The 2025 Web Almanac accessibility report ends on an all too familiar message.
Web accessibility is improving, but painfully slowly.
Automated scores are going up while basic problems like poor contrast, missing labels and broken media stay common. Rules like the EU Accessibility Act are creating pressure, but real-world change lags behind legal deadlines. The report confirms what experts have warned for years. Automated tools and AI cannot replace human judgement.
Nothing in this report really shocks me though. And it shouldn't shock you if you've been paying attention to web accessibility over the past decade.
Progress is extremely slow because accessibility always loses priority.
When features that make money compete with accessibility work that seems optional, of course it gets pushed aside. 1-2% yearly improvements reflect this reality. Old code is expensive to fix. And most developers never even hear about accessibility until they see an audit. Each new generation repeats the same mistakes because we haven't made accessibility education essential.
Automated tools catch less than 50% of real accessibility problems because accessibility is basically a human problem, not a technical one. Good alt text depends on context. Why is this image here? What does the user need to know? No AI can reliably answer these questions. You can pass every automated check and still create a confusing experience.
Technical compliance never equals usability.
Overlays failing is no surprise either.
Any technology promising to "make your site accessible with one line of code" is shit. If your HTML is badly structured, your navigation confusing and your content unclear, no JavaScript layer can fix that. The fact that only 0.2% of high-traffic sites use them should tell you everything you need to know. Companies with resources know they don't work, that's why they don't use them.
AI following the same pattern as every hyped technology is also predictable.
AI models trained on billions of inaccessible websites will copy those patterns. Scale without quality just scales the problem. Garbage in, garbage out. With AI, it's just more of it. AI needs human oversight. We haven't learned our lesson and we're in the same repeated cycle. When big promises come up against real-world limits, you realise humans are still required.
This is not me saying it so I still have a job tomorrow. I'd love it if I didn't have to worry about the boring things, the usual suspects. I'd love it if I could work on interesting problems. I love to program. I love to design. It's still the silly things I have to worry about day to day though.
We've known how to build accessible websites since the 1990s. WCAG has existed for decades. What's not surprising is that without enforcement, most companies don't prioritise accessibility. And even with enforcement, change takes forever because it needs a cultural shift, not just a technical fix.
And this is the main issue I think has me worried.
Organisations are increasingly treating accessibility as a technical problem solvable by automation, when it's actually a design problem that requires human expertise.
This matters because there's a dangerous cycle emerging. Automated testing tools are getting better and AI capabilities are expanding. That would be great if it didn't create the illusion that full automation is just around the corner. Companies see these tools and think: "Great, we can finally tick the accessibility box without the expensive, time-consuming work of redesigning our approach."
But this is backwards.
Automated tools detect less than half of real barriers and many critical problems naturally require human judgement. Goodhart's Law applies perfectly here. When scores become the target, they stop being meaningful measures. That's how we end up with sites that pass tests but fail users.
The stakes are high because rules are tightening. The EU Accessibility Act, updated ADA requirements and various national laws create the legal pressure. But if companies respond by chasing automated compliance scores or using accessibility overlays rather than doing the hard work of inclusive design, we'll only see the surface appearance of progress.
Inclusive design is better design.
Accessibility requires embedding it in your design process, testing with users with disabilities and treating automation as what it really is. A helpful support tool, not a replacement for expertise. It means training your team, involving people with disabilities throughout development and accepting that there are few shortcuts you can take.
So yes, progress is slow. No, automation can't save us. But every company that chooses to do accessibility properly creates a more inclusive web. That's how real change happens. Not through magic solutions.
Accessibility has always been about prioritisation, not capability. And the capability has been there for a while now. But it seems to me we're waiting for automation to catch up.
But we can't automate our way to inclusion.