Accessibility first pass
Combine automated accessibility rule checks with a model's review of screenshots and markup to catch problems that need judgement — unclear labels, confusing reading order, meaning carried by colour alone.
How it works
Rule-based accessibility checkers are established and reliable for what they cover — missing alt attributes, contrast ratios, missing form labels — but they can't tell whether an alt text is meaningful or a label makes sense. The model adds a judgement layer: given screenshots, the rendered markup and the accessibility tree, it reviews whether link and button text makes sense out of context, whether headings describe their sections, whether reading order matches visual order, whether errors are explained, and whether instructions rely on colour, position or shape. The result is a prioritised list for a human tester to verify with real assistive technology.
When to use it
As a fast early check on new screens and components; when an accessibility audit is coming and the obvious problems should be fixed first; on content-heavy sites where text quality matters as much as markup; and in design review, before anything is built.
Limitations
Neither rules nor model can tell you what it is actually like to use the product with a screen reader, keyboard, switch device or magnification; that takes testing with the tools and, ideally, with disabled users. The model's judgements about clarity are inconsistent between runs. Dynamic behaviour — focus management, live updates, timeouts — barely shows in a screenshot. A clean first pass is easily mistaken for conformance.
Watch for
The model approves alt text that reads well but doesn’t carry the content, such as “bar chart of sales” for a chart whose figures appear nowhere else. Check that each meaningful image’s alt text or nearby text gives the reader the same information the image does.
What you still own
Testing with real assistive technology, deciding what conformance level you're checking against, and any claim of accessibility.
Cite this
Qlarify Labs. (2026). Accessibility first pass. Retrieved from https://labs.qlarify.fi/ai-in-testing/accessibility-first-pass