AI will strengthen accessibility workflows, but it won't replace people.
Back in May, we held an internal webinar to mark Global Accessibility Awareness Day. Our Accessibility Champions Team got talking about a topic that's on everyone's mind right now: AI workflows, and where they fall short compared to human expertise.
We wanted to find out for ourselves, not just talk about it in theory. So the team set out to test a range of accessibility workflow tools to see where AI genuinely adds value and where human expertise still makes all the difference. We're sharing what we found because we think other teams working on accessible, ethical tech will recognise a lot of it.
We know AI can generate code, review websites against accessibility standards, or even produce a full AI audit report. But after this year's Global Accessibility Awareness Day (GAAD) webinar, one message stood out above everything else:
AI is a powerful assistant, but it isn't an accessibility expert.
Used well, AI can help teams work more efficiently. Used without human review, it can create a false sense of confidence: a website might look compliant simply because an AI tool hasn't flagged anything, while real barriers still exist for real users.
The current state of digital accessibility: Improving, but the fundamentals still fail
Our accessibility specialist, Martin Waite, shared findings from WebAIM's 2026 analysis of one million homepages. Despite growing awareness and better tooling, common problems like poor colour contrast and missing form labels remain widespread.
It's a reminder that tools alone don't solve accessibility; teams still need the knowledge and processes to act on what they find.
How AI is speeding up accessibility work
AI is making accessibility testing faster, and the tools keep improving. Many can now help identify missing form labels, colour contrast issues, invalid ARIA attributes (the extra bits of code that describe what a part of a webpage does, so screen readers can announce it correctly), generate alt text, and produce accessibility reports.
For teams building websites, that can mean less time spent finding obvious problems and more time improving the actual experience for users.
Accessibility works best when it starts early
A major theme of the webinar: accessibility should be part of everyday delivery, not an afterthought. Too often, teams only think about accessibility at the end of a project, rather than from day one.
Designers can now check accessibility inside Figma using a plugin called Stark, before development even begins, covering colour contrast, typography, touch target sizing, visual hierarchy and component accessibility. The tool can even simulate visual impairments, helping teams understand how a design might be experienced by real users before any code is written.
During development, tools like linting plugins, IDE (code editor) accessibility extensions, and automation tools such as Axe Core and Lighthouse CI can catch issues early. But keyboard navigation testing, screen reader testing, and testing with people who have access needs all remain essential; not every issue can be caught automatically.
Putting AI accessibility tools to the test
Across the team, we experimented with a range of AI-assisted workflows, from generating websites with AI ("vibe coding") to asking large language models (LLMs) to review pages for accessibility issues. We also compared AI-powered tools against established accessibility testing tools to see where each performed well and where they struggled.
Tools we explored included:
- AI-generated code using "vibe coding" tools
- General-purpose LLMs performing accessibility reviews
- WAVE
- Axe DevTools
- Chrome DevTools' AI accessibility panel
- VS Code Chat Agent
- Gemini CLI
- Zed IDE AI Agent
- Specialised accessibility AI agents and an accessibility wizard tool
What we found in practice
AI code generation can help, but it's not enough on its own. We tested an AI website builder called Lovable, and even with prompts asking for WCAG (Web Content Accessibility Guidelines, the internationally recognised accessibility standard) compliance, the team found serious accessibility issues. The AI still struggled with things like semantic HTML (using the correct code elements to describe what something is, a button coded as a button, a heading coded as a heading — so assistive technology understands the page properly) and with content that's simply harder to process for people with cognitive or learning disabilities.
Guy, our Senior Software Engineer, tested tools like WAVE and Axe DevTools on a deliberately flawed HTML document. They caught many issues, but also had clear limitations: they lacked semantic understanding, couldn't consistently interpret the context of an image, and sometimes gave descriptions that were inaccurate or unhelpful. In short: good at spotting patterns, not at understanding purpose.
For example, an AI tool can flag that an image is missing alt text. What it can't always do is explain what that image means. A graph showing a sharp rise in climate change might get labelled "graph" or "image", technically alt text, but not useful to someone using a screen reader. A meaningful description would explain what the chart shows, the trend it reveals, and why that matters. This is exactly where accessibility expertise and human oversight still matter most.
Saicharan, our Automation QA Developer, tested a specialised accessibility AI agent using Axe Core and Playwright, producing a report with scores, critical issues and practical fixes. He found this kind of automation saves real time but only if someone reviews the output carefully.
What we found
AI tools were often very good at identifying common issues, explaining accessibility rules, and suggesting fixes, which also helped developers newer to accessibility understand why something had been flagged. Established accessibility tools were generally more consistent at catching technical issues, while AI tools varied depending on the prompt, the model and the context given.
The biggest takeaway wasn't that one tool beat another. It was that each tool has its own strengths, and the best results came from using them together, with a person reviewing the final outcome.
So where does AI fit?
We think the right question isn't AI versus experts — it's AI plus experts.
AI can reduce repetitive work, highlight common issues, and help teams move faster. What it can't replace is critical thinking, and more importantly, empathy.
AI is good at: detection, pattern recognition, automation, drafting suggestions, speeding up audits.
Accessibility professionals are good at: understanding users, making inclusive design decisions, reviewing complex interactions, validation, and oversight.
Together, they make a workflow that's faster, more scalable, and more reliable than either approach alone.
Accessibility is ultimately about understanding how different people experience digital services, and that takes judgement, curiosity and conversation with real people. Those are qualities AI simply doesn't have.
Why this matters to us
At Hive IT, most of the organisations we work with charities and public sector teams are building services for people who can't afford to be locked out by a badly labelled button or an unreadable form. That's exactly why we treat accessibility as one of our core values, not a compliance checkbox. Testing these tools properly, and being honest about their limits, is part of how we live that out.
Our final thoughts
AI is here to stay, and it'll keep evolving: better testing, smarter code suggestions, more capable accessibility assistants. That's genuinely exciting. But technology alone won't create inclusive digital experiences.
We'll keep using AI where it genuinely improves how we work, spotting issues earlier, cutting repetitive tasks, freeing up time for the problems that really matter. And we'll keep investing just as much in our people.
Because accessibility isn't ultimately about passing automated tests or ticking a checklist. It's about creating digital services that work for everyone.
AI can help us get there faster. It's people who make sure we get there properly.
Want to talk through your own team's accessibility approach, or compare notes on the tools you're using? Get in touch with Hive IT; we'd love to swap what we've learned. Contact us here