AI slop

AI slop is text that reads as obviously machine-made: smooth on the surface, empty underneath. It is fluent, grammatical, and confident, and it says almost nothing. The sentences are all the same shape, the word choices are safe, and the detail could describe anything. Slop is the default a language model reaches for when nothing pulls it toward the specific.

The tells

Readers learn to spot it fast. The usual signs:

• Generic phrasing. Lines that could appear in any article on any topic.
• Padding. Words that add length without adding meaning.
• Flat rhythm. Sentences of one length, marching.
• Hedging. Everything qualified, nothing claimed.
• Em-dashes and stock connectors scattered where no writer would put them.
• Vague description. A room, a feeling, a person, none of them seen.

Why models produce it

A model trained on the whole internet regresses toward the average of the internet. Without strong steering it gives you the most probable next sentence, which is to say the least surprising one. That is the opposite of writing, where the specific and the surprising are the entire point.

How to write with AI and avoid it

Slop is a coherence and specificity problem, and both are fixable. Anchor the model to real material: samples of the voice you want, a story bible full of concrete facts, and characters with actual wants. Edit against the tells with a line pass. Keep a person making the final calls. FireQuill is built around exactly this: it strips the obvious tells automatically and keeps generated prose anchored to your bible and your voice, so what comes out reads like you wrote it. See the full method in how to write a novel with AI without it sounding like AI.