Accessibility testing

On this page 7

Accessibility testing is the practice of evaluating whether a website or web application can be used by people with disabilities, including users who rely on keyboards, screen readers, magnification, voice control or captions. In web testing, it sits beside functional, performance and cross-browser testing, and it measures a product against a defined conformance target, most often WCAG 2.2 level AA.

Effective accessibility testing combines automated scanning, manual expert review and testing with assistive technology. No single method is sufficient, because automated rules can detect only a minority of barriers, while manual methods are slower but reveal whether real tasks can actually be completed.

What accessibility testing measures

Accessibility testing checks a page against the four WCAG principles: content must be perceivable, operable, understandable and robust. Each principle breaks down into testable success criteria at levels A, AA and AAA, such as text contrast, keyboard operability, form labels, captions and the correct exposure of names, roles and states to assistive technology.

Legal and contractual requirements usually refer to these criteria. Section 508 in the United States, the European standard EN 301 549 and the European Accessibility Act all build on WCAG, so a test report that maps findings to specific criteria is directly usable for compliance documentation such as a VPAT or an accessibility statement.

Methods of accessibility testing

Accessibility testing uses several complementary methods, each with a different cost, speed and coverage. The table below compares the main approaches used by web teams.

MethodWhat it findsMain limitation
Automated scanningMissing alt text, contrast, invalid ARIACannot judge meaning or task flow
Manual keyboard testingFocus order, traps, hidden focusRequires time on every template
Screen reader testingNames, announcements, reading orderNeeds trained testers
Expert auditFull criterion-by-criterion conformanceExpensive and periodic
Usability testing with disabled usersReal task barriers and workaroundsSmall samples, slower to organize
Main accessibility testing methods

Automated engines such as axe-core, which also powers the accessibility category in Lighthouse, apply deterministic rules to the DOM and the accessibility tree. Studies of real audits commonly estimate that automation detects roughly 30 to 40 percent of WCAG issues, depending on how issues are counted, so a clean automated score never proves conformance.

Automated and manual checks in practice

The boundary between automated and manual work follows a simple rule: machines verify presence and syntax, humans verify meaning. A scanner can confirm that an image has an alt attribute, but only a person can decide whether the text describes the image correctly or whether the image should be decorative.

CheckAutomatableHuman judgment needed for
Color contrast of textMostlyText over images and gradients
Alternative textPresence onlyAccuracy and purpose
Form labelsPresence and associationClarity of instructions
Keyboard accessPartlyLogical order and visible focus
Heading structureLevel skipsWhether headings describe content
Captions for videoTrack presenceAccuracy and synchronization
Which checks can be automated

A practical manual pass covers keyboard-only navigation, zoom to 200 and 400 percent, reflow at a 320 CSS pixel viewport, and a screen reader walkthrough. Common pairings are NVDA or JAWS with Chrome or Firefox on Windows and VoiceOver with Safari on macOS and iOS, because assistive technology behaves differently across browser combinations.

Integrating accessibility testing into the workflow

Accessibility testing is most effective when it shifts left into design and development instead of arriving as a final audit. Fixing a contrast problem in a design token costs minutes, while fixing it after release can require changes across dozens of templates and a new regression cycle.

Teams usually layer the methods across the delivery pipeline, as the following table shows.

StageTypical activityTools or methods
DesignContrast and focus state reviewDesign annotations, contrast checkers
DevelopmentLinting and component checkseslint-plugin-jsx-a11y, Jest with jest-axe
Continuous integrationAutomated page scansaxe with Playwright, Cypress or Selenium
Pre-releaseManual keyboard and screen reader passNVDA, VoiceOver, browser DevTools
PeriodicFull audit and user testingExpert review, disabled participants
Accessibility testing across the delivery pipeline

Automated checks in continuous integration should fail the build on new critical violations rather than report every historical issue, otherwise teams learn to ignore the results. A maintained issue backlog with severity ratings turns audit findings into scheduled engineering work.

Limitations and common mistakes

The most common mistake is treating a perfect automated score as proof of accessibility. A page can score 100 in Lighthouse and still trap keyboard focus in a modal, announce buttons as “clickable” with no name, or present a checkout flow that no screen reader user can finish.

  • Testing only the home page misses forms, error states and authenticated areas where barriers concentrate.
  • Ignoring dynamic states such as open menus, validation errors and loading content leaves many defects untested.
  • Relying on accessibility overlays does not replace testing and often introduces new problems.
  • Testing with a single screen reader hides browser-specific differences in announcements.
  • Skipping retests allows fixed issues to return in later releases.

Reports also lose value when findings lack context. A useful finding names the affected WCAG criterion, the component and page, steps to reproduce, the user impact and a concrete remediation, so developers can act without repeating the investigation.

The idea in brief

Accessibility testing checks whether everyone can use a website, including people who are blind, cannot use a mouse or cannot hear audio. Testers use the site the way these people do, with a keyboard only, with software that reads the screen aloud, or with the page zoomed in.

Computer tools catch some problems quickly, but people must check the rest. The goal is not a good score but a site where every user can finish the same tasks.

A practical scenario

A city library prepares to launch a new online booking system for study rooms. The development team adds axe checks to its Playwright tests, and the scan reports only four minor issues, all fixed within a day. Before launch, a tester runs a manual pass with NVDA and Firefox and with keyboard-only navigation.

The manual pass reveals that the date picker cannot be operated without a mouse, the time slots are announced only as “button”, and the confirmation message appears visually but is never announced. None of these barriers appeared in the automated report.

The team replaces the date picker with native date inputs, adds accessible names that include the date and time, and announces the confirmation through a live region. A follow-up session with two blind library members confirms that both can book a room in under three minutes, and the system launches meeting its WCAG 2.2 AA target.