Gemini watermark
Google marks text in the Gemini app and web experience with SynthID-Text, a sampling watermark published in Nature and released as open source. It changes how words are chosen, not which characters appear.
In short
Gemini marks text in its app and web experience with SynthID-Text, in production since 2024. It shapes which words are chosen and inserts no character. Google does not document marking for text returned by the Gemini API.
Two things you can do with your own text
Free, and it runs in your browser: nothing is sent. It finds characters, not a watermark in the word choices.
Try it without an account: three rewrites, up to 3,000 characters each, and a trial result is not saved. We show how much of your draft changed and promise no detector result.
What the evidence says
Google DeepMind announced text watermarking for the Gemini app and web experience in May 2024, and the Nature paper published that October describes it as already running in production — by the authors’ account, the first deployment of a generative text watermark at scale. Note the scope: Google documents the app and web experience. It does not document watermarking for text returned by the Gemini API, AI Studio or Vertex AI, so this page does not claim it.
The method is called tournament sampling. Candidate tokens drawn from the model’s own distribution compete in successive rounds, scored by pseudorandom functions derived from a key and the preceding context. SynthID-Text does not affect training and modifies only the sampling procedure, and detection runs without the underlying model.
That last point settles the question this database exists to answer. The Nature paper separates generative watermarking, which shapes the text as it is written, from edit-based watermarking, which alters text after the fact. SynthID-Text is the former and never edits a finished string, so no character is inserted. Google publishes no statement about Unicode specifically; the conclusion follows from the published mechanism rather than from a denial.
Quality was measured rather than asserted. Across nearly twenty million Gemini responses, the thumbs-up rate differed by 0.01 percent and the thumbs-down rate by 0.02 percent between watermarked and unwatermarked models, both statistically insignificant. The reported latency cost was 0.57 percent per token.
The limits are specific, and finer than the usual summary. Cropping, changing a few words and mild paraphrasing are survivable; a thorough rewrite or a translation into another language greatly reduces detector confidence. Watermarking is also less effective on factual answers, where there is less room to vary wording without losing accuracy. There is still no public way to check a passage: the SynthID Detector portal accepts images, video and audio, and text detection, announced as imminent in May 2025, is absent from the product page as of this check.
A common misreading
SynthID is often described as a hidden tag or an invisible character added to the output. The published method modifies only the sampling procedure and never edits finished text, so running a Gemini answer through a Unicode cleaner leaves the watermark exactly where it was. Changing the sampled words is the only edit that reaches it, and DeepMind’s own limits say as much: mild paraphrasing survives, a thorough rewrite greatly reduces detector confidence.
Sources
- Scalable watermarking for identifying large language model outputsNature 634(8035), 818–823 · · Peer-reviewed
- SynthID: identifying AI-generated textGoogle DeepMind · · Provider documentation
- SynthID: Tools for watermarking and detecting LLM-generated textGoogle AI for Developers · · Provider documentation
- SynthID Detector: a new portal to help identify AI-generated contentGoogle · · Provider documentation
Frequently asked questions
Can I check whether Gemini wrote a passage?
Not publicly. The SynthID Detector portal accepts images, video and audio; text detection was announced as coming in May 2025 and is not offered on the product page as of this check. Detection also needs the key and configuration, so no general scanner can substitute for it.
Does SynthID change how the text reads?
Measurably little. Across nearly twenty million Gemini responses, thumbs-up and thumbs-down rates differed by 0.01 and 0.02 percent between watermarked and unwatermarked models, both statistically insignificant.
Does editing or translating the text remove it?
It depends how far you go. Cropping, changing a few words and mild paraphrasing survive. A thorough rewrite or a translation into another language greatly reduces detector confidence.
Related tools
Related reading
Other assistants in the database
Working on your own draft?
GenPolish rewrites text you already have, showing every change before you accept it. A watermark lives in the word choices, so replacing the words is what reaches it. We show what share of your words changed, and make no claim about detectors.