The European Accessibility Act and visual testing

With that said, the Act has created a real engineering problem that is worth understanding properly, because a lot of the marketing around it is misleading.

What the Act actually says

The European Commission's own scope list covers computers and operating systems, ATMs and ticketing and check-in machines, smartphones, TV equipment for digital television, telephony services and equipment, access to audiovisual media services, air, bus, rail and waterborne passenger transport services, banking services, e-books, and e-commerce.

Those last three are why this reached a lot of engineering teams who had not previously considered themselves in scope. If you sell to EU consumers online, e-commerce is on the list and where your company is headquartered does not exempt you.

Where visual testing genuinely helps

Here is the honest version, which is narrower than most vendors imply but still useful.

An accessibility audit tells you whether you conform today. It is a point-in-time assessment, usually expensive, usually done by specialists. What it cannot do is stop a developer three months later from shipping a CSS change that drops your button contrast below threshold.

That is the gap: conformance is achieved by audit and lost by regression. Visual testing is a guard against the second, not a substitute for the first.

WCAG 2.1 AA criteria a visual test can guard

These have measurable visual signatures, so an automated check can tell you when they change:

What visual testing cannot touch

This list matters more than the one above, because it is where the misleading marketing lives. None of these have a visual signature, and no screenshot comparison will ever detect them:

• 1.1.1 Non-text Content — whether alt text exists, and whether it is meaningful. A screenshot cannot read alt text, and no automated tool judges whether a description is useful.
• 1.3.1 Info and Relationships — semantic structure, heading hierarchy, form labels, ARIA. Invisible by definition.
• 2.1.1 Keyboard — whether everything is operable by keyboard, and in a sensible order.
• 2.4.3 Focus Order — a visually identical page can have an incoherent tab order.
• 3.3.x Error handling — whether errors are identified and described in a way assistive technology conveys.
• 4.1.2 Name, Role, Value — whether controls expose correct accessible names and roles.
• Actual usability with a screen reader — which is the thing the standard exists to protect, and which only testing with one reveals.

Widely-cited estimates put the share of WCAG issues detectable by any automated tooling — not just visual — at well under half. Visual regression testing addresses a subset of that subset. It is a regression guard, not an audit, and anyone selling it as compliance is overselling it.

A defensible engineering position

1. 01Get a real audit. Specialists, against EN 301 549 and WCAG 2.1 AA. This establishes conformance; nothing automated substitutes for it.
2. 02Fix what it finds, and record what "correct" looks like for the fixed pages.
3. 03Add automated regression guards on the visually-detectable criteria above, so the fixes stay fixed as the product changes.
4. 04Run dedicated accessibility checks too — axe-core or similar in CI catches a different, larger set than visual comparison does. They are complementary, not alternatives.
5. 05Keep manual keyboard and screen-reader passes for anything new. There is no automated replacement.
6. 06Re-audit periodically. The regression guards buy you time between audits; they do not replace them.

Step 3 is the only one this site is about, and it is deliberately the third item rather than the first.

Why this is worth engineering attention now

Not because a tool will make you compliant — it will not. Because accessibility conformance is unusually easy to lose silently. A contrast ratio that passed in March fails in June after a design-token change nobody connected to accessibility. A layout that reflowed correctly at 320px stops after a new promo banner. Nothing alerts on either.

That is the same structural problem as every other visual defect — nothing in a normal pipeline is looking — with a regulatory consequence attached to getting it wrong.