What the text detector measures
- Burstiness: people mix short and long sentences; language models tend toward uniform length.
- Stock phrasing: words and transitions that appear far more often in LLM output ("delve", "tapestry", "it is important to note").
- Vocabulary diversity: how many distinct words appear relative to the total.
- Human tells: contractions, first-person voice and small informal slips.
Limitations you should know
Every AI text detector, including this one, makes mistakes. Edited or paraphrased AI text can pass as human, and formal human writing (or writing by non-native speakers) can look AI-like. Treat the percentage as a hint, and never as proof of misconduct.
Read more: Can AI text detectors be trusted?