Your alt text passes automated checks. That doesn’t mean it’s any good.

Summary

Automated accessibility checkers reliably flag missing alt text attributes but frequently miss unhelpful descriptions like raw filenames or repetitive labels. To evaluate image description quality without generating excessive false positives, GitHub built an alt text plugin for the GitHub Accessibility Scanner. The tool combines five deterministic, zero-credential rules that evaluate strings and visual layout spacing with an opt-in vision model check for subjective context. Page context including headings and surrounding prose is extracted alongside images to guide the model using structured outputs and explicit anti-nitpick instructions. While deterministic checks catch unwritten text, the model-driven rule serves as an opt-in prompt for human review rather than an absolute verdict.

Takeaways

1. 1

   Repeated alt text detection relies on rendered layout bounding-box gaps rather than document markup order to prevent false positives across distant page elements.

2. 2

   The vision model check uses structured outputs, explicit anti-nitpick rules, and surrounding context like headings and links to prevent the model from critiquing acceptable text.

3. 3

   Model-based image evaluation requires URL query and fragment redaction and functions as an opt-in scheduled check to manage privacy risks and inference costs.