The Slop Score

How much of the writing is made of familiar AI patterns?

A public 0–100 benchmark of words and constructions that language models overuse. Lower is better.

Public leaderboard snapshot

Slop Score

November 7, 2025 · lower is better

Writing sourceScoreSample

Human baseline10.4
10.483 samples
Kimi K2 090518.3
18.3300 samples
Claude Sonnet 4.519.5
19.5150 samples
GPT-5 mini26.3
26.3300 samples
ChatGPT 4o47.7
47.7300 samples
Gemini 2.5 Flash77.6
77.6300 samples

Public benchmark snapshot dated November 7, 2025. Scores are calculated from standardized creative-writing and essay prompts.

Measurement

Three signals make the score

60%Overused words: Listed-word matches per 1,000 words
25%Contrast scaffolds: Repeated “not X, but Y” constructions per 1,000 characters
15%Overused three-word phrases: Listed phrase matches per 1,000 words

Scores are normalized against the benchmark cohort and can shift as that cohort changes. The benchmark uses 300 standardized prompts: 150 creative-writing tasks and 150 essays. Sample counts vary in this public snapshot.

Read the public methodology