← All Notes

LinkedIn is drowning in AI-generated content. You already know what it looks like: five lessons from a founder who's never shipped, three ways AI will change everything from an account that's never built anything, motivational paragraph blocks that sound profound until you read them twice and realize nothing was said. The fingerprint is always the same — smooth, structurally correct, completely generic. Swap "leadership" for "innovation" anywhere and the post still holds. That's how you know a model wrote it. It optimizes for form while being agnostic about substance.

But here's what the discourse keeps getting wrong: the problem isn't AI. The problem is that nobody's editing it.

The Junior Reporter

A good editor at a major paper does not just publish whatever lands on her desk. A junior reporter files a draft. The editor reads it. She asks: Is this accurate? Is this interesting? Does it sound like something a human who lived through an experience actually wrote — or does it sound like a press release? She cuts the corporate softening, restores the specificity, and sends it back if it doesn't pass.

Most people using AI for content have fired the editor.

They prompt the model cold, get a draft, skim it, hit publish. The model does what it was designed to do — produce fluent, agreeable, well-structured text. Text that offends nobody. Text that hedges everything. Text that sounds like a synthesis of a thousand LinkedIn posts because that's literally what it is. And since nobody is reading it against the standard of did a real person think this, experience this, and mean this, it goes out.

The AI isn't failing. The editorial process is missing.

A newsroom desk covered in marked-up pages, a red pen resting on top, the evidence of an editor who reads carefully

Garbage In

AI is a tool. Garbage in, garbage out — always was, always will be.

The correct workflow is not: open ChatGPT, type a topic, publish the result. The correct workflow is the one good writers use with good editors: you write the seed. Your experience. Your angle. The thing you actually know and believe and have friction with. The part that's yours.

Then you give that to the model. Not "write me an article about AI on LinkedIn." Give it your draft — two sentences or twenty, doesn't matter — and let it help you build from what you brought. The model can expand, sharpen structure, find transitions, fill in gaps. What it cannot do is supply the thing you didn't give it: a point of view earned from being somewhere and doing something.

Then you read the output like an editor. Not like someone relieved to have a draft. Like someone who knows what they actually think and is checking whether the text matches it. The model will sand your edges. It's trained to. It will soften the confrontational sentence. It will hedge the claim you intended to be direct. It will add a balancing paragraph to the thing you meant to be unbalanced. Your job is to catch that and put the edges back.

If you seed it with nothing, you get nothing back dressed up in paragraphs.

If you seed it with something real and edit it like you mean it, you can get good work.

What Getting Sanded Looks Like

You write: "Most product managers are decorative."

The model produces: "While product managers play an important role in many organizations, their effectiveness can vary significantly depending on the team structure and company culture."

That sentence is not wrong. It is also not anything. It has been processed into harmlessness. The point — your point, the one you had a reason to make — has been converted into a position no reasonable person could object to, which means no reasonable person will find it interesting.

This is the sanding. It happens at every edge. Every strong claim becomes a nuanced observation. Every direct sentence becomes a paragraph with caveats. Every thing you meant to say becomes a thing that gestures toward what you meant while carefully not saying it.

The model isn't doing this maliciously. It's doing it because its training rewards agreeable output. Your job as the editor is to reverse it — everywhere it matters.

Close-up of a page with heavy red pen marks, crossed-out lines, marginal notes, the evidence of real editorial attention

The Actual Standard

The test for whether AI content is worth publishing is the same test that's always applied to any content: did a real person think this, and does the writing carry evidence of that?

Not: was it written by a human? That's an increasingly meaningless question. AI wrote significant parts of this ecosystem. The question is whether there's a mind behind it — someone who brought knowledge, angle, and friction to the table and used the tool to express it rather than to substitute for having anything to express.

LinkedIn's slop problem isn't an AI problem. It's a publishing-without-editing problem. It's been around since the blog era, the content-marketing era, the ghostwriting era. AI just made it cheaper and faster and more uniform.

The tool is neutral. What you bring to it is not.

Use it the way a good writer uses a good editor: as a collaborator who helps you say what you mean more clearly, not as someone who decides what you mean for you. Keep your edges. Push back when the output softens something you meant sharp. The work is still yours. The voice is still yours.

That's the only version of this that's worth reading.

— J.P. Howlett

Related:

Sources