What SynthID Is and Why It Matters
SynthID is Google DeepMind’s invisible watermarking system designed to help people detect AI-generated content without altering how it looks or sounds. Instead of placing a visible badge on an image, video, or audio clip, SynthID embeds a subtle, machine-readable pattern directly into the file. Google says it has already watermarked more than 100 billion images and videos, plus the equivalent of 60,000 years of audio. The goal is to make AI detection tools reliable even when traditional metadata is missing or stripped out by screenshots, edits, or re-uploads. SynthID now works alongside C2PA Content Credentials, a standardized form of metadata that records how content was created or edited. Together, they provide a two-layer system: metadata for rich context and the invisible watermark as a persistent signal, making provenance more resilient as generative AI and deepfakes become harder to spot with the naked eye.
How Chrome and Search Help You Detect AI-Generated Images
Google is bringing SynthID beyond the Gemini app and into everyday browsing through Chrome and Google Search. In Chrome, you can right-click on an image and choose an option that effectively asks, “Was this generated with AI?” The browser then checks for both SynthID watermarks and C2PA metadata. If it finds a signal, you’ll see a clear response and additional context about the content’s origin, including whether it came from a camera or a generative AI tool. On mobile, you can use Circle to Search: draw a circle around an image or element on the screen, then trigger the same AI detection process. These Chrome AI detection features apply to images, videos, and audio that carry SynthID or content credentials, turning everyday browsing into a quick way to verify suspicious content without needing a separate app or technical expertise.
OpenAI, ChatGPT, and the New Public Verification Tool
OpenAI is now working with Google to embed SynthID into images created through ChatGPT and the OpenAI API, on top of the Content Credentials metadata it has been adding since 2024. This means many images generated by OpenAI’s models will carry two overlapping signals: C2PA-conformant metadata and SynthID’s invisible watermark. To make this practical, OpenAI has launched a public verification site at openai.com/verify. There, you can upload an image and the system will check for both metadata and SynthID watermarks to see if it was generated using OpenAI’s tools. If nothing is detected, the tool won’t claim the image is definitely human-made, because signals can be removed or spoofed and not all AI models support SynthID. Still, this partnership extends SynthID watermark detection beyond Google’s own models and into the broader ecosystem of AI image generators.
Invisible Watermarks vs Metadata: How Detection Actually Works
AI detection tools rely heavily on two complementary technologies: metadata-based content credentials and invisible watermarks. C2PA Content Credentials are pieces of structured metadata attached to files, describing how they were created or edited—for example, whether an image came from a camera or from a generative model. They are easy to read when intact but can be lost through screenshots, compressions, or re-uploads. SynthID solves that by embedding a subtle, algorithmic pattern directly into the pixels or audio signal. This watermark is designed to withstand common transformations, so even a cropped or resized image can still be recognized as AI-generated. When you use Chrome AI detection, Search, Gemini, or OpenAI’s verification site, the tools first look for C2PA metadata, then fall back to watermark signatures if metadata is missing. The result isn’t perfect certainty, but it significantly improves transparency about whether content likely comes from an AI system.
What You Can Verify Today—and the Limits
Right now, you can detect AI-generated images, videos, and audio that carry either SynthID watermarks or C2PA metadata from participating companies. In practice, that means content created with Google’s Gemini tools and partners like OpenAI, Nvidia, Kakao, and ElevenLabs is increasingly likely to be identifiable. Using Chrome’s right-click menu, Circle to Search, Google Search, the Gemini app, or openai.com/verify, you can quickly check suspicious media and get a plain-language explanation of what was found. However, detection has clear limits. Not every AI model embeds SynthID or Content Credentials, and bad actors can still manipulate or spoof signals. If a tool reports no watermark or metadata, that doesn’t guarantee the content is human-made—it simply means no trustworthy signal was detected. Even so, these tools mark a major step toward a more transparent web, especially as deepfakes and AI-generated misinformation continue to spread.