All insights

Public Sector6 October 20261 min read

Accessibility is not a phase

Public sector websites and apps must meet accessibility regulations. With AI-assisted testing there is no longer any excuse for treating accessibility as a remediation project.

Around one in four people in the UK has a disability. Many more use services on old phones, slow connections or in a second language. For public services, accessibility is not a feature; it is a legal duty under the Public Sector Bodies Accessibility Regulations 2018 and a basic test of whether a service works.

Yet accessibility is still too often treated as a phase: build the service, commission an audit, then fix the findings, usually under time pressure and at extra cost.

Build it in from the first commit

At Sukarma, accessibility is part of the definition of done for every change:

  • WCAG 2.2 AA as the baseline, including the newer criteria on focus visibility, target size, dragging alternatives and accessible authentication.
  • Automated checks on every change, run by our tester agents alongside unit and end-to-end tests.
  • Server-rendered pages that work without JavaScript for core journeys, which helps screen readers, low-end devices and poor connections alike.
  • Plain English content, checked for reading age, with clear headings and link text.
  • Manual testing with assistive technology, because automated tools catch only part of the problem.

Where AI helps, and where it doesn't

AI agents are excellent at catching the mechanical failures: missing labels, poor contrast, broken heading order, unannounced errors. They can also draft alt text and plain-English rewrites for a human to approve.

What they cannot do is tell you whether a disabled person can actually complete your journey with confidence. That requires testing with real users, and no amount of automation replaces it.

Bilingual by design

For services in Wales, the same thinking applies to the Welsh language. Designing text, layout and data structures for two languages from the start is far cheaper than retrofitting a translation layer later, and it produces a better service in both.

Keep reading

Related insights