BlogText

SynthID-Text · statistical signals

Do AI text watermarks exist — and can you remove them?

SynthID-Text and statistical detectors work very differently from image watermarks. Here's how AI text signals work, what humanization can and can't do, and where the honest limits are.

Text watermarking is real but works nothing like an image watermark. Systems like SynthID-Text subtly bias which tokens a model chooses, creating a statistical pattern a matching detector can score later. There is no pixel to edit — the “signal” is the distribution of word choices across the passage.

Two different things people mean

  • Watermarks — a deliberate statistical bias inserted at generation time (e.g. SynthID-Text).
  • Detectors — third-party classifiers guessing “AI or human” from stylistic tells, with no watermark required.

They call for different responses. A watermark is a distribution to disturb; a detector is a stylistic profile to move away from. Rewriting affects both, but neither is a guaranteed clear.

What humanization can and can't do

Rule-based humanization rewrites stock AI openers and high-frequency tell-vocabulary and varies sentence rhythm. Optional neural paraphrase (a T5-class model, when weights are present) restructures more deeply. Both reduce statistical tells — but the more you preserve the original wording, the more of any watermark you leave intact. It's a genuine trade-off, not a solved problem.

Try it on your own file

Best-effort, multi-layer, and reported honestly — originals always preserved.

Create an account