Import Memory Claude: A Complete Guide to Migrating from ChatGPT

Anthropic has just launched Import Memory, and ever since, the big question is everywhere: “Can I really transfer my ChatGPT memory to Claude?” The short answer: yes. The honest answer: yes, but probably not what you think you’re transferring.

I did the migration myself. Not in five minutes, and not without surprises. What I discovered along the way is that nearly all the available guides gloss over the process in two steps, present Import Memory as a complete migration, and don’t mention what gets left behind.

This guide is here to change that.

Here you’ll find the crucial distinction between memory and history (the most misunderstood point), the official step-by-step method with the integrated Anthropic prompt, an alternative method for free users via JSON export, Claude’s post-migration actual limitations, and prompts to get the most out of your imported memory.

What “Memory” Really Means: ChatGPT vs Claude

How ChatGPT Stores Your Data

ChatGPT manages memory on three separate levels, and confusing these is the root of most disappointments during migration.

The first level is automatic long-term memory: over time, ChatGPT extracts information about you and summarizes it as discrete “memories.” Things like: “The user is a freelancer,” “prefers concise answers,” “has a finance newsletter project.”

These memories are visible in Settings > Personalization > Memory.

The second level is custom instructions: a text block you write to set ChatGPT’s context and expected behavior. Think of it as your “user system prompt.”

The third level is your conversation history: the raw archive of all your chats, exportable in JSON via Settings > Data Controls.

These three levels are independent. Import Memory only transfers the first one, the automatically summarized memories.

A freelance writer who has trained ChatGPT on their style for two years will recover their tone preferences and some professional information.

Not their chats, not their files, not their conversation-by-conversation style corrections.

How Claude Memorizes (and Why It’s Different)

Claude doesn’t memorize anything by default. This is a philosophical difference, not a technical gap.

Anthropic has intentionally chosen to make memory opt-in, visible, and controllable—unlike ChatGPT’s opaque approach where memories are created without you knowing exactly when or how.

Once Claude memory is activated in settings, it gradually builds up with each interaction. You can see what Claude remembers, edit it, or delete it. Memories are handled via visible “tool calls” in the interface (no black box).

Claude Projects add an extra layer: a structured, persistent memory for each workspace, with extended context.

One project per client, one for your AI research, one for your blog—each space maintains its own active context across sessions.

The combination of passive memory (memories) + active memory (Projects) is what gives Claude the potential to be more powerful than ChatGPT over the long run for power users.

But it requires initial setup that ChatGPT manages automatically.

What Gets Imported—and What Doesn’t

Here’s what you get with Import Memory:

• The automatically generated summarized memories from ChatGPT (preferences, professional context, habits)
• These memories are merged with your existing Claude memory—not replaced

Here’s what stays in ChatGPT:

• Your full conversation history
• Your customized GPTs and their settings
• Files you’ve uploaded
• Saved web search results
• Third-party plugin memory

Remember: Import Memory is like switching phones and moving your address book and app preferences—not your photos or texts. Your profile moves, history stays behind.

An entrepreneur with dozens of business memories in ChatGPT gets their profile information back. A developer who used a custom GPT for code reviews will have to manually recreate this setup in a Claude Project—Import Memory doesn’t touch GPTs.

Official Anthropic Method: The Step-by-Step Guide

Pre-requisites Before You Start

Before starting the import, check these three quick points:

• A Claude.ai account — import is only available to Pro and Max plans (free plan users can use the manual method below)
• Memory enabled in Claude: Settings > Privacy > Enable Memory
• An active memory in ChatGPT: OpenAI Settings > Personalization > Memory — make sure it’s enabled and contains memories

Step 1: Extract Your Memories from ChatGPT with the Anthropic Prompt

Anthropic has published an official prompt to use directly in ChatGPT to extract your memories in a format ready for import. Open a new chat in ChatGPT and paste this prompt:

Official Anthropic Prompt:

Please share all the memories you have stored about me. List each memory as a separate bullet point, using plain language. Include all personal preferences, facts, habits, and any context you’ve learned about me. Format the output as a clean, readable list that I can copy and use elsewhere.

This prompt generates a clear, human-readable text block—not a technical JSON file. It’s editable, and it’s exactly what Claude expects for import.

Good news for users of other platforms: this prompt also works in Gemini, Copilot, and Mistral. The Import Memory feature accepts formatted memories from any source, not just ChatGPT.

