ChatGPT or Claude? I run my business on both
If you work on your own and pay for one AI tool, you've probably wondered whether you picked the wrong one.
I use both, every day. But if I could only keep one for running my business, it would be Claude, because of the kind of work I do: lots of writing, research and thinking things through. That's my answer, not everyone's. If your days look different, yours might be too.
I run a small business part-time, and AI does a lot of the heavy lifting so I can spend my time on the work that needs me.
A quick note on what I mean by each. I mostly use Claude Code and Codex. They're the versions that work on the files on my computer, rather than the chat window on the website. Claude Code is Anthropic's, and Codex is OpenAI's, from the ChatGPT side.
How I ended up with both
I opened my ChatGPT account in January 2023, a couple of months after it launched. For a long time it was the only one I used.
Claude came in at the start of 2026, through Claude Code. By then ChatGPT's answers had got so much worse for the way I work that I moved over to Claude completely for a while, and built my whole working system there.
When ChatGPT got its act together, I brought it back in. Now both sit in the same setup.
There isn't one that does everything well
That's the real reason I use both. Each one is good at different things, and it's nice to be flexible.
On an ordinary day, Claude does most of the work. Researching, talking things through, brain dumps, ideas, writing, and the general business support that keeps everything moving. One morning this week, before I'd opened anything myself, it had read my whole inbox, cleared out what was already dealt with, checked my calendar and drafted the replies that needed me. I read them and sent them.
I go to Codex when I want a second opinion, when I want two versions compared, or when I need an image made. It makes the pictures for my blog posts. I get better results from it for those jobs. This week I had it read a blog post I'd written with Claude. It came back with ten points. I took eight of them, at least in part, and left two.
It changes fast, too. One minute everyone's using ChatGPT, the next it's Claude, then Google's Gemini is doing well, then Grok and the rest.
So it isn't only the two of them for me. Gemini comes with my Google Workspace account, I've got a model installed on my own laptop, and I use OpenRouter, which gets me to lots of different models through one account. I don't use them all the time. They're there because we mustn't be at the mercy of the big tech companies. We've no idea how stable they are or where they're heading, and a price change is all it takes to put a model out of the picture.
It's not about whether it gets things wrong
People ask me which one makes fewer mistakes. I don't think it's as black and white as wrong.
We say "wrong" as if AI should somehow be perfect and know. It doesn't, and it can't. Not yet, anyway.
So for me it's not what goes wrong, it's what I need to question. Every day I ask myself: what more do I need to know about what's been put in front of me? I ask questions, ask it to clarify, and ask for references I can check. And I push back on anything that sounds like opinion dressed up as fact, because these tools learn from what people have put online, opinions, judgements, biases and all.
The only exception is a plain automation that just picks up information and moves it from one place to another. Those need less watching, though I still check they're working. Everything else needs a person keeping an eye on it.
I stay in the driving seat
AI can give you back time. It gives me back a lot of mine. But don't take that so far that you hand it everything, including your thinking.
So whichever tool I'm using, some things stay with me:
- Sending. It drafts my emails. I send them.
- Anything going public. Posts, emails and web pages come to me first, and nothing goes live until I've said yes.
- Decisions. It brings me a recommendation. The choice is mine.
- My own story. Nothing personal about me goes into anything public unless I've said it can.
- Money and security. Payments, spending limits and account settings, I do myself.
I keep the parts where I'm the one answerable.
If you pay for one and wonder about the other
Learn one properly first. Know what it's good at before you pay for a second.
Then try the other on a real job, not a test. Give them both the same piece of work and compare. That's how I found Codex worked better for me for second opinions and images, and Claude for the day-to-day.
Don't tie yourself to one company. No one tool does everything well, and a price change can knock one out overnight. With a second, you're never stuck.
Keep your work in your own files, not inside the app. If you're starting out, that can just be a folder of notes and documents on your own computer that you give to whichever tool you're using. Mine has grown: Claude, Codex and the assistant I use from my phone all work from the same files on my own server. Either way, adding a tool, or dropping one, doesn't mean starting again.
Whichever you pick, you're still the one checking. A second tool is another opinion to question. It isn't a replacement for your own.
If you'd like help working out which suits your business, that's what my AI Power Hour is for. We take one question, and I talk you through it while you do the clicks on your own accounts.
Written in October 2026. What I use right now: Claude Opus 5.5 in Claude Code, and GPT-6.1 Sol in Codex. These tools change quickly, so I'll update this when my setup does.