AI Slop Detector
Paste a chat reply, a story or a post. You get a slop score from 0 to 100, every stock AI phrase highlighted, and a guess at which model it sounds like. It runs in your browser, and nothing you paste leaves this page.
Scores update as you type.
Paste something above and the score lands on one of these five cards.
- 0-15
Artisanal
Specific, varied, and surprisingly human.
- 16-35
Home-cooked
Some familiar habits, but mostly fresh.
- 36-60
Microwaved
Repetition starts to show. Characters and scenes blur together.
- 61-80
Cafeteria
The same phrases, reactions, and story beats keep coming back.
- 81-100
Pure Slop
You can practically predict the next sentence.
How the score works
The detector checks your text two ways. First, exact rules: the stock phrases of AI roleplay, companion and assistant replies, the words and shapes Wikipedia's editors list in its "Signs of AI writing", and em dashes. A structural tell like three short sentences in a row counts 0.5 of a hit and an em dash counts 0.5, because people write those on purpose.
Second, near-misses. Each sentence with no exact hit is compared against 123 real lines people reported, using BM25, a search-ranking formula. A sentence that shares at least 2 content words with a line and covers 70% of its weight is flagged, and counts 0.5 of a hit.
The score is hits per 100 words on a curve that reaches 100 at 14, so the first few tells move it most. Texts shorter than 100 words are scored as if they were 100 words long, so one stray phrase in a single sentence does not read as Pure Slop. The tiers are the same ones every app review on this site uses.
Every phrase and its source is in the AI slop words list, by model. How we score whole apps is on How we test.
Questions people ask
Does the AI slop detector send my text anywhere?
No. The detector runs in your browser. Nothing you paste is uploaded, stored or logged, and it keeps working if you go offline after the page loads.
Can it tell whether AI wrote something?
No, and nothing can with certainty. It counts the stock phrases and shapes AI models overuse, so a high score means the text reads like default AI output, whoever typed it.
How does it guess which model wrote the text?
Each stock phrase in our catalog is tied to the model its source names. When at least 2 tells point at one model more than any other, the detector names it and links to that model's list.