The human edit is the moat

For a while, being able to publish a lot was an advantage. You could out-produce competitors who were still writing everything by hand. That window is closed. Everyone can produce volume now, which makes volume table stakes, not an edge.

What happened next was predictable. A lot of teams pointed a model at their keyword list and shipped hundreds of posts. Most of those posts rank for nothing and get cited by no one. They aren't bad, exactly. They're generic, unsourced, and indistinguishable from the other hundred versions of the same article. A pile of them is a liability, not an asset.

The lesson isn't that AI doesn't work. It's that the model was never the scarce ingredient. The scarce ingredient is everything around it.

The single highest-impact document in my pipeline is the brief, not the draft. A draft is cheap now. A good brief is where the actual thinking lives: who this is for, what question it owns, what angle nobody else took, what proof we have that competitors don't, and what the reader needs to believe by the end.

When the brief is sharp, the draft almost writes itself and the edit is light. When the brief is lazy, some version of "write 1,500 words about X," no amount of model horsepower saves it. You get 1,500 competent, forgettable words.

So most of the human judgment in my system happens before a single word is drafted. That is the opposite of how a lot of people use these tools, and it's most of why the output doesn't read like slop.

The fastest way to spot AI content is that it could be about anyone. Generic advice, no specifics, no point of view. The fix is grounding: every draft is fed the real product, the real customers, the real numbers, and the real opinions before it writes a word.

A piece about allergy-safe dining reads completely differently when the draft knows how the actual product makes its calls, what it refuses to claim, and where it deliberately hedges. It stops sounding like a summary of the internet and starts sounding like it came from someone who works on the problem.

This is unglamorous work. It's assembling the context that only you have and forcing it into every piece. But it is the difference between content that could have come from anywhere and content that could only have come from you.

Here is the rule I don't break: nothing publishes without a human reading it and checking the claims. Not skimming. Reading, with a specific job, which is to cut the filler, kill the sentences that sound like a press release, and verify anything stated as fact.

Models are confidently wrong in a very particular way. They'll invent a statistic that sounds right, attribute a quote to no one, or state a best practice that was never true. The edit exists to catch exactly that. It's also where voice gets added back: the aside, the opinion, the specific example, the thing that makes a person want to keep reading.

I run a de-slop pass as its own step, separate from the substance edit, because they catch different things. One checks whether it's true and useful. The other checks whether it sounds like a human wrote it. Both matter, and bolting them together tends to let things slip.

Almost everything in the pipeline can be automated: research, first drafts, formatting, internal linking, images, publishing. I automate all of it, and that's why one person can produce what used to take a team.

The one thing I never automate is the sign-off. A human decides it's good enough to carry the brand's name. Not because a model couldn't approve it, but because accountability is the whole product. When someone hires a content operation, they're paying for one person to be answerable for whether it's any good. The moment you automate that away, you've rebuilt the content graveyard with extra steps.

That is the moat. Not the model, which everyone has. The judgment about what deserves to exist, and the willingness to put your name on the result. Volume is free now. Taste isn't.