Back in February 2025, I attended axe-con for the second year running. If you haven’t heard of it, axe-con is the world’s largest digital accessibility conference - you can find out more about it and read my highlights from last year’s blog post.
It’s now been two months since the conference, and there are two topics, both of which were major themes in the talks I watched, which seem to be getting hotter and hotter every day: the European Accessibility Act (EAA) and Artificial Intelligence (AI).
The European Accessibility Act
The EAA comes into effect on 28th June 2025, and will apply to any company that operates in the EU. It is focused specifically on improving the accessibility of digital products and services, primarily in the commercial sector, and includes detailed technical requirements that these products and services must follow.
Although the EAA will not become part of UK law (thanks, Brexit…), it does affect any business that provides goods and services to consumers in the EU. Therefore many businesses in the UK (and across the rest of the world) will need to comply with it to avoid fines and damage to their reputation.
This isn’t the first legal act that affects digital accessibility - one of the biggest and most well known is the Americans with Disabilities Act (ADA), which was introduced back in 1990. Although the ADA is broader than the EAA (covering things like physical spaces, not just digital services), and doesn’t give as many specific technical requirements, it is still highly relevant for making the web an accessible place; according to a blog post on Usable Net, over 4,000 ADA lawsuits related to digital properties were filed in 2024.
These two pieces of legislation can also have a positive impact on the accessibility of websites outside of the USA and the EU - as noted in one of the keynote talks at axe-con 2025, a company that operates in one of these markets isn’t going to maintain both an accessible and an inaccessible version of the same codebase, meaning accessibility improvements should be rolled out across websites worldwide.
So, given that the EAA compliance deadline is less than two months away, there’s currently a big drive to make websites accessible - or at least, to make them meet certain accessibility criteria like WCAG (Web Content Accessibility Guidelines) 2.1 or 2.2 AA - but more on that later. Although the focus at the moment is on that June date, It’s important to note that adherence to EAA needs to be continuous - websites are updated all the time, and each update, whether that be in the code or the content, needs to be done in such a way that the website is still accessible.
So how do we ensure our websites are accessible now and after dozens, hundreds, or even thousands of updates? Accessibility testing is key to this.
A segue into automated accessibility testing
Accessibility testing involves a combination of methods to ensure a website can be used by people with disabilities. These methods include automated testing tools that scan for common issues, manual testing by individuals using assistive technologies, and user testing with people who have disabilities.
Manual and user testing are both essential to ensuring a website is accessible, and their importance can’t be overstated. Despite being less accurate and complete than these more manual methods, automated testing can provide very fast results for improving accessibility. Incorporating automated tests into places like your design system or CI workflow can help to flag existing issues and prevent new ones being created.
Automated tests are decent at picking up on simple accessibility issues like missing heading levels or orphaned form labels, but they can’t be relied on for more complex or nuanced issues like colour contrast ratios or how appropriate alt text is. When I attended axe-con last year, I remember hearing a statistic that 57% of WCAG 2.x issues could be found with axe-core automated testing, which leaves 43% of issues that need to be found manually.
This is one area where artificial intelligence could make a huge difference to digital accessibility.
Artificial Intelligence
Several of the talks at axe-con this year gave details on how the coverage of Deque’s accessibility tests can be increased to over 80% with the help of AI; advanced rules, AI-assisted testing for Intelligent Guided Tests (IGTs), accessible coding with AI agents, and Axe Assistant for Slack and Teams are all slated to be added to axe in 2025. Deque have a blog post that discusses each of these new features.
Of these, I was most excited to hear about the capabilities of the new axe assistant in the talks 'The future of the axe platform' and 'Axe innovations: AI in next-gen accessibility tools'. This is an AI agent that can help with writing code while keeping accessibility at the forefront - it can answer questions, explain defects and suggest fixes to accessibility issues, as it has been trained on Deque’s content. Helpfully, it cites its sources for transparency.
Although I haven’t tried it out yet, it sounds like it could be a very powerful tool for finding and fixing accessibility issues and for ensuring that you are writing accessible code from the start. As a team, we have recently been experimenting with a few different AI coding agents, such as Cursor, and so far they have proved to be quite impressive on the whole.
One of the biggest ways that AI-based tools could help with improving accessibility is around content, such as poor or non-descriptive alt text, and use of overly complex language. In my experience, AI tools tend to do well at language-based problems, so these kinds of issues could potentially be easy to fix with AI. The difficulty may be in detecting the issues - although AI is already capable of generating an image description or re-writing a block of text to use simpler language, it may not currently be able to take into account all the context related to an image to decide whether it needs better alt text.
Despite all of the ways that AI could positively influence digital accessibility, is there a chance that it could also have a negative impact? The use of AI for generating code is likely to increase a lot in web development across the board in the coming months and years. However, given how much inaccessible code is out there in the wild, surely AI models have been trained on code with accessibility problems, so it’s not impossible that it will generate code that also has accessibility problems. We need to be very careful about this, and always properly review and test AI-generated code.
Accessibility ≠ Usability
In the current whirlwind of talk about EAA and AI, we shouldn’t forget that digital accessibility is about more than checking the boxes of WCAG and getting your site to pass all the fancy new AI-powered accessibility tests - it’s about usability for real people. Your site could sail through all the automated tests you throw at it and be compliant under EAA, but things like a poorly thought out navigation order could make the experience miserable for a disabled user.
My favourite talk at axe-con 2025, titled 'The Usability Gap: Why Accessibility does not always mean usable' highlighted this particularly well: “When organisations treat WCAG compliance as the end goal, they will miss critical usability issues. And if these same organisations are not involving real users with disabilities in their testing, they may never discover them.“
I hope that in the near future, websites and applications around the world will have fixed all of their technical accessibility issues (with a push from EAA and AI), and so designers and developers can give more attention to crafting experiences that are more usable for everyone.