OPEN RESEARCH
US commercial websites: what the automated check finds
Shops, cafés, clinics, banks, libraries — the sites a person uses every day. Each one is opened in a real browser and checked with the axe-core engine: how many formal accessibility rules are broken in the markup and the styles.
Data snapshot as of September 10, 2026
Two different instruments — and their numbers do not add up
This site carries two measurements, and they answer different questions. The keyboard traversal of government sites checks whether a person can reach the goal pressing only Tab. This page is an automated check: the machine counts rule violations in the code. Neither follows from the other — which is precisely why the gap between them became the main result of the U.S. study. There is no combined figure.
| keyboard traversal | this page | |
|---|---|---|
| what sites | government, CISA registry | commercial organisations |
| how measured | a real Tab keypress | axe-core |
| what it shows | whether a person reached the goal | how many rules are broken |
95,524 · 11,659 · /research/data
How the check ended
| Outcome | Records | Share |
|---|---|---|
| Measured | 91,710 | 68.9 % |
| Bot wall | 21,730 | 16.3 % |
| Did not finish rendering | 7,680 | 5.8 % |
| Page missing, server says otherwise | 4,094 | 3.1 % |
| Empty page | 3,073 | 2.3 % |
| Server silent | 2,798 | 2.1 % |
| Domain not in DNS | 1,875 | 1.4 % |
| Broken encryption | 239 | 0.2 % |
How severe it is
The axe-core scale. «Critical» is an obstacle that makes the page unusable outright: a button with no name, a form field with no label. «Serious» is what makes using it painful: unreadable text, a link that gives no clue where it leads.
What the shares are counted from
Shares are counted from LOG RECORDS, not from the number of organisations and not from the number of sites that opened. One organisation yields several records: several of its pages are checked. Therefore the phrase «N % of sites are inaccessible» does not follow from these numbers and is not said here.
What this study does not say
- — An automated check does not replace a person. It finds a broken rule, but does not say whether it gets in the way in practice — and it misses what gets in the way while breaking no rule at all.
- — A violation is not the same as a barrier. In the U.S. study, 53.8 % of the pages the machine called clean turned out to be impassable for a keyboard user.
- — A fifth of the sites did not answer at all, and almost a tenth closed themselves to the check. These are neither «accessible» nor «inaccessible» — they are unmeasured, and they take no part in the conclusions.
- — Log records are counted, not sites. The same address can yield several records.
Who did this
Maksim Galatin, self-funded research. The scanner was written from scratch; the raw data and the methodology are published in full. The list of organisations used for traversal comes from OpenStreetMap, © OpenStreetMap contributors.
Conflict of interest disclosure: the author provides accessibility remediation services. The research is self-funded; nobody paid to be included in the sample or excluded from it.