Beta

Provenance Analyser

What can actually be proven about who wrote a piece of text.

Coming soon

This one is built but not open yet. It runs several checks at once and reports them separately, and the most interesting of those depends on something that is not publicly available yet. Rather than ship it half-answering the question, it stays closed until it can answer properly.

CW

Vendor watermark

Some providers now mark the text their models generate. Reading that mark requires a key only the vendor holds, so no third party can check it. This check exists, stays switched off, and will not guess.

waiting
HC

Hidden characters

Invisible characters can be embedded in text to carry a payload. This reads them straight from the file, so a result here is a direct observation rather than an estimate. It also finds them when they arrive innocently, from a copy-paste or a PDF.

ready
ST

Writing style

Compares the writing against patterns learned from a large collection of human and machine-written text. Suggestive, never proof, and never allowed to stand in for the checks above.

ready

In the meantime, the AI Detector covers the writing-style half of this and is open now.

← Back to all tools