Original measurement · 2–3 August 2026

99% of European online stores fail WCAG 2.1 AA. We measured 290 of them.

Nobody had published a conformance measurement of European e-commerce, so we ran one. 290 live Shopify stores across six EU markets, each tested on its home page and one real product page, on desktop and on mobile. 287 failed.

Everything below is reproducible. The tool is open source, the stores are public, the date is fixed. We publish the limitations in the same detail as the results, because a measurement you cannot criticise is not a measurement.

Method

We assembled a list of live Shopify stores selling into Germany, France, the Netherlands, Spain, Ireland and Sweden from a public directory of e-commerce sites, then verified each one was Shopify by inspecting its storefront HTML.

Each store was loaded with headless Chromium driven by Playwright and evaluated with axe-core against four rule sets: wcag2a, wcag2aa, wcag21a and wcag21aa. We tested the home page and one real product page, resolved from the store’s own catalogue rather than guessed.

Every store was tested twice: at 1366×900 and at 390×844. The figures reported here are the union of both — for each store, the worst result of the two.

What counts as a “broken element”

One individual element that fails one WCAG 2.1 AA success criterion — a single button with no accessible name, one image with no text alternative, one piece of text without enough contrast against its background. A store with 32 broken elements has 32 specific things to correct, each with an address in the code. We count elements, not rule types, because that is what a person actually has to fix.

The finding we did not expect: one viewport is not enough

Our first pass tested desktop only. When we repeated it on mobile across the first four markets, 22.8% of stores turned out to be worse on mobile than on desktop — scrollable tables, carousels and menus that acquire overflow once the layout collapses and become unreachable by keyboard.

We learned this the hard way, twice in one night: both this site and our parent agency’s site passed a desktop-only audit with zero violations, and both had real failures that appeared only at 390 pixels. A single-viewport audit reports a store cleaner than it is — and it errs in the direction that flatters whoever ran it.

Limitations

These are the reasons a careful reader should discount our numbers. We would rather state them than have them found.

• Automated testing finds roughly 30–40% of accessibility problems. Everything here is machine-detectable. The real figures are worse, not better: we did not test whether a person using a screen reader can select a variant or complete a purchase, which is where most of the harm lives.
• The sample is not random. Stores were drawn from a public directory, not from a census of all EU e-commerce. It skews toward stores that are indexed and active. We make no claim of statistical representativeness.
• Two pages per store. Home plus one product page. A full-site audit would find more. Whole categories of failure — checkout flows behind authentication, PDFs, video captions — were never touched.
• One moment in time. 2–3 August 2026. A store that installs an app next week will differ. Accessibility decays; a single measurement is a photograph, not a trend.
• Market sizes are uneven. Germany contributed 77 stores, Sweden 26. Per-market medians from smaller samples carry wider uncertainty, and we have not computed confidence intervals.
• We sell remediation. We have a commercial interest in these numbers being alarming. That is precisely why the method is published in enough detail for you to run it yourself and check.