Each year I read the WebAIM Million report, and each year the numbers are disappointing.

WebAIM’s latest analysis of the top one million homepages shows that 94.8 percent have detectable accessibility issues. After years of guidelines, awareness campaigns and automated tooling, the vast majority of popular websites still contain basic barriers.

Low contrast text. Missing alternative text. Unlabelled form fields. Misused ARIA. These are not advanced failures. They are fundamentals that should already be part of normal development practice.

Now layer artificial intelligence on top of that.

AI is increasingly being used to generate front-end code, layouts and components. If the web we are starting from is largely inaccessible, automation can either help improve it or scale the same problems much faster. That is why I was genuinely encouraged to see AIMAC released.

 

What AIMAC Is

The AI Model Accessibility Checker is an initiative from the GAAD Foundation in collaboration with ServiceNow. It introduces something the industry has been missing: an accessibility benchmark for AI-generated code.

Today, AI models are compared using benchmarks that measure reasoning ability, coding accuracy and performance on standard programming tasks. Those are useful metrics, but they do not tell us whether the output is inclusive.

AIMAC evaluates whether coding models generate accessible patterns. It looks at semantic structure, correctly associated labels, appropriate use of ARIA and accessible interaction behaviours. In other words, it tests not just whether the code works, but whether it works for everyone.

That is a meaningful shift.

 

Why Benchmarking Changes Behaviour

Benchmarks drive optimisation.

When models are publicly compared on speed or reasoning, vendors optimise for those dimensions. If accessibility becomes part of the measurable comparison, it becomes something model providers are incentivised to improve.

Without measurement, accessibility remains optional. With measurement, it becomes competitive.

That is how progress tends to happen in technology.

 

The Risk Of Scaling The Wrong Patterns

The WebAIM data makes one thing clear. We are not starting from a strong baseline. If 94.8 percent of leading homepages contain detectable issues, then much of the web that AI models learn from already includes accessibility failures.

Unless accessibility is explicitly tested and benchmarked, AI risks reproducing those same patterns at scale.

If generated code defaults to non-semantic structure, unlabelled inputs or inaccessible dialog behaviour, those patterns will spread faster. Automation magnifies whatever it touches. If the baseline is flawed, scale amplifies the flaw.

We have struggled for years to shift accessibility left in the development lifecycle. AI gives us another opportunity to do that properly, but only if we are intentional about it.

 

Moving Accessibility Upstream

Accessibility is too often treated as a final testing stage. Build first. Audit later. Fix what you can. The WebAIM figures show that approach has not delivered meaningful improvement at scale.

If AI is going to generate more of our digital interfaces, accessibility needs to be embedded at the point of generation. Not retrofitted after deployment.

A benchmark like AIMAC moves the conversation upstream. It shifts the focus from asking whether a finished site passes accessibility checks to asking whether the code is being generated in an accessible way from the start.

That is a more sustainable model.

 

My Take

The WebAIM Million report highlights limited progress across the mainstream web. At the same time, AI is accelerating how quickly code can be produced.

Those two realities intersect.

Seeing the GAAD Foundation and ServiceNow collaborate on AIMAC is encouraging because it introduces measurement into a space currently driven by assumption.

Benchmarks influence behaviour. If accessibility is measured, it can improve. If it is ignored, it will continue to lag behind.

AI will shape the next generation of digital products. The real question is whether it will replicate the 94.8 percent problem, or help reduce it.

AIMAC does not solve accessibility on its own. But it introduces accountability at the source. And if we are serious about building inclusive technology, that is exactly where accountability needs to sit.

 

Sources