Published 2026-06-12

Web accessibility in 2026 — the rules are on, and 95.9% of the web still fails

2026 is the year after the year accessibility stopped being "nice to have" and started being "costly to skip." The EU's European Accessibility Act (EAA) has passed its first year in force, and in Korea web accessibility is already a legal obligation under the Anti-Discrimination Against Persons with Disabilities Act. Yet 95.9% of the top one million home pages still fail on the basics [WebAIM Million, 2026]. In a year when the rules switched on, the answer is not more ARIA (Accessible Rich Internet Applications) and not an accessibility overlay — it is fixing the most common violations first, as facts. (This article is information, not legal advice.)

What changed in accessibility regulation in 2026?

Regulation moved from "recommended" to "required." The EU's EAA became enforceable on 28 June 2025, with all 27 member states transposing it into their own national law. Korea already treated web accessibility as a legal duty before that.

The EAA requires accessibility for digital products and services in areas like e-commerce, banking, e-books, and transport. But each member state supervises and penalizes independently, and microenterprises (fewer than 10 staff, under €2M turnover) are exempt [Davis Wright Tremaine, 2025]. So it is less a single EU rule than a patchwork of 27.

Korea is a separate track. Web accessibility is mandatory for all public and private sites under Article 21 of the Anti-Discrimination Against Persons with Disabilities Act and the Framework Act on Intelligent Informatization, and a government-designated body issues a quality-certification mark [Korea Web Accessibility Certification Evaluation Institute, current]. The national standard (KWCAG) was revised in late 2022 to reflect WCAG 2.1/2.2, adding 9 criteria to the previous 24 [NIA, 2022].

Is the EU rule actually being enforced?

The rules are on, but enforcement is still finding its footing. A year in, there are no documented monetary fines from EU regulators under the EAA. Enforcement so far has been market surveillance, formal notices, and audit rejections — "remediate first, fine later" [Lainey Feingold, 2026].

France has gone furthest. In July 2025, the disability associations apiDV and Droit Pluriel sent formal legal notices to four major grocers (Auchan, Carrefour, E.Leclerc, Picard); when remediation fell short, they filed emergency injunctions in November 2025 — Europe's first EAA-related lawsuits [IBA, 2025].

The first ruling came on 5 May 2026, and the Auchan case was dismissed. But do not misread it. The dismissal was on a revenue threshold — the court had to decide which French law applied, applied the higher turnover threshold, and Auchan E-Commerce fell below it. The court did not find the site accessible. It actually found strong or major failures in 13 of the 19 sections audited [Silktide, 2026]. The associations appealed. In short, enforcement in 2026 is not "lax" — it is "still testing the procedure."

Where is WCAG now?

WCAG 2.2 is the baseline; WCAG 3.0 is not finished. The standard to follow in 2026 is WCAG 2.2, not the in-progress 3.0.

WCAG 2.2 was published as a W3C Recommendation in October 2023, updated in late 2024, and adopted as the ISO/IEC 40500 international standard [W3C WAI, current]. WCAG 3.0, by contrast, is still a Working Draft. The latest draft came in March 2026 and points toward replacing the A/AA/AAA levels, but a final Recommendation is expected only around 2028–2030 [W3C WAI, 2026].

The practical takeaway is simple: there is one standard to adopt now — WCAG 2.2 — and no reason to defer because of 3.0.

What is the real state of the web — and what do you fix first?

Regardless of rules or standards, the live web fails on fundamentals. WCAG 2 violations were detected on 95.9% of the top one million home pages (up from 94.8% in 2025), with 56.1 errors per page on average [WebAIM Million, 2026]. So the priority is clear — start with the most common violations.

The details show what to fix first. 41.8% of pages skip heading levels (jumping from h2 straight to h4), 16.2% of home-page images (10.8 per page on average) sit there with no alt text, and of 948,225 tables observed, only 19% had valid data-table markup [WebAIM Million, 2026]. These three are both common and relatively easy to fix.

(WebAIM detects only what automated tooling catches, so the real count may be higher — these are "detected" values.) The starting point for 2026 is not a flashy redesign — it is closing the high-frequency, low-difficulty items first.

Is more ARIA the fix? Are overlays the answer?

Neither. Layering on ARIA is not a shortcut, and installing an accessibility overlay is not either — both have evidence of backfiring.

Start with ARIA. The data runs the opposite way: pages that use ARIA average more errors, not fewer (59.1 vs 42) [WebAIM Million, 2026]. That is correlation, not causation — complex pages tend to use more ARIA. But the lesson holds. The first W3C rule is "use a native HTML element instead of ARIA where you can," and "No ARIA is better than bad ARIA" [W3C ARIA in HTML, current].

Overlays deserve a more honest look. An overlay is a one-line script that advertises automatic accessibility fixes. Its defenders argue it can supplement some user-adjustable controls — contrast, font size, keyboard toggles. But "make the whole site compliant" is a different promise. In April 2025 the US FTC reached a $1 million settlement and order against the overlay vendor accessiBe — finding its claim that "accessWidget can make any website WCAG-compliant" to be unsubstantiated, deceptive advertising [Lainey Feingold, 2025]. Overlays did not even stop lawsuits: one tally found about 25% of digital accessibility lawsuits targeted sites using overlays, with cases where the overlay clashed with assistive technology and became a barrier itself [TestParty, 2024]. The conclusion is not vendor-bashing — it is a measurable limit: an overlay may be a supplement, but it is not a substitute for real fixes.

This also shapes search and AI visibility

One note to add. The structure a screen reader reads well — semantic markup, heading hierarchy, alt, tables — is the same structure an AI extractor reads well. Accessibility does not guarantee AI citation, but it builds the precondition for it. We covered that link separately → How web accessibility affects AI citation.

What to check first — as facts

Rules and standards both come back to one question: what is my page violating right now? That is a matter of direction, not a score — whether the heading hierarchy holds, whether images have alt text, whether tables are marked up as data tables. Each is a countable fact.

zupzup diagnoses these signals across 9 categories and 84 analyzers, and shows what to fix first in priority order across four scores, accessibility included. For Korean sites, heuristics like table accessibility and language tags (BCP47) carry their own weight. zupzup does not track search rankings or AI citation counts — because it can't. It points to the most common violations as facts instead. Only what we can measure.

In the year after the regulation year, the starting point is not a grand declaration — it is one diagnosis. Start by checking what your page needs to fix first.