Automated accessibility scans flag approximately 25% of WCAG issues on a given page. The remaining 75% requires human evaluation. Scans excel at detecting machine-verifiable conditions in code, but most accessibility issues depend on context, meaning, and user experience, which code analysis cannot determine. Categories of missed issues include meaningful alternative text quality, link purpose clarity, heading structure logic, keyboard interaction flow, focus order, form instructions, error message usefulness, and content accuracy across assistive technology output.
| Key Point | What It Means |
|---|---|
| Coverage Limit | Scans identify approximately 25% of WCAG issues on a page or screen. |
| What Scans Detect | Machine-verifiable conditions in HTML, CSS, and ARIA, such as missing attributes or invalid markup. |
| What Scans Miss | Anything requiring human judgment about meaning, quality, context, or user experience. |
| Coverage Method | A human-led evaluation conducted by an accessibility professional addresses the remaining 75%. |
Why Scans Have a Coverage Limit
Scans evaluate HTML, CSS, and ARIA attributes against rule sets derived from WCAG success criteria. A rule can confirm that an image has an alt attribute, but it cannot judge whether the alt text accurately describes the image. A rule can confirm that a heading element exists, but it cannot determine whether the heading reflects the content beneath it.
Most accessibility criteria depend on meaning, not markup. That is the boundary scans cannot cross.
Categories of Issues Scans Cannot Identify
The 75% that scans miss falls into recognizable categories. Each requires a person evaluating the page with assistive technology and judgment.
- Alternative text quality: Scans confirm alt text exists but cannot judge whether it conveys the image’s purpose or meaning.
- Link purpose: Scans see link text but cannot evaluate whether “read more” or “learn here” tells a screen reader user where the link goes.
- Heading structure logic: Scans verify heading levels exist in order but cannot assess whether the headings reflect the actual document outline.
- Keyboard interaction: Scans cannot operate a page with only a keyboard to confirm every interactive element is reachable and usable.
- Focus order and visible focus: Scans cannot watch focus move through a page and verify the order matches the visual layout.
- Form instructions and error messages: Scans cannot judge whether instructions are clear or whether error messages explain how to correct the input.
- Custom widgets: Scans cannot operate accordions, modals, carousels, and menus to confirm the announced state matches the visual state.
- Reading order: Scans cannot determine whether the order content is announced by a screen reader matches the intended reading sequence.
What Scans Do Well
Within their 25% coverage, scans are fast and consistent. They catch missing alt attributes, empty buttons, missing form labels, invalid ARIA usage, and language attribute issues across thousands of pages in minutes. Scheduled scans through a monitoring setup catch regressions when new code ships.
The value of scans is volume and speed at the surface level. The limitation is depth.
How the Two Coverage Layers Work Together
A complete accessibility evaluation pairs automated scans with a human-led evaluation. The scan covers the machine-verifiable layer across the full property. The evaluation covers the contextual layer on a representative sample of pages or screens, applying screen reader testing, keyboard testing, visual inspection, and code inspection.
Scans alone produce a partial picture. A human-led evaluation alone misses the speed advantage scans provide for ongoing monitoring. Used together, the two layers cover the full WCAG conformance requirement.
What This Means for Conformance Claims
A clean scan report does not equal WCAG conformance. The 75% of issues scans miss includes many of the criteria that affect users most directly: whether a screen reader user understands an image, whether a keyboard user can complete a form, whether the page makes sense when read aloud. Conformance claims depend on the layer scans cannot reach.