ChatGPT for content creation: what I learned drafting a whole blog with it

TL;DR: I ran the experiment most people searching this are about to run: I set up ChatGPT for content creation the way you’re supposed to – custom GPTs, project folders full of my voice and process, step-by-step drafting workflows – and produced a lot of posts for my hobby blog with it. It worked, barely, and it fought me the whole way. I’m Julie Kaiser, a science writer, hobby blogger, and AI operator, and this is the honest map: what the setup looked like, where it broke, and what actually matters more than the tool. As of this writing, because models change fast.

Can ChatGPT write blog posts?

Yes – I made it do exactly that, earlier this year, right before I built my current system. And I didn’t do it casually. I had custom GPTs. I had project folders holding how I write, my SEO conventions, my voice, my topics, my framing and philosophy, who I am. I had step-by-step processes designed to keep the chatbot on track from outline to draft.

I was doing everything you’re supposed to do.

Where it broke

Four ways, over and over.

It drifted. ChatGPT could see all the files in the project, and instead of staying on the post at hand, it would wander into the other topics and start talking about those too. We kept drifting, and I didn’t like it.

It ignored the plan. I’d say: we’re doing this section by section. It would write the whole thing in one go anyway – or write a pile of material that had nothing to do with the outline we had already agreed on.

It repeated itself. I was never able to get longer pieces that didn’t circle back and say the same things over and over.

So I took the work back. I abandoned the full-draft ambition and went piece by piece: I’m in charge of the outline, I decide what goes where, help me with this paragraph. That worked – and it was cumbersome enough that I kept asking myself: is this really faster than writing it manually? A little bit. That was the honest answer. A little bit.

The feeling I remember most is: I’m doing everything you’re supposed to do. Why isn’t this working?

Why it broke (I found out later)

It was probably the context window. All those files and step-by-step processes I was forcing on it filled the conversation’s working memory – and ChatGPT gave me no view of that window. I had no idea where I was, at the beginning of it or past the end. The feedback is missing, so you can’t manage what you can’t see. Why an AI forgets as the window fills is its own post.

And everything has to do with the models. What some models can’t do, other models breeze right through.

The wall that made me switch

The clearest moment came from a different project in that same hobby world: video transcripts for a course. I had written them myself – strong drafts, a firm idea of what they should be. But there was repetition from one video to the next, and tightening them meant auditing each transcript in the context of all the others. I couldn’t hold that in my brain. And ChatGPT couldn’t do anything with it either.

When I brought those same files to Claude, it could see the whole set at once – what needed to change in each transcript in the context of everything else. The bigger picture. ChatGPT, at least as I was using it then, couldn’t hold that bigger picture.

I moved my whole system shortly after, and I canceled my ChatGPT subscription – the math is in what I pay for AI. Take what you want from that about brands. The durable lesson is about context: the tool that can see your whole project is the one that can help you with it.

What can ChatGPT do today?

Honestly: I don’t know, and I won’t pretend to. I haven’t used it since I moved. I hear good things about the newest model – GPT-5.6, announced as I write this, though it hasn’t reached regular ChatGPT users yet – and maybe I should test it when it does. Custom GPTs and project folders presumably still work roughly the way they did when I built mine. If you’re using ChatGPT today, weigh my experience accordingly: it’s a true story, but it’s a dated one, and AI moves lightning fast – and everything has to do with the models.

How do content creators use AI without sounding generic?

This is the question underneath all of it, and it matters more than which tool you pick. AI for content creation is sold as a shortcut, and used as one, it produces exactly the generic content everyone can already smell. The AI content creation tools will keep changing. What doesn’t change:

Own the outline. Whoever owns the outline owns the piece. The moment I took the outline back from ChatGPT, the work got better – that rule survived the tool switch and runs my system today.

Keep the judgment. The sentences, the cuts, the call on whether a piece is ready – if I hand those off, what comes back is generic, and I think that’s obvious to my readers even when they can’t name what’s wrong with it.

Say how you work. Every post on this blog is AI-drafted and I say so – the disclosure is its own page, and the method has a name. The system that makes it mine instead of generic is the one I built after ChatGPT.

I’m one person running a content operation with a lot of AI in it and my judgment on every piece. That’s the whole trick, and no subscription includes it.

The one rule

Capabilities move monthly; the principles don’t. Own the outline, keep the judgment, disclose how you work. Hold those three, and any sufficiently capable model is a tool. Hand them off, and the best model in the world will produce content that sounds like nobody.