AI Memory Portability — March 2026
Gemini, ChatGPT, Claude.
Your Memory. Your Choice.
Three AI platforms shipped memory portability features in March 2026. Google and Anthropic added import tools. OpenAI shipped export only.
OpenAI, Google DeepMind, and Anthropic each shipped memory portability features within a 30-day window in March 2026, but not the same features. OpenAI launched ChatGPT memory export as a JSON file containing stored facts and preferences, and export is all it shipped: there is no OpenAI tool for importing a competitor’s history. Google added Gemini memory export to Google Takeout and paired it with two import tools. Anthropic released structured memory export from Claude.ai alongside an import path for ChatGPT archives. The convergence was not coordinated. It was driven by European data portability rules that have made export functions table stakes since GDPR Article 20 took effect in 2018, and by competitive pressure from users demanding the ability to switch AI platforms without losing conversational context.
What Actually Transfers and What Does Not
Transfers successfully: Explicit stored facts (name, job, location), user-stated preferences (format, length), professional context (role, industry), stated goals and ongoing projects.
Does not transfer: Implicit tone calibration from past conversations, task-specific context built over multiple sessions, model-specific reasoning style learned from feedback, conversation history (only summaries, not transcripts).
How the Import Mechanisms Work
Google built two distinct import tools for Gemini. The first is a memory transfer tool: users export their preferences, relationship context, and behavioral patterns from ChatGPT or Claude as a text block, paste it into Gemini, and Gemini ingests the information into its memory system. This is a lossy transfer because it captures stated preferences but not the implicit patterns that emerge from thousands of conversations.
The second tool is a full chat history import via ZIP file upload. Users export their complete conversation history from ChatGPT (Settings, Data Controls, Export) or Claude, upload the archive to Gemini, and Gemini processes the conversation transcripts to build a user profile. This is a higher-fidelity transfer because it captures the actual conversations, not just a summary. However, the processing is one-directional: Gemini reads the transcripts to understand your preferences and communication style, but it does not import the conversations as accessible chat history you can reference.
Why Anthropic Shipped the Same Feature First
Anthropic launched its history import feature three weeks before Google, accepting ChatGPT export archives and building Claude’s memory from the conversation data. The timing was strategic: Anthropic recognized that switching costs are the primary barrier to AI assistant migration. If a user has invested months of conversations building a relationship with ChatGPT, moving to Claude means starting from zero context. The import feature reduces switching costs from “lose everything” to “lose some nuance.”
OpenAI has not shipped an equivalent import tool. This is the competitive dynamic the import tools reveal: the companies that are gaining market share (Anthropic, Google) are building migration tools. The company that is losing market share (OpenAI, to some degree) has no incentive to make migration easier. The absence of an OpenAI import tool is itself a competitive signal.
Why No Shared Standard Exists Yet
OpenAI exports memory as a JSON object with key-value pairs mapping fact categories to stored values. Google uses a similar JSON structure but with different field names and taxonomy. Anthropic exports as structured markdown with labeled sections. None of these formats are interoperable without a converter. Importing ChatGPT memory into Claude requires a manual reformatting step or a third-party tool. This is not accidental. Each company has an incentive to make import easy but export minimally useful, since true frictionless portability reduces switching costs and accelerates churn.
The AI platforms made memory portability sound like a major unlock. In practice, most users who switched from ChatGPT to Gemini did not lose anything they could not recreate in two conversations. The real switching cost is not stored facts: it is learned behavior and task context that accumulates over hundreds of sessions. That context is not exported by any platform and cannot be reconstructed from a JSON file. Memory portability solves a compliance requirement and a PR problem. It does not solve the actual lock-in mechanism.
The EEA Restriction and What It Means
Google’s import tools are not available to users in the European Economic Area. This restriction reflects the regulatory complexity of processing personal data transferred between AI platforms under GDPR. When a user exports their ChatGPT history and uploads it to Gemini, the data crosses organizational boundaries. GDPR requires a legal basis for processing, purpose limitation, and data minimization. Google’s compliance team apparently concluded that the current implementation does not meet these requirements for EEA users.
The EEA restriction previews how data portability regulation and AI competition will interact, and the legal instruments involved do different things. GDPR Article 20 gives users the right to receive their data and transmit it elsewhere. It binds the exporting controller, not the receiving one, so nothing in it forces Gemini to ingest a ChatGPT archive. The Digital Markets Act’s portability provision, Article 6(9), requires designated gatekeepers to provide effective, continuous, real-time export of user data. Its interoperability mandate in Article 7 covers number-independent messaging services, not AI assistants. As of August 2026 the designated gatekeepers are Alphabet, Amazon, Apple, Booking, ByteDance, Meta, and Microsoft across 23 core platform services, no AI chatbot is among those services, and OpenAI and Anthropic hold no designation at all. The import tools are therefore voluntary product decisions, not legal obligations. The nearest binding DMA action points in a different direction entirely: in July 2026 the European Commission issued specification decisions under Article 6(7) requiring Google to give rival assistants, including ChatGPT and Claude, equally effective access to Android features that Gemini uses. That is an obligation on the gatekeeper’s operating system, not a mandate that any AI service import a competitor’s chat history. Whether generative AI becomes a designated core platform service is an open question in the Commission’s ongoing DMA review, which makes any import mandate analysis about a possible future, not a description of current law.
What the Portability Race Reveals About the Market
The AI portability race tells you that AI assistant providers believe switching costs are their primary retention mechanism. If the product alone were sufficient to retain users, portability tools would be unnecessary because users would not want to leave. The investment in import tools is an implicit admission that AI assistants are becoming interchangeable enough that users will switch for marginal improvements in quality, price, or features.
This is the commoditization signal. When AI assistants competed on raw capability (which model is smartest), switching costs were high because capability differences were large. As models converge in capability, the competition shifts to switching costs, pricing, ecosystem integration, and user experience. Portability tools accelerate this convergence by removing the one remaining barrier to switching: accumulated context. The AI assistant market in 2026 is transitioning from a capability competition to a user experience competition, and the import tools are the evidence.
Sources: OpenAI memory export documentation; Google Gemini data portability blog; Anthropic memory feature release notes; GDPR Article 20; Digital Markets Act Articles 6(7), 6(9), and 7; European Commission DMA gatekeepers portal; March 2026. Updated 2026-08-18: corrected the lead, which described all three companies as shipping import capabilities when OpenAI shipped export only, and rewrote the Digital Markets Act analysis to distinguish GDPR Article 20 portability, DMA Article 6(9) export obligations, Article 7 messaging interoperability, and voluntary product import features, none of which currently obligates any AI service to import a competitor’s history. Added the July 2026 Commission specification decisions on Android access for rival assistants as the current state of DMA enforcement touching AI.