Disclosure: I'm Amit Mazuz, founder of Clearwren, an Atlassian Marketplace Partner. We made Clarable, an accessibility checker for Confluence Cloud.
The numbers below come from our checking engine, but you don't need our app to use any of this, and I won't mention it again.
Most accessibility studies look at home pages. Almost none look at documentation: the knowledge bases, how-to guides and policy pages people write in Confluence every day. In September I checked some of them. Here's what I found.
Why now
The US Department of Justice's ADA Title II rule says state and local governments must make their web content meet WCAG 2.1 AA. Under the April 2026 interim final rule, the deadline is 26 April 2027 for entities serving 50,000 people or more, and 26 April 2028 for smaller ones and special districts.
I'm not a lawyer. Whether a given Confluence space is covered depends on who can see it and what it's for, and the rule has exceptions, such as archived content. Ask your counsel. The findings below are useful either way.
What I checked
I took 1,182 pages from 13 Confluence Cloud sites that anyone can read without logging in: seven public knowledge bases, mostly in education, and six software and research-computing documentation sites. I sampled up to 100 random pages per site and checked them against WCAG 2.2. I only counted a failure when a tool can be certain. Anything that needs human judgment was set aside.
What I found
- About 2 in 3 pages had no level A or AA failure. About 1 in 3 did. With 100 pages per site, that number can move by up to 8 points between samples, so treat it as "somewhere between half and three quarters".
- Missing image descriptions were the most widespread problem: 1,474 images on 285 pages, at 12 of the 13 sites.
- Low text contrast came next: 1,396 instances, but on only 114 pages, also at 12 of the 13 sites.
- Another 258 images had a description that didn't describe anything: a file name, the word "image", or a caption copied from the text above.
- Software docs did better than knowledge bases: 71% of pages had no A or AA failure, against 57%.
No site was completely clean, and I'm not naming any of them.
Two things that surprised me
Contrast problems cluster. More than a thousand failures sat on about a hundred pages. That usually points to a few templates or one color choice used again and again, so fixing it once fixes a lot. I haven't confirmed that for every site.
Filled in doesn't mean done. An empty alt text field is easy to spot. A useless one, like "image1.png", isn't, and simple counts miss it.
A correction
The first version of this study counted some things as failures that WCAG doesn't: bare URLs as link text, an H1 inside the page body, skipped heading levels and empty headings. A reader pointed this out. I fixed the method on 7 October, measured a fresh sample, and replaced my earlier "roughly half" figure. If you saw that number, please use the one above.
Where I'd start
- Sample at random. Don't just check the top of the page tree. My first run did that and showed 90% of pages clean, because landing pages rarely have images or tables.
- Fix contrast at the source. Find the template or color and change it once.
- Agree on one alt text rule. If an image carries information, describe it. If it's decoration, mark it as decorative. Then decide who checks.
- Give each space an owner. When everyone writes pages and nobody owns them, the numbers above are the result.
Limits
This is a first look, not a full survey. I found the sites through web search, so they lean toward education and open-source docs, and private spaces may look different. Checks that need human judgment weren't counted, so a page with no failure here can still have problems. None of this says anything about legal compliance.
If you run documentation in Confluence, I'd like to hear what you find when you check a random sample of your own pages, and which fix turned out to be the cheapest.
