Do clean pages stay clean?
Published accessibility statistics almost always measure how common failures are, and WebAIM has done that thoroughly on a million homepages every year. This study asks a different question, the one our customers pay for: if a page is clean on a check today, what are the chances it still is in thirty days? That number is the argument for monitoring, and only time can produce it.
The sample
A frozen, dated slice of the Tranco top-million list: every 3,333rd entry, giving 300 sites, frozen on 18 October 2026. The list is public, the sampling rule is written down, and the sample does not change: a study that quietly reselects its subjects can prove anything. It is a sample of homepages, not of websites and certainly not of pages in general.
What we do, week by week
- One request per site per week, to the homepage only. Not a crawl, not a load test.
robots.txtis obeyed; a site that asks not to be fetched is marked as such and counted.- The same 25 checks run, in the same version, and the version is recorded with every scan. If a check changes, the change is announced here rather than silently improving the numbers.
- A page that cannot be read is counted as unreadable, never as clean. That distinction is the whole reason this product exists, and the study would be a lie without it.
What we publish, and what we never publish
Aggregates, with the number of pages behind every figure, and the number of sites that could not be read. No site is ever named, ranked or shamed. We do not claim that a detected failure means a website is inaccessible, or that it fails any law: we report that a page was found to fail a specific check on a specific day, which is a much smaller and more defensible claim.
What the data says so far
Two weeks, 18 and 25 October 2026. One week of change is not a trend, and we will not pretend it is — but it is already telling us something.
The first week's snapshot
Of 300 homepages: 175 readable, 110 unreadable, 7 asked not to be fetched. Among the readable ones, 69% carry links that open a new window without warning, 57% have images without alt text, 43% have links with no accessible name, and 40% have unlabelled form fields. The full table is below the change figures, because the change figures are the interesting part.
What changed in a week
173 homepages were readable in both weeks — that is the only honest denominator for a change, and it is much smaller than 300. Of those:
- 0 gained a new failure of any check.
- 1 fixed something: a link that used to open a new window without warning stopped doing so.
- 117 sites could not be read in the second week at all — more than a third of the sample. They are not clean and they are not dirty; they are unknown, and any study that counted them as either would be lying to you.
That is an uncomfortable result for a company that sells monitoring, and it belongs on the page rather than in a drawer. Two explanations fit, and we do not yet know which is right. Either homepages genuinely change slowly — a homepage is often a template plus a few hand-edited blocks, and the edits that break accessibility are more likely on articles, product pages and landing pages that are published weekly — or one week is simply too short an interval to catch change.
What we are doing about it, in the open: the sample stays frozen, and from next week each site is asked for a second page as well — its most recent article or blog post where we can find one — so that the study measures the pages that actually change. The frame will be recorded here on the day it changes, not edited quietly afterwards.
Where each check stands
| Check | Homepages carrying it | Gained a failure in a week |
|---|---|---|
| Link opens a new window without warning | 121 of 175 (69%) | 0 of the 53 that were clean |
| Image without alt text | 99 (57%) | 0 of 75 |
| Link with no accessible name | 75 (43%) | 0 of 98 |
| Form field without a label | 70 (40%) | 0 of 104 |
| Heading levels that skip | 60 (34%) | 0 of 114 |
| Duplicate id attributes | 54 (31%) | 0 of 119 |
| Button with no accessible name | 47 (27%) | 0 of 126 |
| Headings start below h1 | 36 (21%) | 0 of 137 |
| Zoom is blocked | 32 (18%) | 0 of 141 |
How to reproduce it
The study runs on our own server with the same rules engine the products use, and the methodology above is enough for anyone with a crawler to repeat on their own sample — which is the point of publishing the sampling rule. If you want the raw aggregates when there are more of them, ask: hello@accessaudit.dev.
Why a monitoring company publishes this
Because the honest version of our sales argument is a measurement, not a slogan. If pages that are clean today tend to break quietly, scheduled checks are worth paying for; if they do not, you should save your money and we would rather find that out in public.