AI Detector

What the detector looks at

The detector weighs patterns that tend to appear in AI-generated text against patterns that tend to appear in human writing. On the AI side: very uniform sentence length and rhythm, generic transitions, hedging boilerplate, list-like structure in prose, characteristic AI phrasing, and a lack of specific personal detail. On the human side: idiosyncrasy, typos, specific lived detail and uneven rhythm.

You get a score from 0 to 100, a confidence level, a one-line verdict, and the signals themselves with the evidence quoted or described. The signals are the most useful part: they show you what the estimate is based on, so you can judge it rather than take a number on faith.

An estimate, not proof

No AI detector can tell for certain who wrote a text, and this one does not claim to. Careful human writers can produce smooth, uniform prose; edited AI text can look human; non-native speakers and very formal writing are especially easy to misjudge. A score is one piece of evidence, never a verdict.

Do not use the result on its own to accuse anyone of anything. If a result worries you, treat it as a reason for a conversation: ask the writer about their process, look at drafts and notes, or ask them to explain the work in their own words.

Sensible ways to use it

Teachers can use it as a prompt for a conversation about how a piece was written. Writers can check whether their own draft reads as generic and then revise it. Editors can flag a submission for a closer read. In every case, the signals are where to look — not the score alone.

• Longer texts give stronger evidence than short ones
• Read the signals, not just the number
• Combine the estimate with drafts, notes and a conversation

Example: how a result reads

Each check returns a score from 0 to 100 for how likely the text is to be AI-generated, a confidence of low, medium or high, a verdict — likely human, mixed or unclear, or likely AI — and three to six signals. Every signal says which way it points and quotes or describes the evidence, such as evenly sized sentences and generic transitions on one side, or a specific personal detail and uneven rhythm on the other.

Read the signals before the score. A high score with low confidence means very little; a medium score with signals you can see for yourself in the text is a reason to have a conversation, not to reach a conclusion.

Limits to know about

Short texts give weak evidence. Under about 150 words, the detector is told to keep the score near the middle and report low confidence. Texts that are heavily edited, translated, written to a strict template or written by non-native speakers can show the same signals as AI writing.

Up to 12,000 characters of text, or a PDF, DOCX, DOC or TXT file up to 10 MB, can be checked. The detector is an estimate from the writing itself; it does not look anything up and cannot prove who or what wrote a text.

• Never use a score as the only evidence in a grading decision
• Compare with the student's earlier writing and drafts
• Ask the writer to talk through how they wrote it