7,328 out of 7,573 evaluable websites showed at least one automatically detected accessibility violation. This is the central finding of Italian Web Accessibility 2026, a study by Giovanni Manetti for PerseoDesign Research, published on Zenodo on 13 October 2026.
That is 96.8% of evaluable domains. It does not mean that 96.8% of Italian websites have formally failed WCAG conformance. The original sample included 10,022 domains, but 2,449 could not be evaluated; furthermore, automated testing covers only a subset of accessibility requirements.
The project is called IWAS (Italian Web Accessibility Study). The scientific paper, research dataset, and reproducible acquisition pipeline are available separately.
The sample and how it was chosen
The starting population consists of the 10,022 .it domains in the Italian slice of Tranco list 46XZX, dated 8 July 2026. Tranco provides a popularity-ranked selection of websites, so this is not a random sample representative of all Italian websites.
The domain cohort exactly matches the previously published Italian Web Security Study (IWSS). Using the same original domains makes future matched-domain comparisons possible without treating security configuration and accessibility as interchangeable measurements.
Data were collected from 9 to 13 October 2026. Each website was rendered in Chromium through Playwright, then evaluated using axe-core 4.14.0.
The protocol selected up to five pages per domain: homepage, a functional page such as a contact form or search interface, a content page, a deeper URL and one deterministic seeded URL. Candidates were discovered only through homepage links and sitemaps. The study did not perform unrestricted recursive crawling.
Checks targeted automatically testable WCAG 2.0, 2.1 and 2.2 Level A and AA rules at a fixed desktop viewport of 1366 × 768 pixels. The protocol excluded authentication, form submission, manual keyboard testing and screen-reader evaluation.
What the data show
| Measure | Result |
|---|---|
| Domains in original cohort | 10,022 |
| Evaluable domains | 7,573 (75.6%) |
| Non-evaluable domains | 2,449 (24.4%) |
| Sampled pages from evaluable domains | 36,038 |
| Successfully evaluated pages | 34,641 |
| Evaluable domains with a detected violation | 7,328 (96.8%) |
| Violation-rule occurrences | 106,815 |
| Affected DOM nodes | 890,680 |
The difference between a rule occurrence and an affected element matters. A rule may fail on multiple elements and across several sampled pages. Accordingly, 106,815 does not represent 106,815 different websites or independent accessibility problems.
The median evaluable domain exhibited four distinct violation types and 13 rule occurrences across its sampled pages.
The most common detectable barriers
| Rule family | Domains affected | Share of evaluable domains |
|---|---|---|
| Insufficient colour contrast | 6,129 | 80.9% |
| Links without accessible names | 4,894 | 64.6% |
| Insufficient interactive target size | 3,080 | 40.7% |
| Missing or unsuitable image alternatives | 3,016 | 39.8% |
| Accessible names mismatching visible labels | 1,962 | 25.9% |
| Buttons without accessible names | 1,727 | 22.8% |
Contrast failures were detected on more than four out of five evaluable domains. These can arise in common interface components, including buttons, navigation, form feedback and secondary text. A visually polished design does not necessarily provide readable colour relationships in every state.
Missing accessible names are another recurrent problem. Links and icon-only controls may appear intuitive to sighted users while being difficult to identify through assistive technologies. Although a tool can detect missing names, evaluating whether a label accurately describes its destination often requires human judgement.
Image alternatives require similar care. Decorative images and informative images have different requirements; mechanically filling every alt attribute with a filename is not a substitute for a meaningful accessible description.
Website popularity is not a clear separator
Across the four Tranco ranking quartiles, the proportion of evaluable domains with at least one detected violation was:
| Popularity quartile | Domains with violations |
|---|---|
| Q1, most popular | 97.4% |
| Q2 | 97.4% |
| Q3 | 96.1% |
| Q4 | 96.2% |
All four groups exceed 96%. This suggests that a binary measure of whether any automated violation was found does not substantially distinguish popularity strata. It does not establish that popularity is unrelated to accessibility quality. A more informative comparison would account for violation types, impact, page coverage and non-response.
The 2,449 domains that could not be evaluated
The most important methodological limitation is the 24.4% non-evaluable portion of the original cohort. Among these domains, 1,150 returned HTTP 403, 386 produced DNS or non-public-destination errors, and 190 encountered navigation timeouts. The remaining failures include certificate, connection, HTTP and browser-navigation errors.
Blocking an automated browser does not imply that a website is inaccessible to people. The proper classification is not evaluated. Treating all these sites as error-free, or all as inaccessible, would distort the results.
The study also records 470 cross-domain redirects. Excluding those sites leaves an observed violation prevalence of 96.7%, close to the full evaluable-domain estimate of 96.8%.
At manuscript preparation, an anomalous selected-page entry and a stratified 88-row manual review still required final quality disposition. These matters should remain visible in scientific interpretation.
What this means for a website audit
The results help prioritise where to look, but the frequency of an automated rule finding is not the same as its impact on a specific person's task.
A practical audit should start with contrast on repeated interface components, meaningful accessible names for links and controls, and appropriate alternatives for informative images. Fixing one shared component can resolve the same issue across many pages.
The next stage is manual: keyboard navigation, focus order, forms, error feedback, interaction states and assistive-technology compatibility. Automated checks are useful screening tools, not complete WCAG conformance assessments.
A large-scale study describes patterns across a cohort. A website owner still needs to understand which tasks, users and components should be evaluated first.
The paper, dataset and research code
Paper: Manetti, G. (2026). Italian Web Accessibility 2026: An Observational Analysis of 10,022 Domains and 34,641 Web Pages (Version 1.0.0). Zenodo. DOI: 10.5281/zenodo.23347092.
Dataset: IWAS dataset on Zenodo.
Code and methodology: Italian web accessibility study on GitHub.
This study shares its input cohort with the IWSS web security paper but reports accessibility findings independently.
Editorial note: IWAS is the correct project acronym. The original Zenodo records and some deposited files use the earlier “IWASS” acronym. Both DOI identifiers remain unchanged.