Missing HTML alt Text Is a Fact. Image Purpose Is a Separate Claim.

A design review exercise examined a synthetic HTML page that contained a single image with no alt text. Three AI agents, Claude Code, Codex and agy, reviewed the page and correctly identified the missing alt attribute and the horizontal overflow, but failed to distinguish between the observed fact and the unverified assumption about the image's purpose. The page's full width was 744 pixels, while the viewport remained at 375 pixels.

The issue lies in the reviewer's interpretation. The first step was to report an observed fact: the image had no alt attribute. The second step is to recognize that the purpose or role of the image was not established by the page's markup. Depending on the image's purpose, it could be informative, decorative, or part of a link or button. The W3C decision tree provides a guideline for determining the appropriate alternative text based on the image's role and context.

The review should separate the claims of the missing alt attribute and the unverified image purpose. The former is a real finding that needs fixing, while the latter depends on the author's intent and context, which were not provided by the page's markup. The reviewer should offer conditional fixes based on possible scenarios: if the image is informative, provide alt text that conveys its meaning; if it's decorative, use alt=""; if it's a link or button, describe its action or destination; and if it's complex, such as a chart, a longer explanation may be necessary.

This approach ensures that the review is honest about its limitations and provides clear guidance for the author to follow. Validation alone does not prove the image's purpose, and a single synthetic case does not prove the model's behavior on different pages. The review process should consider the page's purpose and context, as a missing alt attribute can have significant implications for accessibility.