A Treatise on Writing in the AI Age, Part 4

This story won’t resonate with everyone, but I’ve heard it enough to know that it’s uncomfortably common. You’ve published something you worked hard on. It could be a blog post, a twitter thread, a video, a course, some sort of content that you’ve put time and attention into. Then weeks or even days or hours later, you find your idea living in someone else’s post. I’m not talking about having the same idea as someone else. I’m talking about someone ripping your ideas, presenting them as their own, and taking credit for them. There’s no citation, link, or acknowledgment that you even exist.

On many platforms, there’s a strong stance against plagiarism, including removal of the post, suspensions, and bans. Let’s look closely at what makes that a good policy. It targets the harm itself. It’s checkable in the artifact. And it never asks what tool was involved. Steal by hand or steal by model, the violation is identical, because the reader’s injury is identical.

Now hold the AI disclosure mandate next to it. No harm requirement. Nothing checkable in the artifact, which is why enforcement runs on appearance. And the tool isn’t incidental to the rule. The tool is the rule. The same platform wrote both policies, and only one of them protects anyone.

That asymmetry is the tell. Platforms have decided that a machine touching your own ideas is a policing priority, while a human taking someone else’s ideas is a shrug. Whatever problem they think they’re solving, it isn’t reader betrayal, because this is what betrayal actually looks like.

What actually harms readers

Set aside tools entirely and ask what makes writing bad for the person reading it.

Ideas presented as original when they were lifted from someone uncredited, claims stated confidently that nobody verified, content written to game an algorithm rather than to say anything. Padding, filler, the eight-hundred-word wind-up before the answer. Recycled conventional wisdom wearing a bold headline.

Every one of those predates AI. Every one of them is thriving right now, by human and machine hands alike. And every one of them is measurable in the artifact itself, no process forensics required. You don’t need to know what tool produced an unverified claim to know it’s unverified.

Protecting readers from those harms would mean citation norms with teeth. Plagiarism response that actually responds. Quality signals that reward substance. And I want to be fair here, because that’s brutally hard work. Slow, expensive, full of judgment calls, and most platform teams are small, stretched thin, and staring down a content flood they didn’t create. I have real sympathy for anyone holding that backlog.

A disclosure checkbox is none of those things. It ships in a sprint.

The fear response

I don’t think platform teams are cynical. I think they’re afraid, and I understand why. AI dropped the cost of producing plausible text to nearly zero, and every open publishing platform is staring at a real flood of low-effort content. Something had to be done.

But look at what “something” became. The flood is a quality problem. Low-effort content was always the enemy, and AI just made it cheaper to produce. The honest response is to get better at detecting low effort. That’s hard. So instead, platforms reached for a proxy. AI produces the slop, therefore mark the AI.

The proxy fails in both directions. It misses the humans who were producing slop by hand all along and will keep producing it, badge-free. And it catches the writers using AI carefully in service of real thinking, who are producing exactly the substance the platform claims to want.

We’ve seen this pattern before, and not just in writing. When a real problem is hard to measure, institutions measure something adjacent and easy, then defend the metric instead of the mission. Every developer who’s been judged on lines of code knows how this story goes. The metric becomes the target, the target gets gamed, and the original problem sits there untouched.

The problem we’re not allowed to name

There’s one more layer, and it’s the uncomfortable one.

Some of the resistance to AI-assisted writing isn’t about readers at all. It’s about writers, and about identity. If writing well took years to learn, and a tool now closes part of that gap for others, the instinct to defend the moat is human and understandable. We’ve never accepted “it threatens my position” as an argument in open source, in education, in any community worth belonging to. Gatekeeping dressed as quality control is still gatekeeping. The developers who sneered at Stack Overflow users, the photographers who sneered at digital, the writers who sneered at bloggers. History doesn’t remember the moat-defenders kindly, and it definitely doesn’t remember them as right.

The question was never how to keep writing hard. It was how to keep writing honest. Those are different projects, and the disclosure mandate serves the first while claiming the second.

So if tool-marking is the wrong standard, what’s the right one? What would a norm look like that catches the plagiarist and the fabricator, protects the newcomer and the craftsperson, and doesn’t care which keyboard the sentences came through?

That’s the last essay, and it’s the one this whole series has been walking toward.