1,000,000
Homepages analyzed
A homepage-level sample, not one million complete websites.
Research report
Published 2026
Primary data: Feb 2026
Reviewed: Aug 27, 2026
State of Web Accessibility 2026
An independent synthesis of public accessibility research, centered on WebAIM's evaluation of one million homepages and supported by HTTP Archive context.
Headline finding
95.9% of the one million homepages evaluated had automatically detectable WCAG failures, with 56.1 detected errors per homepage on average. Both figures are worse than 2025, reversing six consecutive years of gradual improvement.
Source: WebAIM Million 2026. Automated results do not establish full WCAG conformance.
- Primary sample
- 1,000,000 homepages
- Evaluation type
- Automated detection
- Primary period
- February 2026
Executive summary
In 2026 the web moved backwards
For six straight years the share of homepages with detectable failures inched downward. In February 2026 it rose instead, from 94.8% to 95.9%, while the average number of detected errors per homepage climbed 10.1%. Two years of measured progress were erased in a single cycle, putting the web back at roughly its 2024 position.
This report keeps measurements from different studies separate and distinguishes automated signals from complete accessibility evaluation.
01
The same six failures, for a seventh year
Six categories account for 96% of the automatically detected errors in WebAIM's sample, and the list has not changed in seven years. Four of the six became more common in 2026.
02
Complexity is outrunning remediation
The average homepage contained 1,437 elements in February 2026, a 22.5% jump in a single year and nearly double the 2019 figure. ARIA attributes grew 27% in the same period.
03
Automated results are incomplete
These findings identify repeatable code-level failures, not full WCAG conformance or the lived usability of a website.
At a glance
The scale of detectable barriers
95.9%
With detectable WCAG failures
Up from 94.8% in 2025. Automatically detectable failures only; not a full conformance result.
56.1
Average errors per homepage
The mean across the February 2026 sample, up 10.1% from 51 in 2025.
3.9%
Elements with detected errors
About one detected error for every 26 page elements.
Six common detectable failures
The six failure types reported by WebAIM across the one-million-homepage sample, and how each moved since 2025.
Low contrast remains the most widespread detected issue and grew sharply, from 79.1% to 83.9% of homepages. Four of the six categories became more common in 2026; only missing alternative text and missing document language improved. Together these six account for 96% of all errors WebAIM detected, and the list has been unchanged for seven years.
1. Low contrast text83.9%
2. Missing alternative text53.1%
3. Missing form input labels51%
4. Empty links46.3%
5. Empty buttons30.6%
6. Missing document language13.5%
Source: WebAIM Million 2026
Six years of progress, undone in one
WebAIM Million results for detectable failures and average errors per homepage, 2019-2026.
Every year from 2020 to 2025 improved on the one before it. 2026 broke the pattern: detectable failures rose from 94.8% to 95.9% and average errors rose from 51 to 56.1, returning both measures to roughly where they stood in 2024.
201997.8%with failures59.6average errors
202098.1%with failures60.9average errors
202197.4%with failures51.4average errors
202296.8%with failures50.8average errors
202396.3%with failures50average errors
202495.9%with failures56.8average errors
202594.8%with failures51average errors
202695.9%with failures56.1average errors
Source: WebAIM Million 2026
Interpretation
What the numbers do and do not show
Automated measurements are useful for tracking recurring code-level barriers. They are not a substitute for manual review or testing with disabled people.
Pages are getting more complex
WebAIM found an average of 1,437 elements per homepage, up 22.5% in a single year and nearly double the 2019 figure. More markup means more places for a barrier to appear.
ARIA is not a shortcut
82.7% of homepages used ARIA, and homepages now carry 133 ARIA attributes on average, up 27% in one year. Pages with ARIA averaged 59.1 detected errors versus 42 without it, an association that also reflects greater page complexity.
A score is not conformance
The HTTP Archive Web Almanac reported a median Lighthouse accessibility score above 85in 2025, improving 1% year over year even as WebAIM's error counts rose. Lighthouse tests only part of WCAG, so even a perfect automated score cannot establish accessibility.
What changed
Why 2026 moved backwards
WebAIM attributes the reversal primarily to two things that grew faster than remediation did: page complexity and ARIA. Homepages gained 22.5% more elements in a single year and 27% more ARIA attributes. In WebAIM's own reading, this likely reflects heavier reliance on third-party frameworks and libraries alongside automated and AI-assisted coding practices. Their conclusion is that accessibility at scale will need both better practices and simpler systems.
