How to share project context between ChatGPT and Claude

By the Flori team · Published 2026-10-22 · Updated 2026-10-22

ChatGPT and Claude do not share memory, and neither can read the other’s chats. There are three practical ways around that. You can paste a short project brief into each new chat. You can copy your ChatGPT memory into Claude once. Or you can connect both assistants to one outside project record, so each reads and saves the same brief, decisions and next steps. The first two are enough if you switch now and then. The third keeps both current when you move between them every week.

Why context does not carry over

Claude keeps memory per project. Anthropic says “Each project has its own memory.” In a project, Claude uses that project’s memory along with your memory from outside projects, but not memory from other projects. Searching past chats is available on paid plans, and inside a project it only covers that project’s chats.
ChatGPT’s reach depends on the project’s memory setting. A project with project-only memory cannot use conversations outside it. Shared projects always use project-only memory. OpenAI also says memory “does not retain every detail from every conversation”.
Neither reads the other. Neither vendor documents a way for ChatGPT to read Claude’s memory, or for Claude to read ChatGPT’s.

So when you plan in ChatGPT and draft or build in Claude, the second assistant starts without the decisions you made in the first. That is the problem of re-explaining the project every session, and it gets worse with every client or project you add.

Option 1: Paste a project brief

Keep a short brief and paste it at the start of each new chat, in either assistant.

PROJECT: [name]
GOAL: [one sentence: what done looks like]
CURRENT STATE: [two or three lines]
DECISIONS (do not reopen without asking):
- [date] [decision] because [reason]
OPEN QUESTIONS: [list]
CONSTRAINTS: [budget, deadline, tone, tools, what needs my approval]
NEXT STEP: [one concrete action]
WHEN YOU FINISH: give me an updated version of this brief.

Works well when you switch occasionally and the project is small. Where it breaks: the brief is only as current as your last paste. The assistant can hand back an updated version, but you have to store it somewhere and remember which copy is the latest. With several clients, the copies multiply.

Option 2: Copy your memory into Claude once

Claude can import memory from other AI services. You ask the other assistant to write out what it remembers, using the prompt Anthropic provides. Then, in Claude, open Settings → Memory, choose Start import, paste the text and choose Add to memory. Anthropic says imports work on every plan, on the web and in Claude Desktop.

Works well when you are moving from ChatGPT to Claude and want Claude to know your role, preferences and ongoing work from the first day.

Where it breaks:

It is a one-time copy, not a sync. Decisions you make in ChatGPT afterwards never reach Claude.
Anthropic calls the feature “experimental and still in active development”.
Memory holds a summary about you, not the working state of a project. It will not tell Claude which task is blocked or what you agreed with a client yesterday.

Option 3: Connect both assistants to one project record

Instead of moving memory between assistants, keep the project’s working state outside both of them. Use a record each assistant reads at the start of a session and updates when useful work is done. Claude calls these outside connections connectors. ChatGPT calls them apps or plugins, depending on your account. If both connect to the same record, a decision saved from Claude is there when you open ChatGPT, and the other way round, within the plan limits below.

Item	Why it matters
Project brief	The first thing any assistant reads: the goal, the constraints, and what needs your approval.
Decisions with reasons	Stops the next assistant from reopening questions you already settled.
Tasks and status	Shows what is done, in progress or blocked.
Results and links	Lets you check the work yourself instead of trusting the assistant’s summary.
Next step	Gives a fresh session somewhere to start.

If your projects already live in a work tool that offers its own connection for ChatGPT or Claude, use it: the method is the same. Flori is a project workspace built for this exact job. Briefs and decisions are first-class items, you choose read-only or update access for each connection (one project or the whole workspace), and you check every saved change in Flori.

Where it breaks: whether ChatGPT can save through a connection depends on your plan (see below). A connection does not import old conversations, so only what you or an assistant saves is in the record. An assistant saying “saved” is not proof. Check the record itself, which is what the test at the end is for.

