AI images have gotten good enough that the uncanny valley is officially behind us. A face that would have screamed "fake" two years ago now passes casual inspection. But they're not perfect yet—and there are ways to check.
The most straightforward approach is to let AI itself do the detective work. Google's Gemini embeds something called a SynthID watermark into every image it generates. Upload a suspect image to Gemini and ask if it was made by AI—if the watermark is there, Gemini will find it. It's not foolproof (watermarks can be removed with effort), but it catches a lot of cases and even identifies which model created the image.
There's also a broader standard called C2PA (Coalition for Content Provenance and Authenticity) that labels AI-generated content. Websites like Content Credentials let you upload an image and check for these markers. Neither tool is a guarantee—a clean result doesn't prove an image is real—but they'll flag many AI generations.
We're a new kind of news feed.
Regular news is designed to drain you. We're a non-profit built to restore you. Every story we publish is scored for impact, progress, and hope.
Start Your News DetoxThe Context Check
Images don't exist alone. A photograph from a reputable news outlet will usually be honestly labeled if it's AI-generated. Social media is messier. Content floats around without attribution or context, which is exactly when fakes thrive.
Look for consistency: Do multiple photos of the same scene show the same details from different angles? For illustrations, can you find the artist's name and portfolio? A reverse image search can reveal where a photo originally came from. If nothing matches, that's a red flag.
What to Actually Look For
AI doesn't take photographs or draw by hand. It generates approximations based on patterns in its training data. That's where the tells appear.
Generic sheen is common—anime characters, trees, city streets that feel like they could be anywhere. There's even a recognizable ChatGPT font that appears when the AI generates text without specific styling instructions.
Physics breaks down in interesting ways. Turrets sprout in illogical places. Staircases lead nowhere. Elevator doors open onto walls. Faces and hands look compressed, with blurred or fuzzy details that don't quite resolve. Fingers especially—count them. AI still struggles with getting the right number.
With a bit of practice, these patterns become visible. You'll start noticing the generic quality, the physics violations, the hands that are almost but not quite right. It's not a perfect system, but it's enough to make an educated guess in most cases.
