The AiEdge Newsletter

The AiEdge Newsletter

Deep Dive: How AI Text Watermarks Work

How a language model leaves a detectable pattern in its word choices, and why copying the text preserves it.

Damien Benveniste's avatar
Damien Benveniste
Sep 11, 2026
∙ Paid

An AI text watermark leaves detectable evidence in the words a model chooses. During generation, a secret scoring rule helps select the next piece of text. Later, a detector with the matching key can reconstruct that rule and check whether the finished passage agrees with it unusually often.

The surprising part is that an exact copy into a plain-text document can preserve the evidence. There is no hidden character to carry along. The ordinary words preserve the choices.

An exact copy of the support reply preserves the same six reconstructed scores, shown separately from the text.
The dots are a detector overlay. Copying the same text preserves the inputs to that check.

If you use AI to draft, proofread, or rewrite, the distinction matters: those operations leave different amounts of freedom for a watermark. And when someone says a passage was “detected,” the test matters too. A keyed watermark check measures a deliberately introduced pattern; it does not infer AI use from a writer's style.

User's avatar

Continue reading this post for free, courtesy of Damien Benveniste.

Or purchase a paid subscription.
© 2026 AiEdge · Privacy ∙ Terms ∙ Collection notice
Start your SubstackGet the app
Substack is the home for great culture