Alt Text Coverage Rate: A High Number That Can Still Mean Failure
Alt text coverage, the share of images carrying an alt attribute, is the easiest accessibility metric to compute and one of the easiest to lie with. A site can report 98 percent coverage while a screen reader user hears image, image, IMG underscore 4021, and the word spacer read out on a product page. Coverage counts the presence of an attribute. WCAG 2.1 Success Criterion 1.1.1, Non-text Content, requires something coverage cannot see: that the text alternative serves the equivalent purpose, and that purely decorative images are hidden from assistive technology with an empty alt, which a naive coverage metric counts as a miss and pressures editors to fill in.
That pressure is the pitfall in one sentence. A team chasing a coverage number will write alt text for decorative images, keyword stuff product alts for search, and accept the file name the CMS filled in automatically, and every one of those moves raises the metric while making the experience worse.
Measure three numbers instead
Split coverage into the numbers that each mean something. Meaningful image rate: of the images that convey information, what share has alt text that describes the information, judged by a human sampling real pages. Decorative handling rate: of the images that are purely decorative, what share is correctly nulled with an empty alt. And junk alt rate: the share whose alt is a file name, a placeholder, the word image, or a keyword string. The third number is the one that tells you whether the first number is worth anything, and in most first audits it is embarrassingly high, because auto generated alt attributes from uploads are junk by default.
Sampling beats crawling here. A crawler can count attributes across ten thousand pages in an afternoon and tell you nothing about quality. A reviewer opening thirty product pages and reading every alt aloud learns more, because alt text is writing, and writing is judged by reading it. Report the sample size with the rate. A meaningful image rate of 90 percent on a sample of 200 images is a claim. The same figure from an attribute count is a guess wearing a lab coat.
Where junk alt comes from
The sources are boring and fixable. The CMS inserts the file name when an editor skips the alt field, so fix the default. The SEO checklist demands keywords in image alts, so correct the checklist: alt text describes the image for someone who cannot see it, and a keyword string fails that reader and the criterion together. The template renders a decorative flourish as a content image, so fix the template once rather than asking editors to null it forever. Each source fixed upstream removes junk from every future page, while editor training alone fights the last upload forever.
Icons and functional images need their own rule in the same audit. A magnifying glass icon inside a search button does not need alt describing a magnifying glass; the button's label, search, is the alternative that serves the purpose. Counting that icon as missing alt text is another way the naive metric manufactures failure, and teams waste sprints describing icons while product photography, the images that sell, still carries file names.
Report it so it stays honest
Publish the three rates together, with the sample size and date, and refuse to publish the single coverage figure on its own. The moment leadership sees one number, that number becomes the target, and the target will be hit by whatever behaviour it rewards. Three numbers with a method note is a slightly harder slide and a much better programme, because it keeps the goal where the criterion puts it: a reader who cannot see the images getting the same information as everyone else.
Related reading: Target Size Pass Rate: Catching Clickable Boxes Too Small to Use, Contrast Failure Rate by Template, Not by Page, Focus Visible Pass Rate: Measuring the Outline Shoppers Actually See.
Frequently asked questions
What is a good alt text coverage rate?
A single coverage figure is the wrong target, because it counts junk alt and penalises correctly nulled decorative images. Track meaningful alt rate, decorative handling rate, and junk alt rate separately, from a human reviewed sample.
Should decorative images have alt text?
They should have an empty alt attribute, alt equals quote quote, so assistive technology skips them. Writing descriptive alt for decoration adds noise for screen reader users and can make a naive coverage metric look better while the experience gets worse.
What counts as junk alt text?
File names, placeholders, the word image, repeated boilerplate, and keyword strings written for search rather than for a reader who cannot see the image. Junk alt satisfies an attribute count and fails WCAG 1.1.1.
Why sample pages manually instead of crawling the whole site?
A crawl counts attributes, not quality. Alt text is writing, and its quality shows when a reviewer reads it on real pages. Report the rate with its sample size so the claim stays honest.