Slop is the word for it now: that grey, bloated, everyone-has-read-this-a-thousand-times prose AI produces by default. “In today’s rapidly evolving landscape…” “It is important to note…” Nothing wrong, nothing right, nothing in it.

Why slop happens. The model aims to please and offend no one, so it averages everything it has ever read. Its most likely next word is usually the most generic one. No stakes, no audience, no consequence — so no edge. Slop isn’t a malfunction. It’s the resting state.

You beat slop with inputs and rules, not hope.

Ground everything in your material. When the model answers from your archive — real projects, real numbers, real site names — it has specifics to use, and specifics are the opposite of slop. Generic questions get generic answers. “Draft from these six reports” gets substance.

Ban the filler. Put the banned phrases in your instructions: “in today’s world,” “plays a crucial role,” “it is important to note.” Sounds petty. Works immediately. The model has endless alternatives once the defaults are blocked.

Demand the concrete. Every section needs numbers, dates, named examples, or a source citation to your documents. If a paragraph contains none of those, it’s slop by definition — cut it or fix it.

Give it your voice. Show it two or three examples of writing you’re proud of and say “write like this.” Style transfers surprisingly well.

Then edit hard. The first output is a quarry, not a building. Your job as the expert is cutting the marble. AI that saves you drafting time only pays off if you spend the saved time on the red pen.

The pattern underneath all of this: slop is what AI produces when it has nothing to say. Give it your knowledge, your data, your standards — and it becomes a drafting tool for work that sounds like you, because it literally comes from you.