The underlying structural measures below show where that complexity lands. They are drawn from the same February 2026 sample and are useful as a benchmark when auditing your own pages.
| Measure | Homepages | Why it matters |
|---|---|---|
| Had at least one landmark region | 84.3% | Landmarks let screen reader users jump straight to a section instead of reading linearly. |
| Had a main element or main landmark | 46.1% | Fewer than half of homepages mark their primary content, so most offer no reliable jump target. |
| Had a skip link | 17.1% | One in ten of those skip links was broken: hidden inaccessibly, or pointing at a target that was not there. |
| Had skipped heading levels | 41.8% | Headings are the primary way screen reader users navigate; skipped levels misrepresent the document outline. |
| Had more than one h1 | 18.1% | Up from 16.3% in 2025. Multiple top-level headings obscure what the page is actually about. |
| Had ambiguous link text | 15.2% | Links such as “click here” or “more” carry no meaning when read out of context in a links list. |
| Used an ARIA menu role | 5.7% | 22% of those menus introduced barriers through missing menu markup or interactions. |
Each of these maps onto a fixable pattern. For the contrast failures that top the list, see 1.4.3 Contrast (Minimum) and the contrast checker. For unlabeled inputs, which were 33.1% of all form inputs in the sample, see the accessible forms guide. For broken skip links, see 2.4.1 Bypass Blocks. And for the ARIA menus that introduced barriers on a fifth of the pages using them, the accessible menu & menu button guide explains why most navigation dropdowns should not use role="menu" at all.
The same February 2026 sample can be cut two other ways. See how the error count varies by technology stack across content management systems, frameworks and libraries, or by industry across 29 sectors.
Data Sources
- WebAIM Million 2026
Automated evaluation of the home pages for the top one million websites in February 2026, published 30 March 2026.
- HTTP Archive Web Almanac 2025
Accessibility analysis based on HTTP Archive data and Lighthouse audits from the 2025 dataset, published January 2026.
Sample Size
1,000,000 homepages in the primary WebAIM dataset
Date Range
February 2026 primary sample; 2025 HTTP Archive context
Limitations
- Automated tests evaluate only a subset of WCAG requirements and cannot establish conformance.
- The primary WebAIM sample covers homepages, not every template or user journey on each website.
- HTTP Archive findings are presented as separate context and are not merged into the WebAIM sample. The two sources use different engines (WAVE versus Lighthouse and axe-core) and can move in different directions in the same period.
- Observed associations, including the relationship between ARIA usage and detected errors, do not prove causation. WebAIM notes that pages using more ARIA were also more complex.
- Year-over-year comparisons reflect a changing sample: the underlying one-million-site list is rebuilt from Tranco rankings each year, so some movement reflects which sites were measured.
Last updated: 2026-08-27
Use the research
Download, cite, and share
Reuse the findings with a clear citation and retain links to the primary sources when publishing extracts.
Download report data
The PDF includes key findings, trend tables, methodology, and source links. The CSV contains the underlying values used here.
Cite this report
Accessibility.build. (2026). State of Web Accessibility 2026. https://accessibility.build/research/state-of-accessibility
Questions about the report
Frequently asked questions
Definitions and limitations that are important when quoting these findings.
Get the data
Every number on this page is available as CSV and JSON under a CC BY 4.0 licence. Cite Accessibility.build and the underlying sources (WebAIM Million; HTTP Archive Web Almanac). Last updated .
- Most common failuresCSV, Most common failuresJSON, Most common failures
- Year-over-year seriesCSV, Year-over-year seriesJSON, Year-over-year series
Essential Accessibility Resources
Comprehensive tools, checklists, and guides to help you create inclusive digital experiences
contrast
wcag
accessibility
+1 more
Top Pick
Accessible Forms Guide
Build forms everyone can complete: labels, required fields, accessible validation and error messages, fieldset grouping, autocomplete, and multi-step patterns mapped to WCAG 2.2 AA
form accessibility
form labels
Top Pick
WCAG 3.3.2 Labels or Instructions Guide
Complete guide to accessible form labels: the four label association methods, why placeholders fail, required and format hints, code examples, and testing
form labels
form accessibility