How to tell if images on a website are AI-generated

AI-generated images on websites are now everywhere: hero images, "team" photos, product shots, blog illustrations. Some sites say so; most don't. Unlike a builder's script in the page source, an image rarely carries clear proof of where it came from. Here is what can give you a real answer, what can only give hints, and why that distinction matters.

1. Content Credentials (C2PA): real proof, when present

Content Credentials are a standard, called C2PA, for attaching a signed record of how an image was made. Several AI image tools and cameras add them. If an image still carries them, a verifier shows which tool created or edited it.

• Save the image, then upload it at contentcredentials.org/verify.
• Or install a browser extension that reads Content Credentials on the pages you visit; Adobe offers one.

The catch: the credentials are easily lost. Screenshots, many image editors and some websites' image processing strip them. A missing credential tells you nothing.

2. SynthID: Google's watermark

Images made with Google's AI tools carry SynthID, an invisible watermark designed to survive cropping and compression. Google offers ways to check for it, such as asking its Gemini app whether an image was made with Google AI. It only works for content from tools that use SynthID, so a negative result says nothing about images from other generators.

3. Reverse image search

Right-click the image and search with Google Lens, or upload it to another reverse image search. Useful outcomes:

• It appears on a stock photo site or in an earlier news story: you've found its source.
• It appears on many unrelated "team" pages under different names: a stock or generated face, not the company's staff.
• It appears nowhere else at all, on a site that claims years of history: worth extra caution.

4. Visual tells (hints only)

Older and cheaper generators leave recognisable mistakes. Newer ones have fixed most of them, so treat these as hints, not proof:

• Text in the image that is garbled, misspelled or nonsense, on signs, labels or screens
• Hands, ears, glasses and jewellery that don't quite resolve
• Backgrounds that melt into patterns, or repeated objects with tiny differences
• Lighting and shadows that disagree with each other
• An airbrushed, too-perfect sheen on skin and products

Why not just use an AI image detector?

Tools that guess whether an image is AI-generated from the pixels alone are unreliable, and they fail badly against images that have been edited, resized or compressed. Even search engines that filter AI images rely mostly on reputation and blocklists rather than pixel analysis, because no automatic method is precise enough. Use detectors as one opinion at most. The engines that do have a filter are covered in how to hide AI-generated images in search results.

Look at the site, not just the picture

When images are ambiguous, the website around them often isn't. A site built with an AI app builder, written in stock AI copy and illustrated with generated pictures tells a consistent story. The build is the part with hard evidence: hosting, scripts and template leftovers, as explained in how to tell if a website was made with AI.

Kitsuvo, our no-AI web browser, focuses on that evidence. It doesn't judge individual images; it reads how each page was built, on your own computer, and filters likely AI-built results out of your searches.

When it matters

A generated illustration on a blog post is usually harmless. It matters when an image is evidence: a product photo for something you're buying, a "team" or "office" photo meant to show a business is real, or a news image. In those cases, run the reverse search at least, and see how to spot a fake online store.