The Cost of Switching: An AI Lock-In Experiment (Post 3 of 6)
Trust and the “AI Fluency” Lock-In
By this stage of the experiment, I thought I had done the hard work. I had exported my ChatGPT history, parsed it into text, and split the files into smaller pieces. Surely now I could start testing Claude and Gemini in real-world use.
Except Claude never even got out of the gate.
No matter how much I trimmed, split, or reformatted, Claude simply couldn’t handle my exported history. I tried plain text, Markdown, even PDF. I sliced the data into 20+ smaller files. Each time, the window was too small. It was like trying to pour gallons of water into a shot glass—technically possible one sip at a time, but completely impractical.
That left me with Gemini as the only workable alternative. Gemini could at least ingest the files via Google Drive (though only after I got deep into command-line work to split and clean them). But even then, I ran into the next barrier: trust.
After two years with ChatGPT, I had built a kind of fluency with it. I knew what to expect from its answers, how to phrase prompts, and when to push for clarification. With Gemini, I didn’t have that muscle memory. Every output carried a question mark. Was the summary incomplete, or did I just not set the context right? Did it miss a detail, or was I failing to prompt properly?
This is where the real cost shows up. The AI Adoption Paradox calls it the Trust Tax the hidden time spent checking, editing, and validating outputs until they feel reliable . Once you’ve paid that tax for years with one tool, starting over with another feels daunting.
The data confirms this: 3 in 4 workers abandon AI mid-task because of accuracy concerns, and nearly half say they don’t trust AI-assisted work from colleagues . It’s not just about what the model can technically do, it’s about whether you’ve built enough fluency to trust it.
For me, Claude never even made it to that stage. Gemini did, but the trust gap was so wide. Which means the gravitational pull back to ChatGPT is strong. There are so many little things:
ChatGPT has custom GPTs - Gemini has Gems.
ChatGPT has projects - Gemini doesn't, Claude does but it means starting over
ChatGPT has image creation - Gemini has nano-bananna :-)
ChatGPT has agent mode - Gemini is working on Opal (still in experiment)
If one of your go-to AI tools suddenly became unavailable tomorrow, would you have the patience to rebuild fluency and trust from scratch in another?
