Google Gemini Now Lets You Drag-and-Drop Your ChatGPT History

Google released two companion tools inside the Gemini web and mobile clients that let users pull their historical context from rival chatbots. The first is a guided prompt that asks your current assistant to “export everything you know about me” and then paste the resulting text into Gemini. The second accepts a zipped archive of JSON or markdown chat exports (up to 5 GB) and automatically maps topics, custom instructions, and so-called “memory” entries into Gemini’s own memory system. Both paths are live for English-language users and will expand to more languages “in the coming weeks,” according to Google’s product blog.

Switching chatbots has always meant trading months of accumulated context for a cold start. If you’ve trained ChatGPT to mimic your brand voice or remember your dietary restrictions, re-teaching that to Gemini felt like losing a notebook you’d annotated for a year. The new import flow collapses that reset time from days to minutes. Early testers told The Verge they spent roughly 90 seconds pasting a memory dump and saw Gemini “instantly mirror the tone and preferences” they had curated elsewhere. That removes the single biggest psychological barrier to experimentation.

For builders, the tooling is more than a convenience; it’s a user-acquisition API in disguise. Google quietly exposed a new endpoint, v1alpha/importContext, that accepts the same 5 GB zip format and returns a structured mapping of imported entities. Apps that embed Gemini can now offer one-click onboarding for users coming from OpenAI-powered products without handling raw chat logs themselves. Expect wrapper startups (think journaling or coding assistants) to bolt this on within days to shorten time-to-value for new sign-ups.

Google isn’t peeking into proprietary model weights. Instead, the pipeline treats the exported text as synthetic few-shot examples. When you upload a zip, Gemini extracts system messages, user instructions, and conversation turns, then runs a lightweight fine-tuning job that lasts 2–3 minutes. The resulting “user profile” is stored as a 128-dimension embedding plus a sparse keyword overlay—small enough to fit in the free tier’s context window. If you later delete the memory, Google purges both the archive and the fine-tuned layer, satisfying GDPR deletion requests without retraining the base model.

Neither company has announced a reciprocal export feature, but the pressure is now on. ChatGPT still only offers per-conversation downloads, not a bulk memory export. Anthropic’s Claude has an “AI memories” panel but no official way to ship those memories elsewhere. Industry analysts quoted by Bloomberg expect at least one of them to match Google’s move within a quarter, if only to stem churn. Until then, Google gains a rare asymmetric advantage: it’s easier to leave ChatGPT than to leave Gemini.

Privacy advocates note that exported logs may contain sensitive prompts, medical data, or proprietary code. Google addresses this by client-side encryption in transit and automatic deletion of the source zip after 24 hours, yet users must still trust that the parsing script won’t leak fragments into training corpora. There’s also a fidelity problem: nuanced style instructions sometimes get flattened into generic bullet points, so power users might still need manual tweaking. Finally, the 5 GB cap excludes heavy enterprise accounts with multi-year histories, a segment Google says it will tackle “later this year.”

Google just turned switching costs into a rounding error. By commoditizing context, it’s betting that model quality and ecosystem integrations—not locked memories—will decide the next wave of AI loyalty. If the gambit works, expect every major platform to race toward portable user states, making the entire market more like email providers than operating systems. For now, Gemini is the only chat app where your past travels with you.