What the Slop score is

Every Desloppy check gives you one number, the Slop score. It runs from 0 to 100%. It tells you how much a text sounds like default chatbot writing. That is all it tells you. A low score means the text sounds natural. A high score means it sounds like it came straight from the chat window. The score never says who wrote the text. Many people write with AI, edit well, and score low. Some people write slop by hand.

The bands

The score comes with a verdict word, so you can read it at a glance.

• 0 to 10: Clean.
• 11 to 30: A little sloppy.
• 31 to 60: Sloppy.
• 61 to 89: Very sloppy.
• 90 to 99: Pure slop.
• Exactly 100: Peak slop.

A check passes when the score is under 50 and Desloppy found no strong stock phrase and no passage that reads mostly like a chatbot. A passing text gets "Good to go". Desloppy then leaves it alone unless you ask for touch-ups.

Two readings, one number

Desloppy reads a text in two ways. The first is a detection model we train ourselves. It reads whole passages and gives a probability that the passage sounds like AI writing. The second is a library of sloppy habits: stock phrases, scene-setting openers, tidy morals at the end of a paragraph, and the rest. Each hit has a severity and a plain fix.

Until 19 October 2026 the score was the model's reading alone. That had a problem. You could fix every sloppy line Desloppy pointed at, and the score would hardly move, because the model had its own opinion of the passage. So we now blend the two. Fixing the sloppy lines lowers the score.

The blend depends on length. From 150 words up, the model's reading counts in full. At 60 words or under, the model's reading counts for 60% and the sloppy lines fill the rest. Between 60 and 150 words the weight slides from one to the other. On top of that, a pile-up of stock phrases adds to the score at any length. The result on short AI text: fixing every sloppy line in an AI text under 80 words lowers its score by about 22 points on average.

The sloppy-lines side has a rule of its own. One stock word on its own never counts. People write "I hope this finds you well" in work emails all the time, and novelists used the fancy words long before any model did. Hits count only when three or more different habits stack up in one passage of about 300 words, or when a line is plain chatbot boilerplate. We measured that rule on 3,000 hand-written texts and 1,739 modern AI texts. The share of hand-written texts shown an AI phrase fell from 24.7% to 1.8%.

The blend showed up in our test against other detectors. On 218 humanized AI texts, the model alone caught 63.8%. The blended Slop score caught 72.0%. The full results are in Desloppy vs six AI detectors on humanized text.

Accuracy by length

Length matters more than anything else. We measured the live model on 19 October 2026 on 6,499 AI texts from current models and 8,157 hand-written texts, with rewritten and humanized AI text included. "Caught" means a Slop score of 50 or more.

• Under 80 words: 30% of AI texts caught, 0.4% of hand-written texts wrongly flagged.
• 80 to 199 words: 74% caught, 0.2% wrongly flagged.
• 200 to 499 words: 95% caught, 0.6% wrongly flagged.
• 500 words or more: 96% caught, 0.2% wrongly flagged.

So check at least 200 words for a reliable result. Under that, Desloppy marks the result as low confidence and tells you to treat the score as a hint. Slides and lists are also hard, because they are fragments. There the line-level fixes are worth more than the number.

By kind of text, the same model caught 95% of everyday LinkedIn posts written by five AIs, 97% of real-world AI posts and answers, including the newest 2026 models, 93% of AI from model families it never trained on, and 74% of AI text that was paraphrased, rewritten or run through a humanizer. Short chatbot answers under 80 words are the weak spot, at 28%. On the other side it wrongly flagged 3 of 303 student essays, 0.8% of short social posts, 0 of 1,000 short replies and 1 of 1,500 web pages, news articles and tutorials. Every group of hand-written text stays under 1%.

Before and after

When you edit and check again, Desloppy compares the two scores. A move of under 5 points counts as noise, and the card says the score barely moved. A rewrite that scores 5 or more points higher gets a plain verdict: keep your original. A chatbot that rewrites a whole draft tends to put its own default voice in, and that is the voice the score measures.

What the number is for

The Slop score is a probability. It can flag text written by hand, especially short text, plain styles and English from people whose first language is something else. Use it to find passages worth another pass. Never use it as evidence against a person. The full results are on About Desloppy. You can try 1,000 words free, with no account, at desloppy.com.