ChatGPT Memory Import: Supported Paths and a Manual Fallback

For a ChatGPT memory import, first identify the source and destination: OpenAI's desktop import supports selected material, including Claude Code project memories. Bringing a reviewed personal context summary into a ChatGPT conversation is a different, manual process; it does not establish that an earlier memory system has been restored.

The useful goal is to carry accurate information into the next session, then check how it is used. fenbs keeps the ongoing work record outside the destination assistant, so a transfer does not have to become the only copy of your project decisions.

What supported import should you check first?

1. Identify whether you are moving coding-project context or personal preferences used in ChatGPT web conversations.
2. For supported desktop-agent material, check Settings > Import, or General > Import other agent setup if that is the available control.
3. Choose the supported source and review the selected items before importing. Complete any connection setup the app flags afterward.
4. For other material, prepare a reviewed summary and supply it explicitly to the relevant conversation or settings. Confirm the destination can use it before continuing.

OpenAI's import guide↗ (opens in a new tab) lists supported sources and destinations. Its import table includes Claude Code project memories. It does not describe a universal importer for every assistant's personal memory, so match your source to the documented path rather than assuming any exported file will work.

Also distinguish the destinations. OpenAI documents separate memory systems for ChatGPT web and local Codex clients↗ (opens in a new tab). Importing supported local project context should not be treated as proof that ordinary ChatGPT web personalization now contains it.

How is a context summary different from an import?

A summary is text you review and give the assistant to use. An import is a supported product flow that handles specific information from a specific source. Both can help you continue, but they provide different evidence about where the information went.

Think of moving a research project. You might paste an accurate page describing the research question, source list, and unfinished checks. The new conversation can work from that page, but the action alone does not recreate the old account's settings, historical conversations, or organization of remembered information.

OpenAI's account-transfer guidance↗ (opens in a new tab) makes a related distinction: uploaded exported conversations are reference material, and that process does not transfer memories or other account settings. Treat a conversation export as evidence to review, not as a memory-restore file.

If your destination is Claude, the separate Claude memory-import guide owns that process. The question here is what you can bring into ChatGPT and how to avoid confusing a useful summary with a supported transfer.

What should you include in a reviewed summary?

Start from information you can confirm yourself. Ask the original assistant for a summary if that saves time, then compare it with the actual decisions and source material. Keep a correction visible instead of quietly trusting a fluent but inaccurate description.

• Purpose: what you are trying to achieve and who the result is for.
• Stable preferences: formatting and communication choices that still apply.
• Project facts: the current source versions and the constraints relevant to this work.
• Approved decisions: what a person chose, with enough context to avoid reopening it accidentally.
• Open work: what remains, what needs checking, and the first next action.
• Exclusions: ideas that were rejected or details that should not guide this task.

For example, “Use a short summary followed by evidence” may be a lasting writing preference. “The customer has approved the draft” is a claim about one project and needs a reliable source. Keep those separate so a temporary project fact does not become an instruction for unrelated work.

Remove obsolete facts and information you lack permission to transfer. Do not include passwords, private identifiers, or unrelated details about colleagues just because they appeared in the old conversation. A smaller record can be easier to check without losing the conditions that matter.

Where should you put each kind of information?

Use the narrowest home that matches the information. Put immediate task context in the conversation where it will be used. Consider project context for material that belongs to one continuing project; the ChatGPT project-memory guide covers that scope separately.

For response preferences you want to set explicitly, OpenAI documents custom instructions↗ (opens in a new tab). Review their scope before adding a project-specific constraint there. A global preference should still make sense when you start an unrelated task tomorrow.

For remembered information, review the controls your account actually offers. OpenAI's memory guidance↗ (opens in a new tab) places them under Settings > Personalization > Memory and explains that availability varies. You can ask what ChatGPT remembers and correct information, but a summary is not necessarily exhaustive.

Keep project rules, decisions, and next steps in a record you can inspect and maintain. fenbs's AI context workflow puts those work details together for connected assistants without turning every project update into a personal preference.

How do you test what the new session understands?

Use three questions with answers you know: what is the current goal, which choice has already been approved, and what still needs checking? Compare the answer with the source record before asking for substantial new work. A sensible-sounding answer can still omit the one constraint that changes the task.

Then ask for one small continuation, such as revising a paragraph or proposing the next check. Review whether the transferred preference or project rule actually affected that result. Keep the original summary so you can correct the input when the output reveals a misunderstanding.

If you are working in Codex, use the Codex overview to keep the product boundaries clear. Do not test one client and infer that another client has the same context. Check the destination you intend to use for the next piece of work.

How should you keep the record current afterward?

Treat the first transfer as a starting point. When a person changes a decision, update the source record and explain which earlier choice it replaces. When a task finishes, record its result and the checks that support it. Old summaries should not silently overrule newer evidence.

For personal personalization, review unwanted or outdated information using the available memory controls. For continuing work, keep a clear next step outside the chat. The switching models collection covers that broader continuity problem. For the underlying personalization behavior, see how ChatGPT memory works.

How to do this with fenbs

1. Create a free account and add the project whose work you want to preserve.
2. Review the source summary. Add the open tasks, record approved choices in Decisions and rules, and keep unresolved questions open.
3. Connect a supported assistant using the integration guide. MCP is a standard way for an AI app to use an outside tool.
4. Ask the assistant to read the project context before continuing and save progress before stopping, including unfinished checks and the next action.

fenbs does not import ChatGPT memories. It keeps work, rather than a personal profile, so it is not the right fit if you only want remembered lifestyle preferences. Use it when project context should remain available across connected assistants.