Automated accessibility testing tools detect approximately 57% of WCAG issues. That number comes from industry research and has been consistent across multiple studies. The American Foundation for the Blind published a widely cited analysis in 2025 arguing that automated tools systematically miss the issues that cause the most frustration for users with disabilities.
This is not a criticism of automated testing. It is a description of its role.
What automation catches well
Automated scanners excel at detecting issues that can be evaluated programmatically:
- Missing alt text on images
- Color contrast failures below WCAG ratio thresholds
- Missing form labels and broken label associations
- Missing document language declarations
- Duplicate IDs and invalid ARIA attributes
- Missing page titles and heading hierarchy problems
- Keyboard trap detection in many cases
These are high-volume, high-frequency issues. On a typical site, automated scanning catches the majority of individual issue instances because the same types of problems repeat across pages.
What automation misses
The 43% that automated tools cannot reliably detect includes issues that require human judgment:
- Alt text quality. A scanner can detect a missing alt attribute. It cannot determine whether
alt="image"oralt="DSC_0042.jpg"is meaningful. Evaluating whether alt text accurately conveys the content and purpose of an image requires understanding context. - Logical reading order. Automated tools can check DOM order but cannot evaluate whether the visual layout creates a reading sequence that makes sense when linearized for screen readers.
- Meaningful link text. A scanner can flag empty links but cannot evaluate whether "click here" or "learn more" provides sufficient context when read out of its visual context.
- Keyboard interaction patterns. Complex widgets like date pickers, modal dialogs, and custom dropdowns require specific keyboard interaction patterns. Automated tools can test for basic keyboard access but cannot fully evaluate whether the interaction model is usable.
- Cognitive accessibility. Plain language, consistent navigation, error prevention, and clear instructions require human evaluation against user needs, not code inspection.
Where the gap is closing
Machine learning is improving automated detection rates. Detection is projected to approach 70% by late 2025 or early 2026 as models get better at evaluating contextual issues like alt text quality and reading order. But even 70% leaves a significant gap that requires human review.
What this means for compliance programs
A compliance program built entirely on automated scanning will miss roughly 4 out of every 10 accessibility issues. Some of those missed issues will be the ones most likely to generate complaints, because they are the ones that most directly affect usability.
The effective approach combines automated scanning with human review:
- Automated scanning provides breadth -- continuous monitoring across all pages, catching the high-frequency programmatic issues, and tracking remediation progress.
- Manual auditing provides depth -- evaluating the issues that require judgment, testing with assistive technologies, and validating that automated fixes actually work.
Neither approach alone is sufficient. Automated scanning without manual review creates false confidence. Manual auditing without automated scanning is too slow and expensive to cover an entire site.
The 57% number is not a limitation to apologize for. It is the reason automated scanning is essential -- it handles more than half the work at scale. The key is understanding what it covers and building the complementary human review process to cover the rest.