What Is an AI Detector?
An AI detector estimates how likely a piece of text was written by a language model rather than a person. Instead of reading for meaning, it reads for statistical fingerprints: how predictable the word choices are, how uniform the sentences run, and how closely the phrasing matches patterns that models produce by default.
The output is a probability. A score of 85% does not mean 85% of the text is AI-written; it means the text as a whole reads more like model output than human writing. That distinction matters whenever a score is used to make a decision about a person.
The detector currently supports English text only.
How AI Detection Works
Four things happen between paste and report:
- Predictability: the model measures how surprising each word is. Human writing takes odd turns; model writing tends to pick the likeliest next word.
- Sentence variety: people mix short punchy lines with long winding ones. Generated text often keeps an even, careful rhythm.
- Pattern matching: stock constructions ("it is important to note", "in conclusion") and hedged, symmetrical arguments raise the score.
- Scoring: the signals combine into an overall probability, a split, and a per-sentence highlight map you can verify against the text.
What Affects Accuracy
No detection is perfect, and text length is the biggest factor by far. Three things move the needle most:
- Length: a 300-word passage gives the model far more signal than a two-line answer. Below 150 characters we will not score at all.
- Editing: AI text that a person has substantially rewritten drifts toward mixed or human scores, which is usually the honest answer.
- Formulaic writing: lab reports, legal boilerplate, and heavily templated prose can read as AI even when a person wrote it. This is the main source of false positives.
Reading Your Report
Start with the verdict badge and the overall probability, then check the split: a 55% score built from a few strongly-AI paragraphs is a different situation than 55% spread evenly across the text. The sentence highlights are the evidence layer; read the flagged sentences yourself and ask whether the flags make sense. Copy the report or download it as a PDF if you need to share or archive the result.
Tips for Fair Results
- Check at least 300 words when you can; short snippets score unreliably in both directions.
- Never act on a score alone. Use the sentence highlights as a starting point for judgment, not a substitute for it.
- Re-run after edits. Scores are a snapshot of the text you pasted, not of its author.
Writing that needs to read more naturally can go straight to the AI Humanizer.