Claude’s Memory Import Is Useful, but It Leaves 99% Behind

Claude’s Free Memory Import Is Good. It Also Only Transfers 1% of Your ChatGPT Data. Here Is What Handles the Other 99%.

When Anthropic launched its memory import tool at claude.ai/import-memory, it solved a real problem. Users switching from ChatGPT could extract their stored preferences and load them directly into Claude’s memory system. The tool is free, it works, and it takes about five minutes.

What it does not do is transfer conversation history. And for users who have spent months or years building context inside ChatGPT, the conversation history is where almost all of the actual value lives.

What Claude’s import tool transfers and what it does not

Claude’s memory import extracts the contents of ChatGPT’s memory panel. These are short preference entries stored under Settings > Personalization > Manage Memory. Facts like the user’s name, job, location, communication preferences, and topics they asked ChatGPT to remember. The entire memory panel holds approximately 1,200 to 1,400 words.

This is genuinely useful data. After importing it, Claude knows who the user is and how they prefer to interact. But the memory panel is not the conversation history. The conversation history is the complete record of every exchange a user has had—every message, every response, and every research session. For a user active for a year, this history typically contains millions of words. The memory import prompt cannot access this data; these are two separate systems stored and exported differently.

Where Memory Forge fits in the migration process

Memory Forge, built by Phoenix Grove Systems, handles the second step: the conversation history that Claude’s import tool does not touch.

The process starts with ChatGPT’s built-in data export (Settings > Data Controls > Export Data). Users receive a zip file containing conversations.json. This file is unstructured and unreadable by any AI assistant in its raw form. Memory Forge processes that file, organizes conversations chronologically, and outputs a single structured Markdown file called a memory chip.

The tool’s advanced mode adds a curation layer. Users see every conversation listed with search and filtering by type (standard chats, Custom GPTs, Siri interactions). This means transferring only the threads that carry meaningful context.

The privacy model: local processing, not cloud processing

One of the most common questions about Memory Forge is whether conversation data is sent to servers. The answer is that it is not. The tool runs entirely as client-side JavaScript in the browser. No conversation data is transmitted to any external server. Users can verify this by opening their browser’s developer tools (F12), clicking the Network tab, and monitoring activity while processing. The tool generates zero outgoing requests containing conversation data. This architecture means Memory Forge functions more like a local file converter than a cloud service.

Complementary, not competitive

The relationship between Claude’s import tool and Memory Forgeis best understood as sequential steps:

Step one: Use Claude’s import tool to transfer stored memory entries (preferences, name, job).
Step two: Export conversation history from ChatGPT, process it with Memory Forge, and load the resulting memory chip into Claude.

This costs$3.95 per month and transfers the complete archive that the import tool does not reach. The same process works for users migrating to Gemini, Grok, Perplexity, or local AI models.

The scale of what users are moving

The #QuitGPT movement has now surpassed 700,000 pledged cancellations, while Anthropic’s user base has jumped more than 60 percent. For heavy users, the data involved is substantial—typically between 2 million and 15 million words of context.

The memory import transfers approximately 1,200 words. Memory Forge transfers the other millions of words. For the scale of migration happening right now, both steps matter.

Memory Forge is available at: https://pgsgrove.com/memoryforgeland