AI Content Detector documentation
The AI Content Detector estimates how likely a text is to be AI-generated from 19 writing-style signals. It runs entirely in your browser: the text you paste is never sent to a server or saved.
Quick start
- Open the AI Content Detector.
- Paste at least 80 words, or open a .txt file. 300+ words works best.
- Press Check text.
- Read the score and confidence, then the evidence and the highlighted sentences.
Use Try an AI example or Try a human example to see how each kind of writing scores.
Every signal and its weight
Positive weights point toward AI, negative weights toward a person.
| Signal | When it counts | Weight |
|---|---|---|
| AI-favored vocabulary | 12+ per 1,000 words (6+ for the smaller weight) | +2.4 / +1.3 |
| Avoids AI vocabulary | 1.5 or fewer per 1,000 words, 100+ words | −0.9 |
| Stock phrases | 4 or more (2 or more for the smaller weight) | +2.2 / +1.1 |
| Chatbot leftovers | Phrases such as "I hope this helps" | +3.5 |
| Pasted chat formatting | 2+ **bold** labels or # headings | +1.6 |
| Even sentence lengths | Burstiness under 0.30 (under 0.42 for the smaller weight) | +2.0 / +1.0 |
| Varied sentence lengths | Burstiness over 0.65 (over 0.52 for the smaller weight) | −1.4 / −0.6 |
| Transition openers | 14%+ of sentences (8%+ for the smaller weight) | +1.5 / +0.7 |
| Em dashes | 6+ per 1,000 words and at least 3 | +1.0 |
| Lists of three | 8+ per 1,000 words and at least 3 | +0.9 |
| Identical paragraph lengths | 5+ paragraphs with variation under 0.20 | +0.6 |
| Repeated sentence starts | Over 30% begin with the same word | +0.6 |
| Personal experience | 3+ references (1+ for the smaller weight) | −1.8 / −0.8 |
| Casual language | 2+ casual words, emoji or !! marks | −1.4 / −0.5 |
| Typos or lowercase starts | A common misspelling or 2+ lowercase sentence starts | −1.3 |
| Fragments | 12%+ of sentences have 4 words or fewer | −0.9 |
| Contractions | 18+ per 1,000 words | −0.7 |
| Concrete specifics | 15+ numbers, prices or dates per 1,000 words | −1.1 |
| Plain wording | No stock phrases, no chatbot leftovers and few transition openers | −0.6 |
How the score and confidence work
The weights are added together and turned into a likelihood with a logistic curve: score = 100 ÷ (1 + e^(−total ÷ 2.2)). A total of 0 gives 50, so no single signal can decide the result.
| Score | Verdict |
|---|---|
| 80 to 100 | Very likely AI-generated |
| 60 to 79 | Likely AI-generated |
| 41 to 59 | Mixed signals |
| 21 to 40 | Likely written by a person |
| 0 to 20 | Very likely written by a person |
Confidence grows with length (full credit from 600 words) and with how strongly the signals agree. It is halved when most of the text is not in the Latin alphabet, because the vocabulary signals are English.
Privacy
- The detector is plain code that runs in your browser. Your text is never sent to our server, never saved and never used for training.
- Opening a .txt file reads it on your device only.
- Like every free tool on this site, it keeps no record of what you check. Page visits are counted without the text or any address you enter.
Limits
- No AI detector can prove who wrote a text. Treat the score as one clue among several.
- Formal writing, non-native English, templates and heavily edited AI text can land in the middle or be misread.
- It is tuned for English.
- Under 80 words there is not enough text to judge.