Word by word, an AI detector that tells AI generated from human written

PaperBleach is a free AI detector that estimates how likely your text was written by AI. Most detectors highlight a whole sentence; PaperBleach scores every word, so you can see exactly what to change. Paste an essay, article, or email and get an AI-probability score with a sentence-level and word-level breakdown, tuned against how detectors like Turnitin, GPTZero, and Originality.ai behave — then humanize the flagged text if you need it to read more naturally.

An AI detector estimates how likely a piece of text was written by an AI model like ChatGPT. It measures signals such as perplexity — how predictable each next word is — and burstiness — how much sentence length and complexity vary — and returns a probability score rather than a yes/no verdict. PaperBleach's AI detector shows that score alongside a colour heatmap painted over your own text: green reads as human, amber reads as AI, and the opacity carries the confidence, so passages near the decision boundary appear pale. You can toggle between the per-sentence view and the per-word view to find the exact phrases driving the score. How it works: sign in, paste between 50 and 2500 words, and run the check. You get one document-level AI probability plus the sentences and words that look most AI-generated, then you can rewrite the flagged text with the built-in AI Humanizer and scan again to confirm the score moved. Reading your score: below about 15% the text reads as human; 15-30% is borderline and worth revising; 30% and above is likely to be flagged. No detector is perfect. They give a probability, not proof of authorship, and false positives are documented — a 2023 Stanford study reported that detectors misclassified a large majority of TOEFL essays by non-native English writers as AI-generated, and formulaic, technical or heavily cited prose is flagged often. Use the sentence-level and word-level breakdown to judge the result. Your first AI check is free when you sign up.