Step 2: Import the Block into Claude

Once your memory block is copied, head over to Claude.ai:

• Settings (icon bottom left) > Privacy > Memory Preferences > Start Import
• Paste your memory block into the dedicated field
• Confirm the import

What happens next: Claude merges imported memories with any it already has about you. No existing memories are overwritten. For large imports, it may take up to 24 hours for all memories to be fully integrated—this delay is due to asynchronous processing, not a bug.

Step 3: Verify and Refine Imported Memory

After the import, don’t assume everything went perfectly. Test it with this prompt:

“What information do you have about me? List all your active memories about me, specifying which ones came from the ChatGPT import.”

Claude lists its active memories with sources visible in the tool calls. For an entrepreneur who imported dozens of business memories, this is the time to confirm that critical info is present and correct.

Correction and refinement directly via chat:

• “Forget that I work in [sector X], that’s outdated”
• “Add that I prefer answers with concrete examples”
• “Correction: I’m based in Lyon, not Paris”

You can also manage memories via Settings > Privacy > Manage Memory, with a toggle for each memory.

For Free Users: The Manual Method via JSON Export

Exporting Your ChatGPT History

OpenAI data export is available to all plans, including free. Here’s how:

OpenAI Settings (on chatgpt.com) > Data Controls > Export My Data > Confirm by email > Download the ZIP within 24h

The ZIP contains several files:

• conversations.json: a complete archive of all your chats
• chat.html: a readable version of the history
• user.json: your account info
• message_feedback.json: your feedback on responses

Building a Summary-Memory from JSON

The conversations.json file can be large (hundreds of MB for an active user).

The trick: load it directly into Claude (or ChatGPT) and ask it to extract your user profile.

Sample prompt to generate your summary-memory:

"Here is my ChatGPT conversation history in JSON format. Analyze it and generate a structured summary of my preferences, work habits, recurring topics, preferred communication style, and any personal or professional context that stands out. Format the result as a clear bullet-point list I can reuse as context in other tools."

The result is often eye-opening: you might discover patterns in your use that you hadn’t consciously formulated.

It’s also an opportunity to filter what you actually want to transfer.

Integrating This Summary Into Claude’s Memory

Two options depending on your plan:

• Pro/Max plan: use Import Memory with your generated summary (same path as the official method)
• Free plan: paste the summary directly into your Claude profile instructions (Settings > Profile > Instructions) or start a conversation with this context

Limitations to Know Before Migrating

What Claude Can’t Reproduce from ChatGPT

Migrating with incorrect expectations is the best way to be disappointed. Here’s what’s missing in Claude and won’t show up after import:

• Native image generation: Claude does not generate images. If DALL-E is part of your daily workflow, a complete migration isn’t for you—at least, not yet.
• GPT Store and custom GPTs: there’s no direct equivalent to the GPT marketplace. A developer who configured a specialized GPT for code review will have to manually recreate this setup in a Claude Project, with structured instructions.
• Voice mode in French: limited compared to the voice experience in ChatGPT.
• Plugins and third-party integrations: OpenAI’s integration ecosystem does not have a direct equivalent on the Anthropic side.

Claude outperforms ChatGPT on deep analysis, reasoning across long contexts, and writing quality. It doesn’t beat it on everything.

The Gray Areas of Confidentiality

The question everyone has but doesn’t ask: what happens to your data once it’s imported to Anthropic?

Anthropic’s official stance: memories are encrypted, and Anthropic commits not to use them for model training without explicit consent.

This is a better position than many competitors.

What remains unclear: server-side deletion. When you ask Claude to forget a memory, it disappears from your interface, but there’s no independent audit to confirm full deletion from Anthropic’s servers.

This is a matter of trust, not a technically verified guarantee.

For European users, GDPR is a real concern: Anthropic is a US company, and data transferred to its US servers falls under a different legal framework than the GDPR.

The AI Act 2026 increases requirements on AI data transfers to non-EU countries.

The pragmatic recommendation: delete the source memories in ChatGPT after import, and only transfer what you really need in Claude.

Should You Fully Migrate or Stay Hybrid?

The answer depends on your profile:

Full migration is recommended if you’re a writer, developer, or analyst working with long texts, complex code, or dense documents; Claude excels at these, and imported memory integrates well.

Hybrid mode makes sense if you regularly use GPT images for image generation, specific custom GPTs, or real-time web search with saved results. Keep ChatGPT for those, use Claude for deep analysis and writing.

Recommendation: Don’t migrate fully if custom GPTs are central to your workflow. Manual recreation in Claude is possible, but time-consuming, and the result is not identical.

