Spot the fingerprints AI leaves in text.
AI writing tools quietly use typographic Unicode characters most people never type — narrow no-break spaces, smart quotes, em dashes. Paste any text to see them highlighted, then clean them in one click.
Everything runs in your browser — your text is never sent to any server.
How it works
This is a transparent, character-based heuristic — not a statistical AI classifier. Large language models are trained on professionally typeset text, so they reproduce its typography: curly quotation marks (U+201C/U+201D), em and en dashes (U+2014/U+2013), the horizontal ellipsis (U+2026), and above all the narrow no-break space (U+202F), which almost no human types by hand. Finding many of these in a supposedly hand-written text is a strong hint.
It is a hint, not proof. Text pasted from Word or a CMS can carry smart quotes too, and a careful writer may use em dashes deliberately. Conversely, absence proves nothing — these characters disappear with one pass through a plain-text editor. Treat the result as one signal among several.
The cleaner normalizes everything back to plain keyboard characters: straight quotes, simple hyphens, three dots, regular spaces — and strips invisible characters (zero-width spaces, joiners, the byte-order mark) that can break search, diffs, and code.
FAQ
- Does this detect AI text reliably?
- No tool does. This one shows you verifiable evidence — actual characters present in the text — instead of an opaque percentage score, and lets you judge.
- Why does AI use these characters?
- Training data: books, articles, and websites are professionally typeset. Models learned that polished text uses "smart" typography, so they emit it — even where a human would type straight quotes and hyphens.
- Are invisible characters dangerous?
- They can be disruptive: zero-width characters break string matching, watermark text, sneak through filters, and corrupt code snippets. The cleaner removes them all.