neither depends on the other — you can let Claude search old conversations without letting it build a standing profile, or the reverse.

Claude's memory is also now shared between Chat and Cowork, so a preference it learns in one product carries over to the other. We covered how that unified memory works in Claude Now Shares Memory Across Chat and Cowork.

How does ChatGPT remember you?

ChatGPT's memory lives under Settings → Personalization → Memory, and it saves facts from your conversations automatically. A feature called Dreaming, announced on June 4, 2026, curates that memory in the background: it might update an entry like "planning a trip to Singapore" to "went to Singapore" once the trip has passed. So ChatGPT's memory isn't a static notebook — it's a profile that keeps rewriting itself over time.

How does Gemini remember you?

Google rebranded and expanded Gemini's memory feature on January 14, 2026, naming it Personal Intelligence. The setting lives under Settings → Personal Intelligence → Memory, and it operates at the account level — applying across every Gemini conversation tied to your account, not just one chat. That gives you consistent context across chats, but it also means a broader pool of data accumulates in one place.

How do the three compare side by side?

All three assistants score "yes" on remembering across sessions; the shared weakness is that none of them can read another's memory. Each vendor's memory is a retention feature for its own product, not a portable record you carry between assistants.

How do you view, edit, or delete your memory?

The logic is similar across all three: open the relevant settings page, review saved entries one at a time and delete what you don't want, or wipe the whole memory at once. Claude requires you to work through two separate toggles, since search and memory generation are independent — turning one off doesn't automatically turn off the other. ChatGPT and Gemini each have a single master toggle, but both also let you delete individual entries by hand.

One caveat: for users in the EU and UK, memory features are still restricted or fully unavailable under GDPR as of mid-2026. If you're in that region, you might not see the feature at all — not because you opted out, but because of a regulatory restriction.

Here's what deleting a single entry versus wiping everything looks like across the three:

How do you have a one-off chat without touching persistent memory?

All three ship a "temporary chat" mode that skips persistent memory entirely: in Claude and ChatGPT, this mode never writes to memory and generally doesn't surface in history either; Gemini's equivalent temporary-chat option sends nothing to your account-level Personal Intelligence. Rather than switching off persistent memory altogether for a sensitive conversation, using this mode gives you a one-off privacy layer without touching your main setting.

Does data in memory get used to train the models?

That's a separate but related question from memory itself, and the answer varies by provider. ChatGPT and Gemini train on your conversations by default on free and personal paid plans; turning that off is something you have to do yourself. Claude works the opposite way: Anthropic only trains on consumer chats if you actively opt in. API access and Enterprise seats are excluded from training by default across all three providers.

One detail worth knowing: revoking training consent only works going forward. Once a model has already trained on your data, that influence can't be retroactively removed from its weights — opting out only affects future training cycles.

When should you keep memory turned off?

The convenience is real, but the cost usually goes unmentioned: every memory feature you leave on slowly turns into a profile about you that you don't see and can't fully edit. If you're working from a shared device, mixing work and personal chats on the same account, or discussing something sensitive — health, legal, financial — turning memory off for that session is a reasonable default. It's cheaper than asking "why does it remember this" after the fact.

My take: memory features get marketed as "an assistant that knows you better," but the honest framing is that every remembered detail leaves a persistent trace on that provider's servers. Use it, but make checking what's actually been saved a monthly habit rather than a one-time setup step. For a broader look at this trade-off, see Protect Your Privacy in the Age of AI Assistants.

Frequently Asked Questions

Can I transfer my memory between Claude, ChatGPT and Gemini?

No, all three operate independently, and none can read another's memory. Switching from one assistant to another means re-explaining your preferences; some independent third-party tools try to offer a unified memory layer, but that isn't an official feature from any of the providers.

Can I fully delete an AI assistant's memory of me?

Yes — all three let you bulk-delete every memory entry from the relevant settings page. Deletion is generally retroactive, wiping the profile built up to that point, but new conversations start accumulating again as long as memory stays turned on.

Can the assistant still see past conversations with memory turned off?

Usually not, but Claude specifically splits this in two: "search and reference chats" can stay off while "generate memory from chats" stays on, or vice versa. The other assistants use a single master toggle that controls all access to your history at once.

Why is memory restricted in the EU?

GDPR's requirements around processing personal data and obtaining consent add extra compliance burden for memory features that build an automatic profile from conversations. Providers haven't fully met that burden yet, so as of mid-2026 the feature can be restricted or disabled entirely in that region.