Getting the Most Out of Claude’s Memory After Migration

Structuring Your Claude Memory With Projects

Imported memory is passive—it informs Claude who you are. Claude Projects are active memory, defining the context for a specific task with 200,000 persistent tokens.

The combination is something ChatGPT can’t offer:

• Project “Client X”: communication instructions, decision history, reference documents—Claude remembers them every session
• Project “AI Research”: your research areas, sources you follow, your editorial angle
• Project “Blog”: your editorial guidelines, published posts, defined style

If you want to see how Claude compares to ChatGPT for advanced use cases, this passive + active memory architecture is exactly what sets the two apart long-term.

Key Prompts for Auditing and Enriching Your Memory

Four practical prompts to manage your Claude memory post-migration:

Comprehensive audit:

"List all your active memories about me, organized by category (professional, preferences, style, context)."

Introduction from memory:

"Introduce yourself by explaining what you know about me and how you’re going to adapt your answers accordingly."

Correcting incorrect info:

"You’ve remembered that [X]. That’s not correct. The reality is [Y]. Update your memory and confirm the change."

Targeted enrichment:

"Here’s some extra information you should remember about my work context: [list]. Add these to your memory and tell me what you've saved."

These prompts work because Claude’s memory is transparent: you see the tool calls, you control what’s stored and removed.

That’s the clearest difference from the opacity of ChatGPT’s system.

If you want to dig deeper into ChatGPT alternatives available in 2024, or you’re considering using Mistral as a French alternative alongside Claude, the export methods described here work with both.

Key point: Import Memory transfers a profile, not a history. True migration to Claude is something you build over time, combining imported memories with well-structured Projects.

The final decision is simple to make. Are you a writer, developer, or analyst working on long and complex content?

Then full migration is worth it, and Import Memory speeds up your start. Were you using DALL-E every day or relying on custom GPTs? Keep both tools, let Import Memory help Claude get to know you, and let ChatGPT handle what it does best.

FAQ

Is Import Memory available on the free Claude plan?

No. Import Memory is only available to Pro and Max subscribers. Free users can use the manual method via OpenAI JSON export, then paste the generated summary into Claude’s profile instructions or directly into a conversation.

Will my existing memories in Claude be erased if I import from ChatGPT?

No. The import merges, not replaces. Imported memories are added to what Claude already knows. If two memories conflict, Claude may ask you to choose, or you can correct them manually in the chat.

Why can the import take up to 24 hours?

For large imports, Claude processes the memory merge asynchronously—in the background, without blocking the interface. This delay mainly affects users with dozens or hundreds of memories. A quick import of a few lines is usually active immediately.

Is my ChatGPT conversation history transferable to Claude?

Not directly via Import Memory. The raw history remains in the OpenAI JSON export. You can load that file into Claude to generate a summary of your preferences, but Claude cannot “replay” your old conversations as if they happened with it.

Can Claude reproduce a custom GPT I had in ChatGPT?

Not automatically. You’ll need to set up the configuration manually in a Claude Project: copy your GPT instructions, add reference files, define expected behavior. The result is functionally similar but requires initial setup.

Are my imported data used to train Anthropic’s models?

Anthropic states that memories are not used for model training without explicit consent. The data is encrypted. Server-side deletion cannot be independently audited—this is a matter of trust, not a guaranteed technical assurance.

Are European users protected by GDPR when importing into Claude?

Partly. Anthropic follows the explicit consent required by GDPR for data transfer. But data moves to US servers, which means it falls under a different legal framework. The AI Act 2026 strengthens requirements for such transfers. The pragmatic advice: only transfer what’s needed and delete the source data in ChatGPT after importing.

How do I know if the import worked?

Use this prompt directly in Claude: “List all your active memories about me, specifying which came from the ChatGPT import.” Claude lists memories with their sources visible in tool calls. Compare to what you had in ChatGPT to spot any gaps.

Does the Anthropic prompt work in AIs other than ChatGPT?

Yes. The extraction prompt works in Gemini, Copilot, and Mistral. The Import Memory feature accepts a clearly structured text block from any source. It works best from ChatGPT, but other sources are supported.

What’s the difference between Claude standard memory and Project memory?

Standard memory (memories) is a passive, global memory—it applies across all your Claude conversations. Projects provide active, contextualized memory: each Project maintains its own 200,000-token context between sessions. For serious professional use, the two work together: memories for your global profile, Projects for specific work contexts.