Does turning on memory in ChatGPT, Claude, or Gemini make them share what they know about me?

Not on their own. Each vendor documents memory as something that lives inside its own app. OpenAI describes ChatGPT memory as drawing on your chats, files, and connected apps; Anthropic describes Claude's memory as building a synthesis of your chat history inside Claude, with a separate memory space per project; Google describes Gemini's memory as learning from your Gemini chats. None of the three vendor pages describes reading from, or writing to, one of the other two apps' memory.

Can I export my ChatGPT or Claude memory and just put it in Gemini?

Google documents a version of this: Import memory to Gemini has you paste a Gemini prompt into the other app and paste its answer back, and a separate option uploads an exported chat history as a .zip file. Both are copies made at one point in time, not a live shared memory. Anything you tell Gemini afterward stays in Gemini. Per Google, the import needs a personal Google Account, age 18 or older, and isn't available in the EEA, Switzerland, or the UK.

What is a remote MCP memory server, and how is it different from each app's built-in memory?

It is a memory store that lives outside any single AI app, reachable over the Model Context Protocol (MCP) — an open standard OpenAI, Anthropic, and Google Gemini Apps each document support for through their own connector features (Developer Mode custom apps in ChatGPT, Connectors in Claude, Connected Apps in Gemini). Connect the same MCP server to more than one app and every app that's connected can read and write the same memory, instead of each keeping its own.

Does Gemini's consumer app actually support connecting a custom MCP server?

Yes, per Google's own Connected Apps documentation — separate from Gemini's built-in memory feature. It's gated by region, age, and account type, and separate again from Gemini CLI's MCP support; see the Gemini setup guide for the current requirements.

Which of the three apps' memory features can I turn off without losing anything else?

Each vendor documents memory as a setting you control. OpenAI puts it under Settings, Personalization, Memory, and notes that "Don't mention this again" does not delete the information; Anthropic offers a Pause option for Claude that keeps existing memory but stops creating new memory, plus a separate Reset that deletes it permanently, and Google documents a Memory on/off setting in the Gemini app.

How to Give ChatGPT, Claude, and Gemini One Memory

What does "memory" mean in ChatGPT, Claude, and Gemini?

Three different things, built and controlled three different ways. Here's a quick comparison, then what each vendor's own docs say.

	ChatGPT (OpenAI)	Claude (Anthropic)	Gemini app (Google)
What it's called	Memory	Memory	Memory (personalization from past chats)
How it builds	Saves relevant details as you chat, plus other available sources depending on plan	Saves individual topics as you chat; a memory summary synthesizes your chat history	Learns from your Gemini chats
Where it's scoped	Per account, with project-level workspaces	Per account, with a separate memory space per project	Per Google Account
Where it lives	Inside ChatGPT	Inside Claude	Inside the Gemini app
Who can turn it off	You, in Settings → Personalization → Memory	You, in Claude's memory settings (Pause or Reset)	You, with the Gemini app's Memory setting

What does ChatGPT's memory actually do?

OpenAI's Memory FAQ says that when enabled, memory "helps ChatGPT automatically remember useful context from your chats, files, and connected apps to personalize your experience." You can see what it holds in a memory summary, which OpenAI notes "will not include everything that ChatGPT remembers based on your chats." Controls live under Settings → Personalization → Memory. As of September 2026, OpenAI says a new memory system is rolling out to Plus and Pro users in the US first.

None of that reaches outside ChatGPT. It's memory OpenAI built for OpenAI's product.

What does Claude's memory actually do?

Anthropic's help center describes a similar shape with different mechanics: "Claude saves memory as a set of individual topics as you chat, rather than summarizing conversations after they end." Alongside that, Claude also builds a rolling synthesis — Anthropic's docs describe Memory summary as automatically summarizing your conversations into "a synthesis of key insights across your chat history," which Anthropic says "is updated every 24 hours," and note that "Each project has its own separate memory space and dedicated project summary, so the context within each of your projects is focused, relevant, and separate from other projects or non-project chats." You can Pause memory (Claude keeps what it has but stops making new memory) or Reset it (permanently deletes everything, including project memories).

Same story: it's memory Anthropic built for Claude, living in Claude's account system.

What does Gemini's memory actually do?

