Claude vs ChatGPT vs Gemini Memory: How They Compare in 2026

Two years ago, AI memory meant a saved list of facts you told the model to remember. Today, all three major assistants have moved past that starting point, though in different directions. ChatGPT builds a rolling synthesized picture of you. Claude keeps individual categorized memory entries and updates them in real time. Gemini weaves memory into your Google account.

Three different design philosophies:

ChatGPT assumes you shouldn't have to think about memory.
Claude emphasizes transparency and control over what's stored.
Gemini makes memory part of your Google account.

The rest of this guide walks through each system with references to official documentation, then looks at where they meaningfully differ.

Claude vs ChatGPT vs Gemini Memory at a Glance

Every feature covered in this post, side by side:

Feature	ChatGPT	Claude	Gemini
Official name	Memory (Dreaming architecture)	Memory	Personal Intelligence (Memory toggle)
How it saves	Continuous background synthesis across chats	Individual categorized entries, updated in real time as you chat	Learns details and preferences from past chats
Default state	On	On, with opt-out toggles	On
Where to manage	Settings > Personalization > Memory	Settings > Memory	Settings > Personal Intelligence > Memory
Legacy option	Yes, revert to the saved memories list	Legacy experience under Settings > Capabilities (retained in Enterprise)	No
Project separation	Project-only memory inside Projects	Every Project has its own memory space	Account-wide memory, notebook-scoped context on paid tiers via Gemini Notebooks
Memory-free mode	Temporary Chat	Incognito chat	Temporary Chat
Import from other AIs	No	Memory from ChatGPT, Gemini, Copilot, or Grok	Memory and chat history from ChatGPT or Claude
Export memory	No dedicated export, not documented as part of the account data export	Manual copy from the memory panel, or through the general account data export	Via Google Takeout

How ChatGPT Memory Works

ChatGPT's Memory has gone through three phases. The current system, Dreaming, launched in June 2026. It replaced both the original saved-memories list and the Reference Chat History layer introduced in April 2025 with a single background process that works across every conversation.

Dreaming synthesizes continuously across your chats. It rewrites facts as they go stale and holds context at a high level rather than preserving exact wording. The Memory summary shows only part of the full synthesis.

A legacy fallback lets you switch back to the pre-2026 editable list for complete visibility. Under legacy mode, ChatGPT still captures memories automatically; what you give up is Dreaming's continuous cross-chat synthesis and the automatic rewrites that keep facts current.

Memory is on by default and managed under Settings > Personalization > Memory.

OpenAI's own example shows how Dreaming evolves. If you tell ChatGPT you're going to Singapore in July, Dreaming may later rewrite that memory to "you went to Singapore in July 2026." You never asked it to update the memory. The system recognized that the old version was no longer accurate.

Memory is designed to hold high-level context, not verbatim text. If you rely on exact templates or specific wording, don't expect Dreaming to preserve them.

Feature	Description
Memory (Dreaming)	Background synthesis across many conversations but picks up context you never asked to remember, and rewrites facts as they change
Memory summary	The visible view of what Dreaming has learned. Per OpenAI's docs, may be narrower than the full synthesis
Editing	Correct or dismiss details through the summary interface (Dreaming), or delete individual entries directly (legacy mode)
Legacy saved memories	A link under the Memory summary reverts you to the pre-2026 explicit list of facts, fully visible and editable
Project memory	Optional per-Project scoping, chosen when you create the Project
Temporary Chat	Skips memory reads and writes for the conversation
Training on memories	Free and Plus content, including memories, may be used to improve OpenAI's models unless you opt out under Data Controls. Business and Enterprise excluded by default
Full conversations	Not stored as memory. Deleting a chat does not delete the memories it produced

How Claude Memory Works

Where OpenAI optimized for you never thinking about memory, Anthropic took the opposite approach. Claude's Memory rolled out through late 2025 and early 2026, with a substantial revamp on July 10, 2026 that shifted the underlying model. The design consistently pushes control toward you. Everything is managed under Settings > Memory.

Claude builds memory as individual categorized entries. As you chat, Claude writes new entries and updates existing ones in real time, rather than synthesizing on a fixed schedule. Every entry is visible in your memory panel, editable, and deletable. You can also update memory by talking to Claude about what to focus on or ignore.

Two independent toggles configure what Claude does with your history:

Search and reference chats: look up past conversations on demand.
Generate memory from chats: build the persistent memory entries.

Neither one depends on the other, so you can turn on either, both, or neither.

Every Project gets its own memory space automatically. Incognito mode skips both memory writes and reads.

Note: On Team plans, Incognito chats still appear in standard data exports and follow your organization's retention policies. Private from memory doesn't mean private from your admin.

Feature	Description
Memory entries	Individual categorized entries created and updated in real time as you chat, every entry is visible and editable
Editing	Delete entries directly from the memory panel, or chat with Claude about what to focus on or ignore
Toggles	"Search and reference chats" (look up past conversations on demand) and "Generate memory from chats" (build persistent memory entries), independent of each other
Project memory	Every Project gets its own separate memory space automatically
Incognito chat	Skips history and memory entirely
Full conversations	Not stored as memory. Past chats remain reachable through the separate Search and reference chats feature

How Gemini Memory Works

Gemini's memory launched in August 2025 as "Personal context" on Gemini 2.5 Pro in select countries. On January 14, 2026, Google rebranded and expanded it as Personal Intelligence.

