GitHub describes an alt-text plugin for its Accessibility Scanner that checks more than whether an attribute exists. Deterministic rules flag problems such as filenames, placeholders, generic labels and duplicated descriptions.
An optional model-assisted rule considers the image and surrounding context. The article also explains why decorative images with intentionally empty alt text should not be treated as missing descriptions.
Binary perspective: accessibility is part of content design. Describe the information an image contributes to the task, and combine automated checks with human review of the reading experience.
The following is original Binary Solutions editorial analysis. It explores the engineering implications of this topic; it does not reproduce the source article or claim independent verification of its results.
Begin with the purpose of the image
Alternative text should convey what the image contributes in its particular context. The same picture can require different treatment on a marketing page, in an instructional guide or in a technical report. Start by asking what the reader would lose if the image were absent.
A decorative shape may add atmosphere without carrying information. A screenshot may show where to perform an action. A chart may support a conclusion that is not stated elsewhere. These roles call for different solutions, and none is improved by mechanically describing every visible colour and object.
Read the surrounding text before writing a description. Avoid repeating a heading when the image supplies additional information, but avoid making users hear the same sentence twice when the image adds none. The goal is equivalent understanding of the page's message. A short, purposeful description is often more useful than an exhaustive inventory of visual details that leaves the reader to infer why they matter.
Give complex visuals more than a short label
A dense chart, workflow or interface diagram may contain more information than a brief alternative description can reasonably carry. In that case, use the short description to identify the visual and provide its main purpose, then place a fuller explanation or accessible data nearby.
For a chart, state the relevant trend in prose and provide a table when exact values matter. For a workflow, describe the sequence, decision points and outcomes. For a screenshot used in a tutorial, write the operational instructions as text rather than requiring readers to locate an unlabeled region in the image.
This approach also benefits people viewing a small screen or a low-resolution image. Important content remains usable when the visual is difficult to inspect. Keep the extended explanation connected to the figure, with a descriptive caption or link. Accessibility is strongest when the information architecture carries the meaning and the visual reinforces it, rather than serving as the only place where the meaning exists.
Combine automated checks with contextual review
Automated checks can detect a missing attribute or certain structural errors. They cannot reliably determine whether a description communicates the purpose of an image. Passing a mechanical check is therefore a starting point for review, not a complete assessment of accessibility.
Review representative pages using the information available to someone who cannot see the image. Can they understand the claim, complete the task and distinguish important controls? Check the order in which descriptions and surrounding text are encountered. An otherwise good description can become confusing when separated from the content it explains.
Include editorial judgment in acceptance criteria. A reviewer should be able to identify why a visual is informative, decorative or redundant. Where image descriptions are generated automatically, treat them as drafts requiring contextual review. A model may recognise objects accurately while missing the reason the image was selected for this particular page. That reason is what the author needs to communicate.
Make accessible publishing a repeatable workflow
The content system should make good descriptions easy to provide and maintain. Include fields for alternative text, captions and source credits, with guidance that explains their different purposes. Avoid automatically using the filename as a description; it rarely tells the reader anything useful.
Review descriptions when images are replaced or when the surrounding article changes. Reusing an image does not guarantee that its old description remains appropriate. Keep the relationship between the asset and its current context visible to editors.
For a site that recolours images, verify contrast and legibility after treatment. Do not rely on colour alone to distinguish chart series or interface states. Preserve labels, shapes and other cues where they carry meaning. Finally, include a small set of real publishing examples in the team's guidance. Concrete examples of useful descriptions make a quality standard easier to apply than a rule that merely says every image must have text.
Follow the evidence
Use the original publication for the author’s full argument, methodology and updates. This perspective is a starting point for a conversation about your own context.
Open original publication ↗← Return to the archive