How AI can Support Accessibility

By June 2025, many companies will have to make their digital offerings accessible - but very few have any idea how complex the tests for this will be. Testers are currently juggling a patchwork of Word, Excel, screenshots, screen readers and open source tools, while over 80 WCAG criteria have to be checked. An integrated test environment combined with the targeted use of AI could help precisely where automation makes sense - for example, when checking whether headings semantically match the text or when intelligently clustering pages.

Podcast Episode: How AI can Support Accessibility

How can artificial intelligence (AI) improve accessibility testing? Valentin Dallmeier explains the challenge of extracting accessibility information from websites and how his team is using AI to leverage language models for better test results. One interesting aspect is the development of an integrated test environment (ITE), which simplifies the test process and makes it more efficient by bundling all the necessary tools in one application. However, data privacy must be ensured when using AI tools, which poses a considerable challenge.

“This has a dark mode. That’s a no-brainer.” - Valentin Dallmeier

Dr. Valentin Dallmeier studied computer science in Passau and Saarbrücken and completed his doctorate at the Chair of Software Engineering in Saarbrücken. In his doctoral thesis, he combined test case generation with model extraction techniques - an achievement that was awarded the Ernst Denert Prize in 2010 as the best dissertation in software engineering in the German-speaking world for its practical relevance. The techniques he researched led to the founding of the Testfabrik, on whose board he has been developing innovative tools for test automation since 2013.

Highlights der Episode

• Accessibility testing doesn’t need 10 tools - an integrated test environment saves media disruption and time.
• ChatGPT checks headline-to-text fits better than humans - language tasks are AI strength in accessibility.
• Crawler plus clustering automatically recognizes similar pages - saves manual selection of test representatives.
• OpenAI is still unstable - local language models like Llama do not yet work well enough for accessibility testing.
• Build iteratively and get feedback beats AI hype - understand the problem before building the tool.

Insights into AI-supported solutions for accessibility testing

This episode is all about accessibility testing and the use of artificial intelligence (AI) in this area. Valentin Dallmeier shares insights into his work and the development of an integrated testing environment to test and improve accessibility more effectively. An exciting behind-the-scenes look at how technology and AI can be used to make websites more accessible.

From research to practice

Valentin Dallmeier’s team has dedicated itself to the topic of accessibility. Their mission? To create technological solutions that help people test better. A bachelor thesis laid the foundation for their current work by investigating what accessibility information can be extracted from websites. This initiative paved the way for the development of an integrated testing environment that is not only more efficient, but also makes testing more enjoyable.

The role of AI in accessibility testing

Artificial intelligence plays a central role in Valentin’s team’s approach to improving accessibility testing. By analyzing the WCAG criteria, they identified areas where automation is possible - around a quarter of the 80+ criteria. Of particular note is the use of AI in linguistic tasks, such as checking the consistency of headings with text content. Their experiments with local language models show the potential for further automation options and data protection-friendly solutions.

An integrated test environment

The development of an integrated test environment (ITE) is at the heart of the project. This environment aims to unify the numerous tools and processes currently used in accessibility testing - from Word documents to Excel spreadsheets to screen readers. By bringing all these elements together in a single application, without media disruption, they not only make the testers’ work much easier, but also break down barriers in the testing process itself.

Feedback from the community: a decisive factor

An iterative approach that incorporates feedback from the community is crucial to the success of the project. Valentin’s team attaches great importance to working closely with users and incorporating their needs and challenges directly into the development process. This close collaboration ensures that the solutions developed are not only technically advanced, but also practically applicable and helpful in the testers’ everyday lives.

What’s next on the agenda?

Although impressive progress has already been made, Valentin still sees plenty of room for improvement and further development. The integration of further AI-based test scenarios and the optimization of existing tools are on her agenda. The question of data protection when using external AI services and how local alternatives could address this challenge is particularly exciting.