Somebody runs an accessibility scanner on your website. It returns a score, or a green tick, or a list of eleven issues. It is tempting to treat that as the finish line.
It is closer to half the job, and the half it misses is the half that actually stops people using the site.
Automated tools find between 30 and 57 percent of problems
Depending on which study you read, automated scanners reliably detect somewhere between 30 and 57 percent of accessibility violations. That is a useful share and worth having. It is not a conformance claim.
Scanners are genuinely good at a few things:
- Colour contrast, which is measurable arithmetic
- Missing alt attributes
- Invalid ARIA
- Form fields with no programmatic label
Those are real failures and they are worth fixing. Contrast in particular is the single most common accessibility failure on the web, so a scanner earns its keep on that alone.
What a scanner cannot judge
A machine can tell you an image has alt text. It cannot tell you the alt text is right. An image of a staff directory described as “image1” passes the check and helps nobody.
The list of things no scanner evaluates is longer than most people expect:
- Focus order. Whether tabbing moves through the page in an order that matches what you see.
- Keyboard traps. Whether you can get back out of a menu or a modal once you are in it.
- Whether alt text is accurate, as opposed to present.
- Dragging alternatives. Whether a slider or a reorderable list can be operated with a single tap.
- Consistent help. Whether your contact link sits in the same relative place across pages.
- Authentication. Whether logging in requires memorising or transcribing something, with no alternative.
- The screen reader experience, which is the whole point and cannot be inferred from markup alone.
A real conformance claim needs a scan, plus manual keyboard testing, plus screen reader testing. Anyone selling you a certificate off a scan alone is selling you the 30 percent.
The manual pass that finds the rest
This is the minimum worth doing on each distinct page template. It takes under an hour once you know the moves.
- Keyboard only. Put the mouse away. Tab through the whole page. Every interactive element reachable, focus always visible, focus order matching the visual order, no traps, and the sticky header never covering the field you are on.
- 200 percent zoom, and 320 pixels wide. No sideways scrolling, nothing clipped.
- Text spacing. Increase line height, letter spacing and word spacing to the values in criterion 1.4.12. Nothing should overlap or get cut off.
- Screen reader. NVDA on Windows, VoiceOver on a Mac. Navigate by heading, then by landmark, then by link. The headings should form a sensible outline on their own. Form fields should announce their label and their error.
- Reduced motion switched on at operating system level. Animations and hover transforms should stop.
- Forms. Every input labelled. Errors described in words, announced, and never signalled by colour alone.
The failures that turn up most often
Aggregate audit data puts these in roughly this order, most frequent first:
- Contrast that does not meet the minimum ratio
- Missing or useless alt text
- Custom controls built from divs, with no role and no accessible name
- Focus outlines removed in CSS with nothing put back
- Headings chosen for size rather than structure, and tables without header cells
- Placeholder text used instead of a label
- Focus hidden behind a sticky header, the most common of the criteria new in WCAG 2.2
Four of those seven are decisions made once in a design system. That is the argument for doing this at build time rather than as a remediation project later.
What to ask for
If you are buying an accessibility audit, ask which of the three parts you are getting: the scan, the keyboard pass, the screen reader pass. Ask how many templates are covered rather than how many pages, because templates are what actually get fixed.
And ask what happens after. An audit produces a list. Somebody still has to do the work, and on most sites the fixes belong in the theme, not in two hundred individual pages.
Sources: W3C, How to Meet WCAG quick reference; WebAIM annual accessibility analysis of top home pages.