Most SEOs know embeddings help Google understand on-page language, but fewer realize it’s doing the same for visuals—and comparing them to your content and keywords.
Google has a free public tool that lets you upload two images and see how visually or semantically similar they are, using cosine similarity.
In the screenshot below, we compared two AI-generated images that share similar themes (green goblins, chaos, and cash)—but Google’s model still sees them as only scoring .34 similar in cosign similarity.
So what does this mean for SEOs?
• If your images don’t match the topic or reinforce your keyword intent, Google can detect that.
• If you’re using decorative, stocky, or off-topic visuals, there’s a chance they’re hurting more than helping.
• It’s another reason to think about image relevance, not just alt text or file names.
This adds a new layer to content optimization: Does your visual content actually support your topical authority?
Have you explored how Google handles image embeddings? Want a link to test the tool?
(And yes, the image below is a nod to that now-viral description of SEOs as goblins rolling in cash and fighting alligators.)
This post was originally shared by on Linkedin.