Unlike ChatGPT and Claude, Gemini's memory isn't a standalone system. It's part of your Google account. The Memory toggle lives at Settings & help > Personal Intelligence > Memory. Instructions for Gemini (called "Saved info" in some regions) also sit inside the Personal Intelligence section rather than in a separate menu.

Account-level memory applies across all your Gemini chats. Since April 2026, when Google merged NotebookLM into Gemini, you also have notebook-scoped context on paid tiers: each Gemini Notebook holds its own sources, past chats, and custom instructions, which don't leak into your account-wide personalization.

Whether the Memory toggle is available depends on your account type, age, and region.

If you delete a conversation from Keep Activity, it eventually stops contributing to personalization, though Google notes there may be a short delay before the change takes effect.

Feature	Description
Personal Intelligence	The account-level layer that learns key details and preferences from your past conversations
Memory toggle	Turn Memory on or off under Settings > Personal Intelligence > Memory
Instructions for Gemini	Standing instructions applied to every chat (labeled "Saved info" in some regions), sits in the same Personal Intelligence section
Gemini Notebooks	On paid tiers, per-notebook workspaces hold their own sources, past chats, and custom instructions (from the April 2026 NotebookLM merge)
Keep Activity	The renamed Gemini Apps Activity, the conversation record personalization draws on, where you review and delete chats
Temporary Chat	Conversations that skip history, Keep Activity, personalization, and model training, retained up to 72 hours to respond and process feedback

Key Differences Between ChatGPT, Claude, and Gemini Memory

The implementation details above boil down to a handful of practical differences.

How Often Each System Updates Its Memory

Tool	Update behavior
ChatGPT	Continuous background synthesis
Claude	Real-time, as you chat
Gemini	No published cadence

Claude reflects your corrections almost immediately. ChatGPT is likely to reflect changes in your next conversation. Gemini's behavior is harder to predict because Google doesn't publish how often Personal Intelligence updates.

How Much You Can See of Stored Memory

The most auditable mode of the three is a surprise: ChatGPT's legacy saved-memories list, if you're willing to give up Dreaming. Beyond that:

Claude: View, edit, and delete individual memory entries directly in the memory panel. You can also make changes by chatting with Claude.
Gemini: Instructions for Gemini (Saved info) is fully visible, but the automatic Personal Intelligence layer is not.
ChatGPT (Dreaming): Shows a Memory summary, which OpenAI says may be narrower than the full synthesis.

How Each Tool Handles Project-Level Memory

Tool	Project memory behavior
Claude	Automatic. Every Project gets its own memory space.
ChatGPT	Optional. Choose Project-only memory when creating a Project.
Gemini	Account-wide memory. Notebook-scoped context available on paid tiers via Gemini Notebooks.

Importing and Exporting Memory Between AI Tools

The asymmetry here is sharp:

Claude has the widest set of import sources. Its memory import prompt works with ChatGPT, Gemini, Copilot, or Grok. Export, however, is manual: copy the memory out of settings, ask Claude to write it out verbatim, or use the general account data export.
Gemini imports the most types of data. It supports memory handshakes from ChatGPT or Claude, and can import your ChatGPT chat history as a ZIP upload, up to 5 GB of text.
ChatGPT imports nothing. There is no built-in way to bring memory in from Claude or Gemini, and memories are not documented as part of the standard account data export.

Every import is a snapshot. Once it's finished, anything you teach the source assistant stays there until you run the import again.

Why AI Memory Doesn't Sync Between AI Tools

Every one of these systems is powerful inside its own product, and stops at the app's edge.

ChatGPT's memory lives in ChatGPT. Claude's memory lives in Claude. Gemini's memory lives in Gemini.

Each assistant ends up learning a different part of your work:

The tool you code with knows your stack.
The one you write with knows your voice.
The one in your inbox knows your schedule.
None of them can see the others.

Claude's and Gemini's import features narrow that gap once, but only once. They're snapshots. Whatever you teach the source assistant afterward stays there until you import again.

If you regularly work across more than one AI assistant, that means maintaining multiple versions of your context, or re-explaining yourself every time you switch.

There is no native fix. These systems were designed as products, not as parts of an ecosystem, and none of the companies has much incentive to change that.

A Shared Memory for ChatGPT, Claude, and Gemini

This is the problem MemoryPlugin was built to solve.

Instead of keeping separate memories inside each AI assistant, MemoryPlugin maintains a single shared memory that ChatGPT, Claude, and Gemini all read from and write to.

Tell Claude about your project's constraints, and ChatGPT and Gemini have them in your next conversation. Make a decision in ChatGPT today, then ask Gemini about it months later. Instead of maintaining three separate memories, you get one that follows you everywhere you work.

The assistants' native memory systems continue to work exactly as they do today. MemoryPlugin doesn't replace them. It gives them a shared source of truth.

Capability	ChatGPT	Claude	Gemini	With MemoryPlugin
Information learned in another AI assistant	Not available	Available after a manual one-time import	Available after a manual one-time import	Available automatically in your next conversation
Memory stays synchronized as you continue using different assistants	No	No (imports are snapshots)	No (imports are snapshots)	Yes, continuously
Search conversations across multiple AI assistants	Own conversations only	Own conversations only	Own conversations only	All conversations, in natural language

If you work across more than one of these tools, that's the difference worth having. Try MemoryPlugin free for 7 days.

For how memory systems work under the hood, the Field Guide to AI Memory covers the machinery in plain language.