Perplexity
LLMs pick predictable words. We estimate predictability with compression entropy: text that compresses well against everyday English is more machine-like.
Find out which sentences were likely written by ChatGPT, Gemini, Claude and other LLMs — with a confidence score for every line.
Improve your writing's flow and variety using sentence highlights.
LLMs pick predictable words. We estimate predictability with compression entropy: text that compresses well against everyday English is more machine-like.
Humans mix short and long sentences; models write in an even rhythm. Low variation in sentence length and structure raises the AI score.
Overused vocabulary (“delve”, “tapestry”, “pivotal”), stock phrases, formulaic transitions and lists of three are classic LLM tells.
Sudden shifts in sentence length, vocabulary level or British/American spelling between paragraphs can reveal pasted passages.
No detector is 100% accurate. Scores are probabilities, not proof — formal human writing, non-native English and very short texts can be misclassified. Use results as one input alongside human judgement.