Sourced and dated
Guides to AI text watermarks
Most writing on this subject confuses invisible Unicode characters with statistical token watermarks, and both with AI detectors. These guides keep them apart, cite primary sources, and record the date each claim was last verified.
Start here
The mechanisms, and what removes them
How AI Text Watermarks Work
How hidden-character marks and statistical token watermarks differ, how detection works, and why removal claims should always be bounded.
Updated · 4 sources
Hidden Unicode vs Statistical Text Watermarks
The two mechanisms confused most often: invisible Unicode characters versus keyed token-choice watermarks. What each one is, what removes it, and what neither proves.
Updated · 4 sources
How to Remove AI Watermarks From Text
A step-by-step method: identify which mechanism you are dealing with, remove hidden characters exactly, rewrite the statistical layer properly, then verify.
Updated · 3 sources
Mechanisms
How the signals actually work
What Is SynthID Text?
SynthID Text explained: how Google's text watermark is applied during generation, how detection works, what weakens it, and how it differs from image watermarking.
Updated · 4 sources
AI Watermarks vs AI Detectors vs Provenance Metadata
Three different things routinely confused: keyed watermarks added at generation, statistical AI detectors that guess, and C2PA metadata attached to files.
Updated · 4 sources
Provider status
What each provider documents
Provider Text Watermark Status Tracker
Which AI providers document text watermarking, by what mechanism, whether a public detector exists, and when we last checked the primary sources.
Updated · 7 sources
Does Claude Watermark Text?
Anthropic's August 2026 documentation on Claude text watermarking: the mechanism, its scope, what removes it, and what it cannot prove.
Updated · 3 sources
Does Gemini Watermark Text?
What Google documents about SynthID Text in Gemini: the mechanism, the deployment claim, the absence of a public detector, and the stated limits.
Updated · 3 sources
Does ChatGPT Watermark Text?
What OpenAI documents about provenance for text, images and audio — and why invisible characters in copied ChatGPT text are not evidence of a watermark.
Updated · 3 sources
Tools
Or go straight to the workspace
Deterministic character inspection is free and unlimited. The rewrite is what addresses a statistical watermark.
- AI Watermark RemoverText only · free to try
- Gemini Watermark RemoverSynthID Text · documented
- SynthID Watermark RemoverMechanism-specific
- Claude Watermark RemoverDocumented as of August 2026
- ChatGPT Watermark RemoverCopy artifacts · no text watermark documented
- Invisible Character RemoverFree · runs in your browser