Why does my AI assistant forget when I switch between ChatGPT and Claude?

Nothing is broken and you haven't missed a setting. All three of the big assistants can remember you in 2026 — once the switch is on, and each one separately, inside its own account. Here's why, the workarounds that genuinely help, and the point where every one of them runs out.

The short answer

You spend forty minutes with one assistant working through a client problem. The next morning you open the other one — because it drafts better, or because the first was busy — and you're explaining the engagement from scratch.

The reason is duller than it feels: each assistant's memory is scoped to its own account, by design. ChatGPT remembers what happened in ChatGPT. Claude remembers what happened in Claude. Gemini remembers what happened in Gemini. There is no shared standard between the vendors and no connection between the three stores, so switching tools means starting cold. Each of them remembers its own conversations well. None of them remembers across.

What each one actually remembers in 2026

It's worth being precise here, because a lot of writing on this subject is a year out of date. The old line that these assistants have no memory is simply wrong now — all three can remember you. Whether they are doing it for your account is a separate question, and the answer differs by vendor, by plan, and in Gemini's case by account type, so the table says where each switch lives. The problem is the boundary, not the capability.

Every row read against the vendor's own current help page on 2026-08-25. ChatGPT: OpenAI's Memory FAQ (help.openai.com, article 8590148, updated August 2026) for the controls, the sources, and the memory-summary quote; OpenAI's "Dreaming" announcement of 4 June 2026 for the rollout wording and the April 2025 origin of background curation. Claude: Anthropic's chat search and memory article (support.claude.com, article 11817273) for both toggles, the plans they're offered on, and project memory; its import-and-export article (12123587) for the import and export quotes; "What are projects?" (9517075) for project plan availability and RAG. Gemini: Google's "Get personalization with memory of your past Gemini chats" (support.google.com, article 16598469) for the Memory control, Keep Activity, the 18-and-over and personal-account requirements, and the Gems exclusion. Neither OpenAI's nor Google's page states whether memory is on or off by default, so this page doesn't claim one. All three vendors move these controls often — open the settings screen in each app before relying on any row.

Why there is no shared memory between them

Three separate reasons stack up, and it helps to see them apart:

• Memory is an account feature, not a file. What each assistant "knows" about you isn't a document sitting somewhere you can move. It's built and rewritten inside that vendor's account as you chat, in that vendor's own shape.
• There's no agreed format between vendors — and little incentive to build one. Nobody has published a common standard for personal memory, and your accumulated context is precisely what makes leaving a product expensive. That's not a conspiracy; it's just where the interests sit.
• Context can split even inside one product. In Claude, each project keeps a memory space separate from your ordinary chats, so work you did in a project doesn't inform a standalone conversation. Gemini's personalization runs on a personal Google account, so if your work life is on a workspace account it isn't there either. You can lose the thread without switching vendors at all.

The workarounds that help — and where each one runs out

These genuinely work. Use them. But it's worth knowing the ceiling on each before you build a habit around it.

1. Hand the next assistant a summary

The most reliable move, and the one people arrive at on their own: before you switch, ask the assistant you're leaving to write out everything it knows about the project — decisions, constraints, preferences, where things stand — then paste that into the other one.

Claude has productized exactly this. Its Settings → Memory includes an experimental memory import: run a supplied prompt on your current service asking it for all the context it holds about you, then paste the result into the import box. Anthropic is refreshingly plain about the limits — imports are "experimental and still in active development", "Claude may not always successfully incorporate imported memories", and Claude's memory "is designed to focus on work-related topics", so personal details may not carry over (verified 2026-08-25).

The ceiling: it's a snapshot, not a sync. It captures what one assistant will admit to knowing at one moment — and in ChatGPT's case, OpenAI itself says the summary "will not include everything that ChatGPT remembers". Everything that happens after the paste stays behind, and you have to do it again next time.

2. Keep one running context document

A single document per engagement — the brief, the decisions, the open items, the way this client likes to be spoken to — that you paste in at the top of any new conversation, in any tool. It makes you vendor-independent overnight, which is genuinely valuable.

The ceiling: you are the one keeping it current. It captures what you remembered to write down, in the ten minutes you had, and it goes stale the moment the work moves on. The detail you'll want in six weeks is rarely the one you thought worth typing up today.