Google's Gemini Apps Help describes it as a learning feature, not a settings toggle you fill in by hand: "With this feature, Gemini can learn from your chats to understand more about you and your world." Google's requirements say you must "Be 18 or over," that the feature "isn't available when you sign in to a work, school, or supervised Google Account." It's also limited to specific surfaces: the Gemini mobile app, the Gemini web app, Gemini in Chrome where available, and Gemini on a smartwatch — and it's not available in every Gemini feature, such as Gems or Live chats.

Same pattern again: memory Google built for the Gemini app, gated by Google's own account and region rules.

So why doesn't memory just travel between them?

Because none of the three vendors built it to. Each memory feature is a private store tied to that vendor's own account system, with its own settings, and none of the three vendors' docs above describes reading from or writing to the other two. Google is the one exception worth naming, and it's a manual one: Gemini Apps documents an Import from other AI platforms feature: paste a Gemini prompt into another AI app and paste its answer back, or upload an exported chat history as a .zip file. Either way it's a copy made at one point in time, in one direction, not a live shared memory. Nothing you tell Gemini afterward flows back to the app you imported from.

If you want ChatGPT, Claude, and Gemini actually drawing on the same up-to-date memory — not a copy-paste snapshot — that memory has to live somewhere none of the three apps owns.

The general approach: a memory server all three apps connect to

The fix is architectural, not a setting inside any one app: put your memory in a remote MCP server — a service that speaks the Model Context Protocol, an open standard for connecting an AI app to external tools and data — and connect that same server to each app. Every app you connect gets read and write access to the same memory store, so something you tell one shows up when you ask another.

This works because all three vendors document support for connecting outside MCP servers, through each vendor's own connector feature:

ChatGPT connects a custom MCP server through Developer Mode, a setting under Settings → Apps → Advanced Settings (paid plans only, web only — see the ChatGPT MCP guide for the current plan and platform requirements).
Claude connects one through Connectors, either from Anthropic's connectors directory or by pasting a server URL — see the Claude Desktop & Web setup doc for both paths.
Gemini connects one through Connected Apps, where Google's own docs state you "can connect personal or third-party apps by linking their Model Context Protocol (MCP) server" — a feature separate from Gemini's built-in memory, and separately gated by region, age, and account type (see the Gemini MCP guide).

None of this requires the memory server to be any particular product — any MCP server built to store and serve memory works the same way once it's connected. The steps, in outline:

Stand up or sign up for an MCP memory server that exposes tools an AI can call to save and retrieve what it knows about you.
Connect it to ChatGPT via Developer Mode → Create App, pasting the server's URL. Full walkthrough: add-mcp-server-to-chatgpt.
Connect it to Claude via Connectors — find it in the directory, or add it by URL. Full walkthrough: Claude Desktop & Web setup.
Connect it to Gemini via Connected Apps on the web, pasting the server's URL. Full walkthrough: add-mcp-server-to-gemini.
Ask each app to load context at the start of a session — memory servers can expose a tool for loading context. If an app doesn't call one on its own, ask directly ("what do you know about me?").
If you already have memory built up in one of the three apps, some memory servers offer a one-time import that reads what an app already knows and writes it into the shared server — worth checking for before you start over by hand.

What this does and doesn't fix

It gives all three apps the same underlying facts, preferences, and history, kept current from whichever app you last used. It does not touch each app's own built-in memory — ChatGPT's, Claude's, and Gemini's native memory features keep working exactly as their vendors document them, independently, alongside the shared server. And it's bounded by what each vendor's connector feature allows: ChatGPT's Developer Mode is paid-plan and web-only per OpenAI's docs, and Gemini's Connected Apps is gated by the same account and region rules as the rest of Gemini's personalization features. The shared memory is only as available as the weakest connector among the apps you're using.

A worked example

Tempreon is one implementation of this pattern: a hosted MCP server built to hold a person's knowledge, preferences, and decisions, and to serve that same context to the AI apps they connect. Its own setup docs cover ChatGPT, Claude, and a universal flow for other MCP clients; check those pages for the apps it currently documents rather than assuming every app above. The universal connection flow documents the same six-step pattern described above — find the MCP setting, add the server URL, choose Streamable HTTP and OAuth, authorize, and confirm the tools appear — plus the connection details and per-app guides linked throughout this page. Tempreon also documents a Memory Import flow that reads what ChatGPT, Claude, Cursor, or Mem0 already knows and brings it in as a starting point, rather than starting from zero.

This page describes the general pattern; it isn't an endorsement of one server over another. Any MCP server built for memory, connected to more than one app, produces the same result.