How the AI detector works
The detector looks at a text in two ways. First, an AI model reads it and judges how likely it is that another AI wrote it, giving a score from 0 to 100% and a few reasons. Second, your browser measures simple writing signals: how long the sentences are, how much their length varies, and how many stock phrases AI models overuse (“delve”, “furthermore”, “it is important to note”).
Reading the score
| Score | What it suggests |
|---|---|
| 0–29% | Likely written by a person |
| 30–69% | Unclear, or a mix of AI and human writing |
| 70–100% | Likely written by AI |
Why no detector is proof
Every AI detector, including the paid ones, makes mistakes. Short texts, very formal writing and English written by non-native speakers are often wrongly flagged, and edited AI text can pass as human. That is why this tool shows its reasons and highlights sentences instead of giving a yes or no answer. Teachers and employers should never accuse someone on a detector score alone.
Worked example
A 300-word essay in which every sentence is 18 to 22 words long and which uses “furthermore” and “in conclusion” will usually score high and show “Low” sentence variety. A personal story with short and long sentences, names and specific details usually scores low.
If your own writing is flagged, remember that the score is only an estimate. Keeping your drafts and notes is the best way to show that your work is your own. To check spelling and punctuation instead, use the grammar checker.