Alt text coverage is lying to you: count decisions, not images

Alt coverage is usually computed as images with a non empty alt divided by all images. Watch what that formula pays for. A decorative flourish with alt text reading gold divider flourish gets a point. A product image with alt text copied from its filename, IMG 2043 final final, gets a point. A chart with alt text that says chart gets a point. The metric rises while the experience rots, and because it rises, nobody funds the fix.

WCAG 1.1.1 Non text Content never asked for coverage. It asks that the alternative serve the equivalent purpose. A decorative image has no purpose to serve, so its correct alternative is empty, implemented as alt with no value, which the naive coverage metric scores as a failure. Your dashboard is literally penalising correct work. That inversion is the whole argument for changing the measure.

How the coverage metric rewards the wrong work

Classify a sample of images by role before judging their text. Informative images convey content: product photos, editorial images, charts, diagrams. Decorative images are layout and mood: spacers, flourishes, background texture presented through image elements. Functional images do a job: a logo linking home, an icon button, a social link. Each category has a different correct answer, which is why one coverage number cannot describe them.

Sample where the templates live, not just at page level: product template, listing template, article template, home page sections, email capture and footer. Alt quality is a template behaviour in most builds. Twenty images per major template, classified and scored, tells you more than ten thousand images counted by a crawler, because it points at the handful of templates generating the problem.

The three categories, and how to sample them

The honest test for an informative image is substitution. Replace the image with its alt text and read the page. If nothing is lost, the alt passed. If the reading makes no sense, or a key fact, a colour, a size comparison, the data in a chart, exists only in the pixels, it failed. Product images fail this test constantly: alt text repeats the product name from the heading next to it, while the image itself showed the back view, the scale, or the wear that the shopper needed.

Scoring informative alt: the replacement test

Keep reading: Keyboard trap incidents: the accessibility metric that should always read zero, ARIA label coverage: when one hundred percent is the warning sign.

What to report instead, and how it changes the work

Charts need their data, not their genre. Alt text for a chart should give the trend and the key values, or point to an adjacent data table that does. Alt text that says sales chart 2025 is a label on a locked door. Score it as a fail and watch how fast chart producing teams discover tables, which help every reader, not only screen reader users.

Report three numbers: informative images with a passing alternative as a share of informative images sampled, decorative images correctly nulled as a share of decorative images, and functional images labelled by purpose as a share of functional images. Expect the decorative number to be the embarrassing one at first. Sites that chased coverage filled their decorative images with text, and emptying them is quick, template level work with an outsized experience payoff: screen reader users stop wading through described wallpaper.

Publish the sampling method beside the numbers, categories, sample sizes, the replacement test, so the figures survive scrutiny. A smaller, honest metric that a practitioner can re run beats a big automatic one nobody can defend. That, in the end, is what a metric is for.

Frequently asked questions

Why is 100 percent alt text coverage a bad target?

Because decorative images should have an empty alt attribute, which a coverage metric counts as missing. Chasing 100 percent coverage pushes teams to write descriptions for images that should be silent, adding noise for screen reader users. The correct target is right decisions by image category, not maximum filled attributes.

How can I tell if an image is decorative or informative?

Ask what is lost if the image is removed. If the page loses information, a product view, data, a step in a process, the image is informative and needs a real alternative. If only the decoration changes, it is decorative and should carry an empty alt so assistive technology skips it.

What should the alt text for a logo or image button say?

Its purpose or destination, not its appearance. A logo linking home should be labelled with the company name and, where it is not obvious, its function as the home link. An icon button should say what it does, search, close, add to basket, never what it depicts, magnifying glass, cross, bag.

How big a sample do I need for an alt text audit?

Enough per template to see the template behaviour: around twenty images per major template type is a practical start, classified into informative, decorative, and functional. Alt problems are usually generated by templates, so sampling by template finds the fixable causes faster than a giant undifferentiated crawl.