AI content detector
Paste English text to see how much it looks like AI writing – with the sentences and patterns behind the score. Free, no signup, and your text never leaves your browser.
🔒 Runs in your browser – your text is not uploaded to our server or anyone else.
Deep check more accurate
Runs a small open AI language model (DistilGPT-2) in your browser and measures how predictable each sentence is. Downloads about 85 MB the first time (then your browser keeps it). Reads the first ~1,100 words. Works best on a laptop or desktop.
Highlighted text
Tap or hover a sentence to see why it was marked.
What we measured
| Signal | Found | Points to |
|---|
This is an estimate, not proof. AI detectors – including well-known ones – sometimes flag human writing (especially formal writing or writers whose first language is not English) and miss edited AI text. Don't use this result alone to accuse anyone.
Questions
Can this tool prove that a text was written by AI?
No. No AI detector can prove it – ours, ZeroGPT, GPTZero or any other. They estimate how much a text looks like typical AI output. People who write formally can be flagged, and AI text that was edited can pass as human. Never use the result as the only evidence, for example to fail a student or reject a writer.
How does the check work?
The quick check looks at writing patterns that are common in AI text: very even sentence lengths, stock phrases such as "delve into" or "it's important to note", many transition words, lists of three, few contractions and no informal touches. The deep check runs a small open language model (DistilGPT-2) in your browser and measures how predictable each sentence is – AI text is usually very predictable and evenly so.
Is my text sent anywhere?
No. Both checks run inside your browser. The deep check downloads the open model files (about 85 MB, once – then your browser keeps them), but your text is never uploaded to our server or to anyone else.
Which languages are supported?
English only. Hindi, Hinglish and other languages are not scored – the patterns and the model are built for English, so a score would be meaningless.
Why does the deep check give a different score?
It measures something different (how predictable the words are) and counts for more in the combined score. DistilGPT-2 is a small, older model, so text from the newest AI tools can still look human to it. When the two checks disagree, treat the result as unclear.
How long should the text be?
At least 50 words; 150+ words give a steadier result. The deep check reads the first ~1,100 words.