Alt Text Coverage Lies When Decorative Images Count as Covered

Alt text coverage is the first metric in most accessibility dashboards and one of the easiest to make meaningless. Count images, count those with an alt attribute, divide. A site can score near a hundred while its product images say image123.jpg and its decorative swirls carry paragraphs of keyword stuffed prose that screen reader users must endure.

The attribute being present is not the outcome. The outcome is that a user who cannot see the image gets the information the image carries, and is not forced to listen to the information it does not. That takes two metrics, not one, because the two failures point in opposite directions.

Split the population before you measure

Classify images into informative and decorative, template by template. Product photography, diagrams, charts, colour swatches: informative. Flourishes, spacers, background texture, the lifestyle banner whose content the adjacent headline already states: decorative. The classification is a judgement call, so make it once, document it, and keep it stable or your trend line is fiction.

Then measure each population against its own standard. Informative images need alt text that conveys the content and function, not the file name and not the word image. Decorative images need alt empty, alt equals quote quote, so assistive technology skips them in silence. Coverage becomes two honest numbers instead of one soothing one.

Judge quality with a sample, not a vibe

Pull a monthly sample of informative images per template and score the alt text against a short rubric: does it convey what the image shows that the page does not already say, is it concise, does it avoid starting with image of, does it name colour or variant when the image is the only place that information lives. Report the pass rate next to the coverage rate.

The swatch case deserves its own line in every ecommerce audit. When colour options are images, their alt text is the product information. A screen reader user choosing between unlabelled swatches is not having a slightly worse experience. They are being denied the variant list sighted shoppers get free.

Decorative failures are user harm too

Alt text on decorative images is not a harmless surplus. Every paragraph of flourish description is time taken from a screen reader user and trust spent. Audits that only hunt for missing alt teach teams to fill everything, which manufactures this second failure at scale.

Your metric should therefore count decorative images with non empty alt as defects, named as such in the report. Teams respond to what the dashboard calls a defect. Call it coverage and they will keep producing it. Call it noise imposed on users and the templates get fixed.

Tie the metric to templates and releases

Alt behaviour is set by templates and upload workflows, not by individual editors heroics. Product templates that pull the file name into alt by default will fail forever, one upload at a time. Fix the default, add an alt field with guidance at upload, and let the metric verify the system change rather than chase individual images.

Retest on the same classification each cycle and publish both numbers with the sample pass rate. Three numbers, honestly defined, will tell a steering group more than a single percentage ever did, including the part nobody wants to say out loud: which templates the programme has not actually reached yet.

Related reading

These earlier guides go deeper on connected questions: Form Label Coverage: A Metric That Counts Labels and Misses Meaning, Contrast Failure Rate: The Metric Lies When the Denominator Moves, and Keyboard Trap Count: The Metric That Should Always Read Zero.

Frequently asked questions

What should alt text coverage actually measure?

Two things separately: the share of informative images carrying useful alt text that conveys the image's content or function, and the share of decorative images correctly given null alt so they are skipped. One blended percentage hides both failures, empty description where it matters and noise where it does not.

Is empty alt text ever correct?

Yes, for decorative images it is required. alt with empty quotes tells assistive technology the image carries no information, so it is skipped entirely. Empty alt on an informative image, by contrast, deletes information the user needs. The same markup is correct or a failure depending entirely on the image's job.

How long should alt text be?

Long enough to convey the information, short enough to respect the listener. Most product images are served by a concise phrase naming the item, variant and relevant detail. If an image carries complex data, a chart for instance, the alt should summarise the takeaway and the page should provide the full data in text nearby.

Why do scanners report high coverage on sites with bad alt text?

Because most automated checks test presence, not usefulness. An alt attribute containing a file name, a keyword string or the word image passes a presence check. Quality requires sampling and human judgement against a rubric. Presence metrics are a starting screen, never the programme's headline number.