The Cost of Switching: An AI Lock-In Experiment (Post 2 of 6)
The Illusion of Portability
When I started this experiment, I assumed moving my history from one AI to another would be straightforward. After all, we live in a world where data exports and imports are supposed to be standard. Download, upload, done.
But that’s not what happened.
I quickly discovered that ChatGPT’s export isn’t really portable. What you get isn’t a clean set of text documents, it’s a massive JSON blob hidden inside a JavaScript file inside an HTML file. My Chat.HTML file was over 100mb which isn’t big per say, but the HTML is hiding a massive amount of text and structure. Claude doesn’t have an import function at all. Gemini can technically read from Google Drive, but my files were too large to handle.
So I found myself writing command-line scripts, parsing JSON with jq, and splitting files into dozens of smaller chunks. At one point, I was chasing down corrupted text blocks that refused to upload. This wasn’t portability. This was surgery.
And even after I got the files moved, I realized something bigger: the content itself wasn’t enough. Years of context weren’t really “transferrable” because there was no way to bring along the shared understanding I had built with ChatGPT. It was like switching doctors and handing them a box of medical records. They will have the paperwork, but they don’t know me and they can’t read all the records.
Research backs this up. The AI Adoption Paradox report (source) calls this the illusion of portability the idea that you can carry your history with you, when in fact, your most valuable asset is the fluency you’ve built with a specific system .
So here’s the takeaway: in AI, your data may move, but your trust and workflow don’t. That’s where the real lock-in lives.
If you could export everything from your current AI assistant today—but none of the shared understanding or “fluency” you’ve built, would you still feel like you were bringing your work with you, or would it feel like starting from zero?
