People ask me which AI tool is the best one, like there's a single right answer sitting in a drawer somewhere. There isn't. I touch three of these tools most weeks, and by now I've got a pretty firm read on what each one is for.
ChatGPT gets the messy first pass. Claude gets the version I'm willing to ship. Gemini gets opened maybe once a month, and I'm annoyed within five minutes.
None of them are bad tools. I just stopped treating them like interchangeable chat boxes a while back.
Where ChatGPT Earns Its Keep
ChatGPT is where a thought gets to be ugly for a minute. I'll dump half a sentence, a rant, three unrelated bullet points, whatever's rattling around, and it'll meet me there without asking me to organize first. That's the whole value of it for me: low friction, quick back and forth, no setup required before it's useful.
It's also just a good front door into working with a language model at all. If someone in one of my workshops has never touched one of these tools past some little chatbot that popped up on a website once, ChatGPT is where I point them first. The interface doesn't ask anything of you. You open it, you type like you're texting a smart friend, and you get something back.
I don't build anything real in it, though. Not documentation, not a workflow, not anything I'd hand to a client. It's where the idea starts, not where it finishes.
Where Claude Takes Over
Once an idea has actual shape, I move it into Claude, and that's where the real work happens. Documentation, structuring a multi-step process, turning a rough outline into something a client can follow along with. This is the stuff that has to hold together across a few thousand words, and Claude is the one that doesn't lose the thread halfway through.
My read on it, and this is just an opinion from using it daily and not something out of a spec sheet, is that a lot of what Claude was originally built and trained on leaned hard into handling long documents and giving back something structured on the other end: big reference material, long-form summarization, that kind of work. It shows. I can hand it a messy fifteen-page brain dump and get back something I can hand off as-is, not a summary that flattens the whole thing into three bullet points.
It's not effortless, to be fair. Claude wants a project set up, files organized, more upfront structure than just opening a tab and typing. ChatGPT will meet you at zero effort. Claude asks for a little more before it gives you the good stuff.
Why I Can't Get Behind Gemini
I want to like Gemini. I really do. This is what did it for me. I gave it a reference doc and asked for a clean markdown file out the other end, nothing fancy, just formatted text I could drop into another system. What I got back had links to the original file scattered through it, ones I never asked for and couldn't get it to drop no matter how I phrased the follow-up. Not a one-time formatting quirk either. It just kept adding them back.
That's a small thing on paper, but it turns a five-minute task into twenty minutes of cleanup, and once that's happened two or three times, I stop reaching for the tool at all.
The Real Split
So here's where I've landed: ChatGPT for brainstorming and getting oriented, Claude for the deep work and anything that has to be structured and correct, Gemini for basically nothing unless I have no other choice.
That's not a permanent verdict. Models change every few months, and I'll happily eat my words if Gemini fixes the thing that's been bugging me. But that's where things stand right now, and it's the setup that's saved me time instead of costing me more of it. That's the stack. Use each one for what it's good at.
- Kan
