How this AI detector works
AI-written text tends to leave statistical fingerprints: sentences of oddly similar length, a narrower vocabulary than human writing, evenly sized paragraphs, few contractions, and a fondness for certain stock phrases — "delve into", "in today's fast-paced world", "a testament to" and dozens more. This tool measures six of those signals, weighs them, and shows you exactly what it found, signal by signal. Nothing is hidden behind a mystery percentage, and nothing you paste ever leaves your device.
How to use it
- Paste at least 80–100 words — longer samples give steadier results.
- Press Analyze text and read the overall signal score.
- Open the signal table to see why the score is what it is — sentence rhythm, vocabulary variety, phrase hits, repetition, paragraph shape and contractions.
- Check the highlighted sentences: solid highlights contain known AI stock phrases; dotted underlines mark suspiciously uniform sentence lengths.
Use the result responsibly
No detector on the market can prove authorship, and independent studies have repeatedly shown that both free and paid detectors mislabel some human writing — especially short texts, technical writing, and text by non-native English speakers. Treat a high score as a reason to look closer (ask about drafts, sources and revision history), never as evidence on its own. Equally, a low score doesn't guarantee a human author: edited or "humanized" AI text can pass any detector.
Frequently asked questions
How accurate is this AI detector?
Like every AI detector — free or paid — it provides indications, not proof. It transparently measures six statistical signals typical of AI text. Short texts, technical writing and non-native English can trigger false positives, so treat scores as a starting point for judgement.
Can it detect ChatGPT, Gemini and Claude?
The signals it measures (uniform sentence rhythm, stock phrases, low vocabulary variety, even paragraphs) are common to text from all major AI models, so it flags AI-typical writing regardless of which tool produced it.
Will my text be saved or used for training?
No. The analysis runs entirely in your browser; the text is never uploaded, stored or shared with anyone.
Can teachers use this to fail a student?
Please don't use any detector score alone for that. No detector can prove authorship, and false positives are well documented. Use results as one signal alongside drafts, revision history and a conversation with the writer.