# Alt Text Coverage vs Alt Text Quality: Two Metrics, Two Different Lies

Source: https://complymetric.com/blog/2026-11-03-alt-text-coverage-vs-quality-metric.html

Alt text coverage is the most reported accessibility metric in the world and one of the least informative. A site can sit at 100 percent coverage while a screen reader announces IMG underscore 2048 final final dot jpg on every product image, because coverage counts the presence of an attribute. It says nothing about whether the attribute does the job.

The job is defined by WCAG 1.1.1, Non text Content: the text alternative should serve the equivalent purpose. Equivalent purpose is a quality judgement, so quality needs its own metric, sampled by humans, reported next to coverage, and never averaged into it.

## The coverage number and what it hides

Coverage still matters: a missing alt attribute is a definite failure, and decorative images need their empty alt, which is a deliberate value, not an omission. Report coverage as two numbers, images with any alt attribute, and images with a non empty alt where one is required. The gap between them is usually where the decorative images were never decided about, which is its own finding.

Then stop. Coverage cannot detect filename alt text, keyword stuffed alt text written for search engines, alt text that duplicates the caption word for word, or the classic, image of, followed by a description of the photograph rather than its purpose. Every one of those passes a coverage audit and fails a user.

## Sampling quality without boiling the ocean

Quality is measured by sample: pull a random set of content images per template family, fifty is usually plenty, and score each alt against a short rubric. Does it convey the image's purpose in context, is it free of filename and stuffing, is length proportionate, does it avoid repeating adjacent text. Score pass or fail per image and report the rate per template, the same discipline we applied to error identification in [Error Identification Rate: Do Your Forms Tell People What Went Wrong](https://complymetric.com/blog/2026-11-02-error-identification-rate-forms-metric.md).

Expect the failures to cluster by source, not by page. Images from one feed, one migration, one agency upload will share a failure style, because alt text is written by processes more than by people. Finding that a supplier feed ships filenames as alt text is a one line fix upstream that repairs ten thousand images, which no amount of page level editing would ever catch up with. The target size sampling in [Target Size Failure Rate: Measuring the Buttons People Cannot Tap](https://complymetric.com/blog/2026-11-02-target-size-failure-rate-metric.md) found the same upstream pattern in component libraries.

## What good looks like, by image type

Product images: the alt should name the product and the distinguishing feature the image shows, colour, angle, detail, matching what a sighted shopper learns from the photo. Informational graphics: the data or the conclusion, in text, or a link to it, because chart as alt text is a surrender. Functional images, icons and linked logos: the function or destination, never the picture. A magnifying glass icon's alt is Search, not magnifying glass.

Publish both numbers together, always: coverage, and sampled quality, with the sample size stated. A program reporting only coverage is reporting the easy half, and everyone in the field knows it, including the auditors who will eventually sample the quality themselves, in a context with deadlines and penalties attached.

## Frequently asked questions

### What is a good alt text coverage target?

One hundred percent of images should carry a deliberate alt attribute: meaningful text for content images, an empty alt for decorative ones. But coverage alone is a weak target, since filename alt text passes it. Pair it with a sampled quality rate and report both numbers together.

### How do you measure alt text quality?

By human sampling. Take a random sample per template family, around fifty images, and score each alt text against a rubric: conveys purpose in context, no filenames or keyword stuffing, proportionate length, no duplication of adjacent text. Report the pass rate with the sample size stated.

### Should decorative images have alt text?

They should have an empty alt attribute, alt equals empty quotes, which tells assistive technology to skip them. That is a deliberate value and counts as correct coverage. A missing attribute is different: some screen readers will announce the filename instead, which is the failure the empty alt prevents.

### What is the most common alt text quality failure?

Filename or placeholder text shipped by a feed or migration, followed by alt text written for search engines rather than listeners. Both cluster by image source, which is why quality sampling groups results by template and source: the fix is usually one upstream process, not thousands of page edits.