How to maintain context between ChatGPT and Claude.

The free version of this is one document — a Google Doc, a Notion page, a plain text file — that holds what you’ve figured out so far. You keep it open, and after a session in ChatGPT or Claude, you copy in whatever’s new. Next time you open the other assistant, you paste the whole thing in, or attach the file, so it’s starting from the same place. This costs nothing and it genuinely works, for a while. It’s worth doing before you reach for anything else — you don’t need a tool to keep two assistants roughly in sync for a week or two.

The document is only as current as the last time you remembered to update it. A budget moves, a plan changes, an approach you’d settled on gets dropped for a better one — and if you don’t immediately go update the shared file, the next assistant you open is working from the old version with no way to know it’s stale. It states the outdated number with exactly the same confidence as the current one, because nothing tells it otherwise. Keep this up for a few weeks across two assistants and you end up with two documents that have quietly drifted apart, with no way to tell just by looking which line is actually still true.

The real fix is architectural, not habitual. Instead of a document you carry between assistants by hand, the project lives in one place outside of both of them, and you connect ChatGPT and Claude to it directly — each one can then read what’s already there and add to it itself, inside the same conversation you’re already having. Gemini can join too, through Gemini Spark rather than an ordinary Gemini chat. You still do the actual thinking wherever you prefer. The project just stops living inside any one conversation, or any one document you’re responsible for keeping in sync.

Reading the same facts from two assistants is the easy part. The harder part is what happens when something changes. Before it adds something new, your assistant checks what the project already says — and if you’ve landed somewhere different, the current answer updates instead of just sitting next to the old one as a second, contradictory line. Facts update in place and keep their history: the old number doesn’t vanish, it’s just no longer the one anyone reads as current. A decision that gets reversed works a little differently — it doesn’t get replaced, it stays exactly where it was, marked reversed, with the reason attached. So if you ever wonder whether you already ruled something out, the answer is sitting right there, instead of buried forty messages back in a transcript neither assistant can see anymore.

Say you’re apartment hunting, using ChatGPT and Claude for different parts of it. In ChatGPT, you land on a number and say: “Save it to my apartment hunt project — budget is $2,400 a month.” A week later, in Claude, after actually touring a couple of places, you land somewhere different: “Update the apartment hunt project — we’re now at $2,800, the extra parking made it worth it.” Claude checks the project, finds the $2,400 figure already there, and updates it — $2,400 is kept as what the budget used to be, not deleted. Back in ChatGPT the next day, you ask it to check the project before you start narrowing down listings. It reads $2,800 as current, because that’s what the project says now — not because ChatGPT remembers the conversation you had with Claude, which it never saw. You told it where to look, and it looked.

Open the project directly and it doesn’t read like a database of memory entries. It reads like a page someone actually wrote: the current budget, the decision about parking, whatever’s still undecided, in plain sentences — that’s the Simple view. Structured view breaks the same project into its labeled parts, if you’d rather scan it that way. Either way, it’s the same project underneath. You’re just choosing how much structure you want to see, and neither view needs an AI open to make sense.