The three options compared

	Paste a brief	Copy memory into Claude	Shared project record
Setup	None	A few minutes, once	A few minutes per assistant
Stays current	Only if you paste again	No, it is a one-time copy	Yes, when assistants save to it
Keeps decisions with reasons	If you write them down	Not reliably	Yes, as saved items
Works across ChatGPT, Claude and coding tools	Yes, by hand	A manual copy, one direction at a time	Yes, for each assistant that can connect
Main risk	Stale copies	An incomplete import	Plan limits on saving; saves nobody checked

What each plan allows

Assistant and plan	Add a custom connection	Save through it
Claude Free	Yes, one custom connector	Yes, within the access you grant
Claude Pro and Max	Yes	Yes
Claude Team and Enterprise	An owner adds it for the organization, then each member connects	Yes
ChatGPT Business	Only admins and owners can add it	Yes
ChatGPT Enterprise and Edu	Admins, owners and members they authorize	Yes
ChatGPT Pro	Yes	No, read and fetch only
ChatGPT Free, Go and Plus	Not listed in OpenAI’s current article	Not listed

Custom connections in ChatGPT work on the web only, not in the mobile apps. On a ChatGPT plan that can only read, ChatGPT can still pick up the brief and decisions that Claude saved. If you plan in ChatGPT on that kind of plan, ask it to give you each decision as one line of text, then save it from Claude or from the project tool directly.

Set it up with Flori

Create the project and write the brief. Use the fields above, and note which actions need your approval.
Connect Claude. Follow the Claude setup guide. At Flori sign-in, choose the project and the access level: read-only lets Claude read, and update access lets it save.
Connect ChatGPT. Follow the ChatGPT setup guide. If your plan only supports reading, connect with read-only access.
Save one decision. In Claude, ask: “Read my Flori project brief. We have decided to [decision] because [reason]. Save that decision to the project and give me the link to the project.”
Check it in Flori. Open the project and confirm that the decision appears with its reason.

Test that it worked

Run this two-minute check before you depend on the setup.

Open a new chat in ChatGPT, not the one where you set things up.
Ask: “Read the brief and decisions for [project] in Flori. What did we decide about [topic], and why? What is the next step?”
Compare the answer with what Flori shows.

Pass: ChatGPT states the decision and the reason Claude saved, and names the same next step.

If it fails:

It cannot find the project. You granted access to a different project or workspace. Reconnect and choose the right one.
It finds the project but not the decision. The save never happened. Save the decision again and check the link.
It answers from memory instead of the record. Ask it to read the brief from Flori and name the item it used.

Keep the record useful

Use one project per client or initiative, and share only the project an assistant needs.
Write decisions as decisions. “Chose the annual plan because the client pays upfront” is useful. “Discussed pricing” is not.
Review before it lands. Let assistants propose changes, and approve them before anything client-facing changes.
Retire stale items. Update the brief when the goal changes, so the next session does not start from an old plan.

Frequently asked questions

Can ChatGPT and Claude share memory directly?

No. Each keeps its own memory, and neither vendor documents reading the other’s. Use a pasted brief, a one-time memory import into Claude, or a shared project record.

Can I import my ChatGPT memory into Claude?

Yes, as a one-time copy. Anthropic gives you a prompt that exports memories from another service, and Claude’s memory settings have an import option. Anthropic labels the feature experimental, and it does not keep the two in sync.

Does this work with Cursor, Codex or Claude Code too?

The shared record works with any assistant that can connect to the same tool. For coding tools without a connection, a handoff file in the repository does a similar job.

Does connecting a tool give the assistant my whole account?

You choose. In Flori you pick a workspace or a single project, and read-only or update access, and you can revoke the connection at any time. Other tools have their own permission settings.

My ChatGPT plan can only read. Is this still worth it?

Often, yes. ChatGPT can still read the brief and decisions that Claude or you saved. Save decisions from the assistant or tool that can write.

Sources

Checked October 19, 2026. Vendors change these pages often, so check the source before you rely on a plan detail.

Keep exploring

Start your workspace