Google opens SynthID Detector: what it can tell you about AI media

On October 7, Google announced public access to SynthID Detector, launching globally in English. The synthid.com portal checks images, video and audio for supported SynthID watermarks. The tool had previously been tested with journalists and media professionals.

What it looks for

According to Google DeepMind’s explanation, SynthID adds an imperceptible mark when content is generated. The mark is embedded in the media itself and designed to survive common changes, including compression. The public Detector covers supported content from Google and partners; it is not a universal test for every generator.

SynthID technology also has a text branch. However, the new public portal’s description names images, video and audio. That does not mean you can paste any ChatGPT or Claude article into it and obtain a reliable verdict about its author.

No watermark does not prove human origin

Google’s help page for verification in Gemini notes that content can originate from another AI system, and repeated modifications can sometimes make a watermark undetectable. This helps explain the limits of the approach. Gemini and the separate Detector portal should not be assumed to have identical partner support or usage rules.

Our interpretation is that a result provides evidence about a particular mark, not an automatic verdict on the truth of the story. A real photograph can carry a false date; an honestly labelled educational illustration can come from a generator. Both require context.

Applying it to a questionable news post

First separate the file from the claim: what does the author want you to accept as fact? Find the earliest publication, check the date and location, and compare the account with independent sources. A watermark check can add provenance information, but it cannot replace those steps.

In particular, do not extend trust in an image to a request for money or a verification code. For that situation, see our guide to recognizing AI phishing.