3. Use each product's projects and spaces

All three offer some form of container where a body of context and instructions stays attached to a set of conversations — Claude's Projects, ChatGPT's projects, Gemini's Gems. Set one up per client, and at least you stop re-explaining the basics every session.

The ceiling: it's scoped to that vendor, so it doesn't travel — and each product adds an edge worth knowing. Claude's Projects are on every plan, free accounts capped at five; what's paid-only is the RAG-backed expansion that takes over when a project's knowledge outgrows the context window. The ceiling that matters is a different one: Anthropic says each project "has its own separate memory space", kept "separate from other projects or non-project chats" — so the container you set up to hold context together also fences it in. Google lists memory of past chats as "not available in certain features, like Gems or Live chats", so the container and the memory don't overlap there either.

4. Export what you can — but know what an export is

Where a product offers a way to get your material out, take it. Claude documents one aimed specifically at memory: you can view what it has stored and, in Anthropic's wording, "save this exported memory as a backup or bring it to another AI service by copying and pasting it into a local file on your computer" (verified 2026-08-25). Elsewhere, treat "export" as whatever that product's own data-controls screen offers on the day, and check it rather than assuming.

The ceiling: an export is an archive for you to read, not a memory another assistant will adopt. The one exception is Claude's import box, which takes a pasted export from another service — and Anthropic itself calls that "experimental and still in active development". No other vendor ingests a competitor's export, so in practice the only cross-vendor path is still the copy-and-paste one above.

So should you just pick one?

It's a fair question and the honest answer is that it wouldn't fix much. Choose between them on output — which one writes and reasons the way you want for the work in front of you — because on memory they've largely converged. Both remember your past chats once memory is switched on, and both now let you look at what they kept. If you're weighing whether to keep paying for two, we work through the split in how to use ChatGPT and Claude together.

Consolidating means one assistant remembers a slightly bigger slice of your typing. It doesn't touch the actual problem, which is the next section.

Where every one of these workarounds runs out

Look at what the four methods have in common. All of them make you the courier: you write the summary, you keep the document current, you remember to paste it in. And all of them move only one kind of material — what you typed into a chat box.

That's the smaller half of your context. The part that actually decides how Friday's call goes was never in a chat window:

• The thread your client has been replying to all week, now three messages deeper than the last time you read it properly
• The revised scope attached to Tuesday's email, which nobody has opened since it arrived
• What the finance lead actually committed to on last quarter's call — a sentence in a transcript, not a chat
• The voice memo you dictated in the car park, because you knew it wouldn't survive the drive home

No assistant you're switching between has ever seen any of that. So carrying context from one to the other, however diligently, recovers a fraction of what you were missing in the first place.

A different answer: keep the record outside the chat window

This is the shift worth making. Instead of ferrying context between chat tools, keep the record somewhere that isn't inside any of them — and let the chat tools be what they're good at.

That's what Equerry is built to be: a personal assistant who holds what you send it and hands it back with the source attached.

• Send it what matters — nothing gets in unless you do. CC or forward a client thread to your own Equerry address (something like alex@mail.getequerry.app) and every later reply on that thread lands too. Share a PDF or a screenshot from the share sheet, dictate a voice memo after a call, save a note. You never hand over your mailbox.
• Ask in plain language later. "What did the finance lead commit to on the scope change?" pulls across the thread and the attachment and your note from the call at once — because they're all in one archive rather than three vendors' accounts.
• Answers show their source. When an answer draws on your notes, emails or the web, it shows where it came from. Not "here's what I remember" — here's the artifact, one tap away.
• See and delete everything it kept. Every remembered fact is listed in one place and removable on tap, and every note shows what Equerry took from it. You're never guessing what your assistant thinks it knows about you.
• Get briefed before the call. Ahead of a client meeting, Equerry can brief you from what you've shared — including the reply that came in while you were double-booked and never opened.

The switching problem doesn't get solved so much as retired. Draft in one assistant, think in another, change your mind next quarter — the record isn't in either of them, so it doesn't go anywhere.

If you're paying for two assistants and losing the thread between them, the fix isn't a better handoff ritual. It's not keeping the record in a chat window at all. See how an assistant that remembers across your tools works, or read the honest side-by-side of ChatGPT memory vs Claude memory.

Last updated 2026-08-25.

All product names, logos, and brands are property of their respective owners. Equerry is not affiliated with, endorsed by, or sponsored by any other product mentioned; comparisons are provided for informational purposes.