A photograph online can now depict an event that never happened. Artificial images may show convincing crowds, damaged buildings, or invented products. Google is expanding its tools for checking where such content originated. Its SynthID Detector became available worldwide in English during October 2026. The change gives ordinary users access to technology previously tested with media professionals.
Google’s New Access Goes Beyond Professional Newsrooms
Google first introduced SynthID Detector to selected testers in 2025. The company opened the verification portal to everyone in October 2026. Its October 7 announcement was updated on October 29.
Users can upload supported image, audio, or video files for checking. The system searches for invisible SynthID watermarks placed during AI generation. Google says its partners include OpenAI, NVIDIA, and Kakao. This is different from guessing authorship based only on visual appearance. The detector checks for an identifiable signal embedded by participating systems.
Google’s Watermark Numbers Show the Scale
Google reports watermarking over 180 billion images and videos since introducing SynthID in 2023. It also reports watermarking 240,000 years of generated audio content. These are company figures covering supported media, not all AI content online.
Google says verification features across Search, Gemini, and Chrome handle over one million requests daily. The demand reflects an important problem for anyone sharing pictures online. People increasingly need evidence about how digital media was produced.
How Invisible Watermarks Survive Everyday Editing
Traditional watermarks place visible marks across a photograph or its corner. Those marks can disappear after cropping or simple image editing. SynthID instead embeds an invisible pattern into generated image content.
Google designed the pattern to survive common changes to images. These include cropping, color adjustments, filters, and certain compression methods. The watermark does not visibly change the picture during generation.
However, the system cannot promise detection after every possible alteration. Extreme manipulation may damage the embedded signal or prevent identification. A missing watermark therefore cannot prove that an image is genuine.
Checking an Image Takes More Than One Click
Anyone using an AI image detector should understand what it actually checks. SynthID Detector examines content from participating watermark systems, not every image generator. Other detection services might estimate artificial origins by analyzing visual patterns.
You can follow a straightforward verification process before reposting suspicious media:
- Save the original image when possible instead of a compressed screenshot.
- Upload the file to SynthID Detector and review its reported result.
- Check whether the image contains Content Credentials showing creation history.
- Search for earlier appearances of the picture using reverse image search.
- Contact the original publisher before treating an uncertain claim as fact.
Google’s Gemini app also provides built-in checks for supported content. Its help documentation explains that Gemini’s SynthID verification currently recognizes Google’s own generated media. The separate Detector portal supports participating partners more broadly.
New Research Shows Why Scores Require Caution
An updated October 2026 research preprint evaluated 16 detection methods. Researchers tested roughly 2.6 million images from 291 generation systems. Across the benchmark, the leading method averaged 75% accuracy. The weakest averaged 37.5% under those evaluation conditions.
These figures concern pattern-based detectors, not a direct evaluation of SynthID. They show why one score should not settle a dispute. Results changed considerably across different datasets and image generators.
Consider a convincing photo that circulates during an emergency. A detector might flag it, yet the accompanying caption could still misidentify its location. Alternatively, an authentic photo may be attached to an invented story. Verifying an image’s origin and checking its factual context require separate work.
For example, a fabricated photo of a flooded street might spread through several social networks. Even with a watermark result, readers still need the location and original source. A website republishing it should document those checks before claiming the image records an actual disaster.
Content Credentials Offer Another Type of Evidence
Content Credentials use signed metadata to document a file’s origin and editing history. Supported software can record whether artificial generation or editing was involved. Unlike a visual prediction, this information comes from participating creation tools.
Still, files can lose metadata during ordinary sharing and editing. The C2PA standards organization also warns that provenance alone cannot establish factual truth. A genuine camera photo might still show an event out of context.
What This Means for Online Creators and Readers
Bloggers should verify illustrations before presenting them as documentary photographs. Online stores should check generated product pictures against their actual inventory. News publishers need original sources when images support significant allegations. Google’s wider launch gives users another useful verification route. Yet responsible checking still needs source research and human judgment. The important question is not simply if AI was involved. Readers also need to know if the accompanying claim is true.
