Google expands SynthID verification portal to check AI-generated content

On October 7, 2026, Google announced it was expanding access to SynthID: anyone can now go to synthid.com to check whether a file carries a SynthID watermark. This is the most direct action in the announcement—taking a verification capability originally embedded in products and spinning it out into a standalone public-facing entry point.

The objects this entry point can handle are image, video, or audio files, and its purpose is to determine whether they are AI-generated. It does not start from zero: Google explicitly states that the new portal sits alongside built-in verification features already available in Google Search, the Gemini app, and the Chrome browser. In other words, the public portal supplements existing verification touchpoints rather than replacing them.

The files the portal can detect are not limited to those generated by Google itself. Google named industry partners including OpenAI, NVIDIA, and Kakao, as well as Apple, which is 'coming soon.' This list determines the portal's actual coverage: it seeks to be a cross-vendor verification entry point, not a closed-loop tool that only recognizes its own watermarks.

But the list itself also exposes boundaries. Apple is described as 'coming soon,' indicating that support for its files is not yet live; as for whether files from OpenAI, NVIDIA, and Kakao can all already be detected by the portal, the source does not specify. In other words, the partner list is a mix of intentions and current status, and readers should not treat it as a fully effective mutual recognition list.

In the announcement, Google gave three scale numbers: it has watermarked 180 billion images and videos to date, watermarked more than 240,000 years of audio content, and frequently processes 1 million verification requests per day. If these numbers hold, they mean SynthID already covers a fairly broad range of generated content and detection traffic.

But attribution must be made clear: all three numbers come from Google's own account, with no third-party verification. The source is a post on Google AI's X account, which is a vendor self-report rather than independent reporting. They should therefore be read as 'the scale Google claims,' not as externally confirmed facts.

The statistical basis is also missing. The cutoff time and calculation method for the 180 billion images and videos and 240,000 years of audio are not disclosed; whether the average of 1 million verification requests per day is a peak or a long-term average is unclear; and the statistical basis, time range, or geographic distribution of verification requests is not explained. These gaps do not prevent the numbers from being cited, but they directly affect how strong a conclusion they can support.

An easily overlooked limitation is that the portal can only detect files carrying a SynthID watermark. The source does not mention its ability to identify unwatermarked content. This means the detection result is directional—detecting a watermark can indicate that the file is related to the SynthID system; but a non-detection does not imply that the file is not AI-generated.

More critically, the accuracy rate, false positive rate, and tamper resistance of SynthID watermarks are not given in the source. For any tool claiming to 'verify AI generation,' these three are precisely the core indicators that determine its credibility. In their absence, the portal's practical value can only be described as 'providing a detection method,' not as 'a reliable criterion for judgment.'

In addition, whether the portal is completely free for ordinary users and whether there are usage restrictions are also not explained in the source. The 'public' degree of public access currently only extends to 'anyone can go there.'

From an industry structure perspective, the significance of this announcement lies in pushing content provenance from a feature inside individual products toward a cross-vendor public verification entry point. If files from OpenAI, NVIDIA, and Kakao can indeed be detected, and Apple joins later, then synthid.com could become a public reference point for AI content verification. This is an inference, based on the partner list Google named and the portal's cross-vendor positioning.

But the reasons to keep distance are equally sufficient. The sole source for the entire announcement is a post on Google AI's X account, with about 34,900 views, which is a vendor self-report. The scale numbers lack third-party verification, partner support status is partly unclear, watermark technical indicators are missing, and Apple's joining time and covered product lines are unknown. Putting these together, the reasonable judgment is: clear direction, limited evidentiary strength.

For readers, the actionable conclusion is layered: the portal's launch and detectable modalities, and the partner list, are confirmed facts; 180 billion, 240,000 years, and 1 million per day are Google's own claims; and judgments such as 'SynthID has become industry-level verification infrastructure' can currently only be considered inferences, and depend on technical indicators not yet disclosed and partners' actual coverage.