The 2-minute test: tell Claude, find it in ChatGPT

Updated 30 October 2026. The menus mentioned come from Anthropic's help centre and OpenAI's documentation, read on 30 October 2026.

To check that Claude and ChatGPT share the same memory, give Claude a precise fact, let it save it to Pryzm Memory, then open a new chat in ChatGPT and ask it to use that fact without repeating it. If it gets it right, the memory lives neither in Claude nor in ChatGPT: it lives in Pryzm, a persistent memory shared across AIs over MCP, and each AI reaches it with its own token.

Before you start

• Pryzm is connected to Claude (custom connector, available on every Claude plan) and to ChatGPT (custom MCP server, on the web, with a paid plan).
• On ChatGPT, read access is enough. Claude does the writing. Per OpenAI, the Pro plan is read-only: that's enough for this test.
• The “main memory” instructions are pasted into both AIs. Without them, an AI often waits to be told explicitly to save or to search.

Connection guides: Claude, ChatGPT, and the instructions on the connection page.

No ChatGPT subscription? The same test works between Claude and Claude Code, Cursor or VS Code. Mistral's Le Chat also offers custom MCP connectors.

The steps

1. In Claude, start a new chat and check that Pryzm is on: “+” button at the lower left → Connectors.
2. Give it a made-up but precise fact, for example: “Note for the Atelier Lumen project: website quote at €4,800, delivery end of November. The client wants a clean look, no purple at all. Save it to Pryzm.”
3. Watch Claude call the save tool (memory_store). The call shows in the chat, then Claude confirms what it kept. The first time, it may ask permission to use the tool: allow it.
4. (Optional) Check in Pryzm: in your dashboard, under Memory, the new memory is in the list.
5. In ChatGPT, start a new chat. Type @ in the prompt box and pick the Pryzm plugin.
6. Ask without repeating anything: “Write the email sending the quote to Atelier Lumen.”
7. Watch ChatGPT search the memory (memory_search), then write the email with the right price, the right deadline, and no purple.

To go further, run it the other way, if your ChatGPT plan allows writes: tell ChatGPT “The Atelier Lumen delivery moves to 5 December”, confirm the save, then ask Claude for the delivery date. It should answer 5 December. Nothing is erased: when the AI marks the new memory as replacing the old one, the old one stays visible, marked as replaced.

What this test proves

• The memory lives outside the AIs. Neither Claude nor ChatGPT owns it: each reads it over MCP, with access you can remove.
• Something said once works everywhere. No more copying a project's context from one AI to the next.
• A correction applies to all of them. The new version is available in every connected AI.

If something doesn't work

• ChatGPT says it knows nothing about the project: the plugin isn't called in this chat. Type @ and pick it, or say “search my Pryzm memory”.
• Claude saves nothing: the connector isn't on in the chat, or the instructions aren't pasted. Say “Save this to Pryzm”.
• An AI answers with old information: its built-in memory competes with Pryzm. Turn it off, or say “according to Pryzm”.
• The memory isn't in Pryzm: one of the AIs may be signed in to a different Pryzm account. Remove the connector and reconnect it with the right username.
• “The free plan includes 1 project” message: Pryzm's free plan has 100 memories and 1 project. Save to your existing project or without a project.
• Your message starts with ##: that's private mode, nothing is saved. By design.

When this test isn't for you

If you only use one AI, its built-in memory is enough: a shared memory only helps if you move between them. To know what to put in it once the test works, see what to store in your AI memory. To bring in what ChatGPT already knows about you, see import your ChatGPT memory into Pryzm.