How to keep the same memory across ChatGPT, Claude and Gemini.

With two assistants, a shared document is a reasonable habit: keep it open, copy in whatever’s new, paste it back in next time. Add a third assistant and the same habit gets harder to keep up — now there are three places to remember to update, and it only takes forgetting one of them to have a document that’s wrong in exactly the copy you happen to be looking at. It’s still worth doing for something short-lived. It’s the maintenance that stops scaling, not the idea.

The failure mode is the same as with two assistants, just harder to catch with three. You update the document in one place, mean to update the other two, and life happens. Now one assistant is answering from last week’s version and has no idea it’s behind — it isn’t lying to you, it’s reading the only copy it has. The more assistants you’re juggling, the more likely exactly this happens, and the harder it is to notice which copy is the stale one without comparing all three side by side.

The fix is the same one, regardless of how many assistants you’re using: stop keeping the project in a document you carry between them, and connect each assistant to the same project directly. ChatGPT and Claude can both reach it from an ordinary conversation. Gemini reaches it too, through Gemini Spark rather than an ordinary Gemini chat — a separate mode built for exactly this kind of task, not the regular chat box you land in by default. Once each one is connected, all three are reading from, and adding to, the same current state — not three versions of it.

Reading the same facts everywhere is the easy part. What matters more is what happens when your thinking changes, because with three assistants in the mix, it will. Before it adds something new, whichever assistant you’re talking to checks what the project already says. A fact that’s changed updates in place, with the old value kept as history. A decision that gets reversed is handled differently on purpose: it isn’t deleted or written over, it stays exactly where it was, marked reversed, with the reason attached — so any assistant that reads the project afterward sees not just what you decided, but that you’d decided something else first, and why that changed. An open question you resolve moves from open to answered the same way, and can reopen if it turns out you were wrong to close it.

Say a few of you are planning a trip. In ChatGPT, you settle it: “Decision for the trip project — we’re doing it as one big week, everyone together, one house.” Two weeks later, in Claude, it’s clear that’s not going to work — half the group can’t get the same week off — so you tell it: “Reverse that decision. We’re splitting it into two smaller long weekends instead, easier for people to actually make.” Claude doesn’t erase the first decision. It stays in the project, marked reversed, with the scheduling conflict as the reason. Before you start drafting invites, you open Gemini, switch to Spark, and ask it to check the trip project first. It reads the two-weekends plan as the current decision, sees the one-big-week plan sitting right below it marked reversed, and drafts invites for two smaller weekends — not because it remembers your conversation with Claude, which it didn’t see, but because you told it exactly where to look.

Open the project on its own and it reads like a page, not a memory log: the current plan, the decision that got reversed and why, anything still unsettled, all in plain sentences — that’s the Simple view. Structured view shows the same project broken into its labeled parts, if that’s how you’d rather scan it. Both are the same project underneath, and neither needs an assistant open to make sense of.