How does image SEO actually work? The four mechanisms

Image SEO works through four separate, individually documented mechanisms: how machines actually read an image, how image weight affects page speed, where the law now requires image descriptions, and how rare readable image metadata still is. None of them is a growth-hack. Each is a plain, checkable fact with a primary source, and this post walks all four. If you're still deciding whether any of it is worth doing at all, that's the pillar's job, not this one; this post is the "how it works."

How does Google actually read images?

Through the text layer, mostly. Google's own image documentation is explicit: it "uses alt text along with computer vision algorithms and the contents of the page to understand the subject matter of the image." Vision reads the pixels; the words tell it what the pixels mean.

The filename is part of that layer too. Google says the filename "can give Google very light clues about the subject matter of the image," which is why a descriptive name beats a camera dump. The documentation's own example prefers something like my-new-black-kitten.jpg over IMG00023.JPG.

Structured data is the third piece. Add an ImageObject block and, in Google's image best practices, it "can display your images in certain rich results, including a prominent badge in Google Images." That is a machine-readable label you write, not a guess the crawler has to make.

None of this means Google is blind to your pictures — modern vision genuinely recognizes objects. It just can't infer meaning or relevance from pixels alone, which is the fuller argument I make in the pillar's section on whether Google can already see your images.

Do slow images really cost rankings?

Indirectly, yes, through page speed. Oversized images are one of the most common causes of a slow Largest Contentful Paint, and page speed feeds Core Web Vitals. Google states plainly that "Core Web Vitals are used by our ranking systems."

Now the boundary, from the same page-experience documentation, because it's the part most guides quietly drop. Good scores are not a ranking guarantee. Google writes that "getting good results in reports like Search Console's Core Web Vitals report or third-party tools doesn't guarantee that your pages will rank at the top of Google Search results," and that Search "always seeks to show the most relevant content, even if the page experience is sub-par." Speed is a factor, not a switch you flip for rankings.

The mechanical fix is compression: convert to a modern format like WebP and the file gets lighter, which is the half of optimization you can feel immediately. I documented what that looks like on one real image in the before/after post's file-size section. How big the win is depends entirely on how heavy the original was, so I won't turn one image's result into a general number.

Is alt text really required by law in the EU?

In the EU, yes, and the deadline has already passed. The European Accessibility Act's requirements have applied since 28 June 2025, per the directive's own text: "Member States shall apply those measures from 28 June 2025."

What the Act covers beyond that date — which businesses, which products, which services — is more nuanced than one sentence, and I won't characterize that scope from memory. The full scope breakdown is its own post, coming.

The mechanics underneath the law are older than the law. A screen reader can't show a blind user your hero image; it reads out whatever sits in the alt attribute, or nothing. That same attribute is what a search crawler indexes and what an AI answer engine parses. Write it once and three readers are served at once: the person using assistive tech, the crawler, and the AI. You rarely get a single fix that pays off in three directions.

How big is the metadata gap?

Bigger than you'd guess. In WebAIM's 2026 analysis of the top one million home pages, missing alternative text was detected on 53.1% of them. More than half of the most-visited sites on the web ship images a machine can't read.

That figure barely moves between WebAIM's annual surveys, so this isn't a blip; it's the web's steady state. Even a top-tier site fits the pattern, which is why I crawled anthropic.com and recorded the exact numbers in the pillar's data section.

Whether that gap reads as depressing or promising depends on which side of the audit you're standing on. If nearly everyone's images are unreadable, being the site whose images aren't is a cheap, uncontested edge.

The honest limit of all four

None of these four mechanisms is a growth guarantee. They make your images machine-readable. They don't make them ranked, and they can't make an AI cite you. Those outcomes aren't something any tool can measure today, which is the entire argument of the pillar's section on what nobody can honestly promise.

The mechanisms are the part you can verify on your own site: alt text present or missing, schema valid or not, files light or heavy. That's the work. The rest is marketing.