Why I don't allow AI-generated content on my blog

I will never include any AI-generated content on this website. In general I’m a big AI believer, and I use large language models a lot: both in my day job, to help me program, and in this blog, to help me explore ideas and find sources. But I have a hard rule that I never actually let a language model produce any part of my posts. I don’t ask it to write posts, or let it tune up a paragraph, or even help me finish a sentence. Every word on this website is written by my human brain.

I don’t have this same standard for software. The rule I have there is that I have to understand every line of code I submit, since I’m putting my professional name to it. Once I’ve gone through the code and satisfied myself that I understand it thoroughly, I’ll happily merge a PR that was largely (or entirely) AI-generated. Why am I stricter with English language posts?

The first reason is that AI-generated English has a certain smell to it. The ChatGPT house style is always present. It’s at its worst in 4o, but even non-OpenAI models have it. The snappy sentences and the overuse of rhetorical opposition… it’s slop, even if the content is useful. I have a visceral reaction to being made to read slop unknowingly. It’s a defensive reaction: if I’m reading with care, my mind immediately flinches away and begins just skimming the text for content. I never want to inspire that reaction in my readers.

Would I be willing to use future models that don’t have this style? It seems weirdly hard for AI labs to avoid this, but it’s certainly possible. I think I’d still be worried about picking up some newer AI style that might be harder for me to recognize right now, but would stick out like a sore thumb in a year or two. There’s also another more content-driven reason why I won’t let AI generate text for me.

That second reason is that I write when I have something to say. Often that something is a bit controversial or off-putting:

• Tech companies contain predators that will try and steal time from competent-but-naive engineers
• Good engineering work is work that serves the company’s interests, not just technically sound work
• Particularly for more senior roles, you actually have to be correct, not just epistemically virtuous
• Getting promoted as a software engineer means trying to impress people you may not respect

Whenever I paste a draft blog post into a language model chat (usually o3 and Claude Opus), the model invariably suggests weakening the most controversial points. Surely I didn’t mean to say “predators”, isn’t that too harsh? Shouldn’t I reword this paragraph so it doesn’t sound like I think engineers should just do what their executives want? But I did mean to say that, and I do think that! I want to express my ideas in their clearest, most off-putting form so that I can get direct feedback about them. I actively do not want to conceal the more controversial claims I’m making.

People write for a lot of different reasons. I write so that I can get clearer about my own beliefs, so that I can get feedback on how I see the world of software engineering from smart people who know exactly what I believe, and hopefully so that more junior engineers can avoid some of the early mistakes I made¹. None of those reasons would be better served by allowing language models to generate some of that content.