Google opens site that can detect AI fakes, but it has clear limitations

Google has launched a tool meant to make it easier to tell whether content was made by artificial intelligence. It is called SynthID Detector and was first shown at the Google I/O developer conference in 2025.

You upload an image, an audio file, a video or a text to the tool. It then scans the file for a hidden mark and shows whether the mark was found. If it was, the tool highlights the parts of the content where the mark is likely located.

An invisible watermark

The mark is called SynthID. It is a watermark that cannot be seen with the naked eye. Google embeds it when Google's own AI models create content. That includes Gemini, Imagen, Lyria and Veo.

The mark sits in the image's pixels themselves. It therefore often survives even if information in the file (metadata) is removed or the image is altered. The tool can also detect that only parts of an image were created or edited with AI.

Here are the limitations

The tool only finds content marked with SynthID. That is primarily content from Google's own AI tools.

Images and videos from other services, such as ChatGPT, cannot be detected if they do not use SynthID. Microsoft, Meta and OpenAI have their own methods for marking AI content, and these cannot be read by Google's tool.

That means a result with no findings is not the same as proof that the content is genuine. It could just as well have been made with an AI tool that does not use Google's mark.

What does it mean for you?

SynthID Detector is a useful extra tool if you are in doubt about an image or a video you have seen. If the tool finds the mark, it is a strong sign that Google's AI created the content.

But you should not trust it blindly. Also use common sense: check who shared the content, whether it has been covered by credible media and whether it matches other sources. Until more companies use common marking, no single detector will be able to catch all AI fakes.

What it means for you

If you have seen an image, a video or an audio file that seems suspicious, you can upload it to SynthID Detector and see whether Google's hidden mark is in it. If the tool finds the mark, it is a strong sign that Google's AI created the content. If it finds nothing, you cannot assume the content is genuine, because it may still come from other AI services. So also check who shared it and whether credible media are writing about it.