Skip to content
Simple Unmark

Simple Unmark guide

Provider Text Watermark Status Tracker

Published
Sources checked
7 primary references

Short answer

Google documents SynthID Text for generated text. Anthropic documents a SynthID-Text-style token-choice watermark in Claude output, applied globally from August 2026. OpenAI documents provenance for images and supported audio, but no text watermark. None of them offers a public detector for arbitrary text today.

Provider behaviour changes, and most content about it goes stale silently. This page records only what primary sources support, with the date each claim was last verified.

How to read this table

"Documented" means a first-party page from the provider supports the claim. "Reported" means credible coverage exists without first-party confirmation, and "unverified" means neither. We do not upgrade a claim on the strength of press coverage alone.

The verification date is when a person last re-read the linked sources, not when this page was deployed. Provider status is reviewed monthly, and within 72 hours of a major announcement.

What is true across all of them

Three patterns hold regardless of provider, and they are more useful than any single row.

  • Where text watermarking is documented, it is statistical and adds no characters. Both Google and Anthropic describe token-choice methods, and Anthropic states explicitly that nothing is added and there are no hidden characters.
  • No public detector exists for arbitrary text. Google publishes its approach and a reference implementation; Anthropic says a detection API is planned. So third-party claims to verify removal should be treated as unsupported.
  • Thorough rewriting is the only documented lever. Light edits, synonym swaps and mild paraphrasing are explicitly described as things these methods survive.

What would change these rows

An OpenAI text watermark shipping to production models would change the ChatGPT row from "not documented" to a live mechanism. Anthropic's detection API becoming public would, for the first time, let anyone measure a claim about removal — including ours, which we would then benchmark rather than describe.

Our methodology page explains how we would test that, and what we would refuse to claim without it.

Status by provider

Text watermark status by AI provider, with mechanism, detector availability and verification date.
FieldGoogle GeminiAnthropic ClaudeOpenAI ChatGPT
Text watermarkIn productionRolling outNot documented
EvidenceFirst-party documentedFirst-party documentedFirst-party documented
MechanismA logits processor applied during generation. Google describes it as augmenting the model's logits with a pseudorandom g-function, so the watermark lives in which tokens were chosen rather than in any added symbol.A key-based pattern in low-stakes token choices. Anthropic describes using the key plus the preceding words to settle which word the model picks, and states the approach is a version of the SynthID-Text method published in Nature.OpenAI's published provenance work combines C2PA Content Credentials with SynthID watermarking, and its verification tooling covers supported images and audio. Text is not included.
Hidden characters?No. SynthID Text does not insert hidden characters; nothing is appended to the visible text.No. Anthropic states plainly that nothing is added to the text and that there are no hidden characters.No documented hidden-character watermark. Invisible characters in copied ChatGPT text usually come from the surrounding copy-and-paste path, not from a provenance system.
Public detectorGoogle publishes the watermarking and detection approach and has open-sourced a reference implementation, but a public detector for arbitrary Gemini text is not offered as a general consumer service.Anthropic says a watermark detection API will be offered but that implementation details are still being worked out, so no public detector was available at the verification date.OpenAI offers verification for supported images and audio. It retired its own AI text classifier in 2023 and does not publish an equivalent text checker.
Last verified18 August 202618 August 202618 August 2026

Google Gemini: what weakens it

  • Google states detector confidence "can be greatly reduced when an AI-generated text is thoroughly rewritten"
  • Translation into another language
  • Factual responses, where there is less opportunity to augment generation
  • Short excerpts, though the method is robust to cropping and mild paraphrasing

Anthropic Claude: what weakens it

  • Anthropic states light editing probably will not remove it, while a complete rewrite in which every word is replaced will
  • Text Claude only lightly edited, because few of the words are Claude's
  • Factual passages and code, where there are fewer free choices to encode
  • Small samples, where detection does not work well

OpenAI ChatGPT: what weakens it

  • Not applicable: no text watermark is documented, so there is no published text signal to weaken

Sources

Published by Simple Unmark. Our methodology page explains how claims on this site are sourced, what we test, and what we refuse to claim.

Apply this to your own text.

Three guest cleans are free, with no account required. Deterministic character inspection is free and unlimited.

AI Watermark RemoverClaude Watermark RemoverGemini Watermark Remover

Keep reading