Alt Text Passing Automated Checks Does Not Mean Quality

According to WebAIM's 2026 report, more than a quarter of images on the top 1 million homepages have missing or ambiguous Alt text, revealing accessibility issues. Among images with Alt text, 10.8% contain non-descriptive content such as 'image' or filenames.

Automated tools accurately determine the presence of Alt text but have limitations in evaluating content quality. This is because quality standards are subjective, and false positives can lead development teams to disable the tools. Therefore, the GitHub Accessibility Scanner plugin combines deterministic rules and AI models to address this issue.

By default, it applies five deterministic rules to detect objectively incorrect Alt text, such as filenames, placeholders, and duplicate text. At this stage, it uses Playwright's role-based locator to exclude decorative images (empty Alt) that are omitted from the accessibility tree from inspection.

Repeated Alt text checks are based on screen layout rather than DOM order. Images are considered part of the same group only when the gap between their bounding boxes is within a certain multiple, preventing the misinterpretation of identical logos that are far apart on the screen.

Ambiguous text that these deterministic rules fail to capture is designed to be judged by AI models analyzing image content and context, providing reliable accessibility checks.