Why I switched from ChatGPT to Claude

I dropped ChatGPT as my main tool and moved to Claude, but not because one is good and the other bad. I'll tell you what pushed me, what I gained, what I miss and what I still open OpenAI for. With luck you'll come away with a way to decide whether switching is worth it for you, and a way to do it without losing what you already have.

You've surely seen them: "ChatGPT is dead, use this", "this AI destroys everything". Every month there's a new champion, and whoever was behind jumps ahead as soon as a new model comes out. If I listened to those videos I'd change tools more often than socks, and spend more time moving house than working.

I did switch, but not because of a video. It was in spring 2025 and it was triggered by a very specific task: setting up a new server with several applications, which with ChatGPT meant copying and pasting commands for hours and going back and forth every time something failed. That's when it became clear I needed something else.

A warning from the start: this is my experience with my tasks, not a verdict. There's no "best" in the abstract; there's the one that fits your work best. What fits me may not fit you.

ChatGPT was my way in, like almost everyone's. I started with the free version in December 2022 and a few months later I started paying for the subscription, which I kept for nearly two years. I used it to write and proofread texts, summarise long documents, prepare proposals for clients and, more and more, to explain how to configure Linux servers.

And it worked, I say that without resentment. With it I set up my first Linux server without ever having touched that system. It explained things patiently, never tired of my silly questions and got me out of more than one jam.

My way of working back then was everyone's. I'd describe what I wanted to do on the server, it would give me a list of commands, I'd copy them into the terminal one by one, paste back the error, it would suggest a fix, I'd copy that in again… and so on for a whole afternoon. I was the go-between for the assistant and the machine: the errand boy running back and forth. If you recognise yourself in that scene, this article is for you.

It wasn't a spectacular failure, it was a pile-up of small stumbles. First, long conversations: past a certain length it started forgetting what we'd decided at the beginning and proposing things we'd already ruled out. Second, made-up facts: once it confidently gave me a configuration option that didn't exist, and I lost half a morning working out why it didn't work. Third, model changes: one day it answered beautifully and the next it felt like a different assistant, without me having changed anything.

Before switching I tried the usual advice. I fine-tuned the custom instructions, broke big tasks into small pieces and tried the different models it offered. Things improved a bit, but the underlying problem was still there: I was still the errand boy between the chat and the machine, and every trip was a chance to make a copying mistake.

If I had to put a number on it, I reckon I was redoing one in four technical tasks and losing about five hours a week going back and forth with errors. Not ChatGPT's fault, mind you: a good part was mine, for asking a chat to do what a chat isn't designed to do. But I didn't know that then.

What I found wasn't a better chat. It was Claude Code: a Claude that works directly on the files on my computer, reads a whole project, runs commands and checks what it does. In other words, an agent, which is what we call an AI program that takes several steps on its own to complete a job. I no longer had to copy and paste: I explained what I wanted and it did it, asked permission for the delicate bits and showed me the result.

As soon as it came out I said: "This is exactly what I need." Love at first sight. On top of that, I noticed it handled long documents better and its writing was closer to mine, with less filler.

The first task I solved with Claude that I hadn't managed with ChatGPT was migrating a client's old website to a new VPS: with ChatGPT I'd spent two afternoons stuck on permissions and certificates; with Claude Code it was up and running in about three hours, and most of that time was me reviewing what it did. I sat staring at the screen with a silly look on my face.

Now the practical warning, and it's a serious one. Claude Code touches real files. It asks permission before delicate actions, but if you give it free rein in the wrong folder, whatever it deletes is gone. Start in a test folder, with a backup and, if you can, with version control (a history that lets you go back). I learnt that by losing a configuration file I hadn't backed up.

What I've ended up doing with it I tell in I built a website with the help of AI and in I built my own tool without being a programmer.

I'm not going to idealise it. I miss having image generation and voice conversation as close at hand as I had them in ChatGPT; for some things I still have to step out to another tool.

Usage limits have bitten me too. In heavy weeks I've run out of quota mid-afternoon, right in the middle of a task, and had to wait or finish it with another assistant. I'm not giving you official figures because they change and depend on the plan; I'm telling you it's happened to me, more than once.

And it gets things wrong too. Once, fixing a bug on a website, Claude Code decided to "simplify" a part that worked perfectly and left it broken; I only caught it because I reviewed the change before publishing. An agent that acts can also act badly, and faster than you.

Finally, any comparison between ChatGPT and Claude goes stale in weeks. What one does worse today may be fixed next month, so don't trust tables with an outright winner, mine included.

I haven't left OpenAI completely. I still ask it for images to illustrate quick ideas, some searches where I like how it presents sources, and second opinions on important texts. There, for the way I work, it performs better or is simply more comfortable for me.

And there's one use I recommend even if you change nothing: cross-checking. When an answer matters (a quote, a technical decision, a text a client will read), I give it to both and compare. If they agree, I rest easier; if they disagree, I know I have to look closely myself.

You don't have to marry one. The reasonable thing is to have a main tool and a backup, just as you have a favourite screwdriver and another just in case.

What doesn't move over on its own is your conversations. What's in ChatGPT doesn't magically appear in Claude, so don't obsess over taking everything with you. What is worth taking, and what I did by copying it into a separate document, is your custom instructions, the templates you use often and the business details you'd been feeding it. You give them to Claude at the start and save yourself weeks of explanations. The step by step is in how to move from ChatGPT to Claude.

The two lived side by side for about six weeks. During those days I missed shortcuts and habits: where each button was, how I picked up an old conversation, how it answered certain questions. That's normal; you get used to it quickly.

My homespun advice: don't switch in the middle of an urgent job. Do it in a quiet week, with tasks you can repeat if something goes wrong.

Is switching worth it for you? It depends on your work. If you mostly work with short texts and it's going well, probably not. If you spend the day with long documents, or want the assistant to touch files and connect to other tools, Claude is worth trying. If your thing is images, look carefully first. With spreadsheets and code, compare using your own files, not the ones in a video.

The test I suggest is simple: one week with real tasks, on the free versions of both (ChatGPT's and free Claude), before paying for either. Keep the one that saves you the most time. Check the price in each app, which is where it's up to date; I'm not putting it here because it changes. And remember there's no universal answer between Claude and ChatGPT.

Verdict: the switch cost me about six weeks of living with both and the odd scare with files, and it saves me several hours a week that I used to spend running errands between the chat and the machine. For me it was worth it. For you, let your tasks decide.