What slop sounds like

The one-line definition

The features that compound into slop

None of these are wrong by themselves. A single tricolon is fine; a personal essay can use a hedge. Slop emerges when several of these stack at once, evenly, throughout the text:

• Hedging without earning it."Often," "may," "tends to," "can sometimes", used as default, not because a real claim is being made under uncertainty.
• No first-person stance. The writer doesn't appear. The text could be attributed to any organization or person without changing.
• No concrete particulars. No proper nouns, no dates, no numbers, no named example. Could be the front of any article on this topic.
• Tricolons everywhere."Faster, cheaper, smarter." "Bold, agile, data-driven." A rhythm reinforced by RLHF preference data, which rewards copy that sounds confident without committing to anything testable.
• "It's not just X, it's Y." The reframe construction. "Less about discipline and more about clarity." "It stops being a tool and becomes a partner."
• Both-sides closers. The final paragraph commits to nothing it didn't already imply: "Ultimately, the answer depends on context."
• Uniform paragraph shape. Every paragraph is three to four sentences, sentence lengths cluster in a narrow band, no rhythmic surprise.
• Em-dash density. Used as a default beat, not for the specific interruption it's good for.
• Relentless positivity. Nothing has a downside; no choice has a cost; every option "empowers," "unlocks," "elevates."

The 0–100 scale, in plain language

• 0–20 · Reads human. Concrete particulars and a stance the writer would defend, including the downside. Children's books, literary essays, opinionated argument. Most published writing pre-2022 lives here.
• 20–40 · Mostly clean. A real voice with some generic patches. Most decent newsletters and personal blogs.
• 40–60 · Sloppy in places. The voiceless register peeks through. Corporate memos, mid-quality marketing, the "clean but generic" band.
• 60–80 · Heavy slop. LinkedIn thought-leader posts, SEO blog filler, AI listicles. Confident-sounding; says nothing testable.
• 80–100 · Industrial-grade slop. What a default-prompted LLM produces on a marketing brief: cliché stacked on reframe stacked on both-sides closer.

Slop is not the same as low quality

This is the load-bearing point. A 2023 study by Zhang & Gosline (MIT Sloan / Berkeley Haas, 1,203 participants) found that blind readers rated AI-generated marketing copy higher than human-expert copy on both satisfaction and willingness-to-pay. The slop register works for ad copy because the reader came for clarity about a product, not a voice.

Slop becomes a problem when the genre is one where voice is the value: personal essays, opinion writing, ghostwriting — anywhere a reader wants to feel that one specific person is talking to them. For those genres, high slop is failure. For a product page, the same register is often the right tool.

How to read a high score

The Slop Score is decomposable. If your composite is high but the breakdown is uneven — lexical at 80, structural and epistemic at 30 — read it as a register signal rather than a slop verdict. Formal academic prose, L2 English, and technical writing all trigger it. Read the per-engine breakdown before you rewrite. The known limitations page covers the cases where the headline number is most likely to mislead.

The calibration set, in public

Every reference text we pin the engines against is on the calibration page, in full. If our definition of slop doesn't match yours, the anchors will tell you that faster than any argument here will. Read them before you decide whether the score deserves a hearing on your own writing.