Accessibility in the age of AI

Posted on Monday, 13 July 2026 by Ela Gorla in Design and development, Strategy

Artificial Intelligence (AI) is now used by many teams, in many organisations. With its promise of improved efficiency and speed, it's easy to see why. But what does the widespread adoption of AI mean for accessibility?

While AI has the potential to help digital teams deliver more products faster, it can also have a negative impact on the accessibility and usability of those products if not used responsibly. This post explores some of the challenges posed by the use of AI, and ways organisations can ensure accessibility remains at the forefront of digital teams' minds.

A double-edged sword

AI seems to be a double-edged sword when it comes to accessibility. It can help teams prioritise accessibility but can also be the source of new accessibility issues.

Many of the tools used by digital teams now include AI-powered features that allow to easily and quickly complete accessibility tasks. Designers can check their designs comply with accessibility requirements, developers can use AI to help generate and review accessible code, the QA team can ask AI to run accessibility tests. This means accessibility can more easily become an integral part of teams' day-to-day work. Instead of being a separate task carried out by few people, it becomes a standard aspect of many people's role at all stages of development. Which is great news.

Why, then, does accessibility seem to be getting worse?

According to The WebAIM Million, a survey of one million homepages conducted by WebAIM every year:

The 2026 WebAIM Million analysis found notable increases in both the number of detected accessibility errors and number of pages with WCAG conformance failures, reversing a trend of gradual accessibility improvements in recent years.

It's likely that the rapid adoption of AI by digital teams has played a role in the above result. As with any new technology, until teams and organisations understand how to use it responsibly and safely, AI poses a risk of causing unexpected, and potentially serious, harm.

AI uses and accessibility challenges

From vibe coding to AI-assisted coding and AI agents, every use of AI comes with some challenges in terms of accessibility. Fast product development also introduces challenges of its own.

Vibe coding

Vibe coding makes it possible for anyone to build a functional digital product simply by feeding AI a plain-language prompt. This creates exciting new opportunities for people with no or little coding skills, such as designers, product owners, project managers, and so on.

However, the lack of coding skills makes it challenging to review the quality of the AI-generated code. This applies to accessibility, as well as other key aspects of digital products such as security. While this may not be an issue when vibe coding is used for prototyping purposes only, it can become a significant problem if vibe-coded products are deployed to production.

AI-assisted coding

Using AI as an assistant when writing code can greatly improve efficiency. AI is able to generate large amounts of code in just few seconds, and can provide advice on how to improve existing code. Some developers have therefore started using AI to help produce and review accessible code.

However, too often AI's output and advice about accessibility are inaccurate or simply wrong. When we asked generative AI help write accessible code, for example, we got many inaccurate and incomplete answers. Similarly, when we asked generative AI to write text descriptions, some models hallucinated and produced incorrect text descriptions.

Because LLMs are generally trained on publicly available websites and apps, many of which are inaccessible, they are likely to replicate the accessibility issues they've found in those websites and apps.

AI agents

Similarly, trusting AI agents when it comes to accessibility can be risky.

Many organisation have built their own AI agents to help with development, testing, and remediation work. This can greatly improve speed and efficiency across several projects.

In many cases, the AI agents are provided with accessibility-focused skills, to ensure they take accessibility requirements into consideration. While this is good practice, it can create a false sense of security that accessibility is fully accounted for. Teams may believe they no longer need accessibility expertise, which can be harmful in the long term.

Fast product development

The integration of AI in digital teams' workflows is driven by the desire to build products faster. However, faster product development can have an impact on the quality of products and, ultimately, on the experiences of people using them.

When AI is used to accelerate product development, key activities traditionally conducted at the beginning of projects start to disappear - discovery research, problem definition, solution design, wireframing, and so on. A team may move from just an idea to a working prototype in a matter of days. This is likely to result in a product or service that doesn't fully meet the needs and expectations of people.

Faster coding production also means that the QA team can no longer thoroughly test all code, increasing the risk of issues going unnoticed.

New technology, old issues

These are not new issues. Well before the advent of generative AI, organisations have tried to find cheap automated solutions that could replace human expertise and fix accessibility issues, and have always aimed to speed up processes often at the expenses of quality. AI is however amplifying these issues.

It's now vital that organisations acknowledge and address them, before the accessibility of their products is impacted.

Bringing AI and accessibility together

AI can play a role in helping deliver accessible products and services. However, it needs to be carefully led, managed, and overseen by people.

Your organisation needs to thoroughly consider how to make the best use of AI to support its long-term accessibility goals, where AI can bring the most value, and how digital teams should guide its work. AI cannot automatically solve your organisation's accessibility issues but, when used responsibly, it can be a useful ally.

Robust governance

A first step that your organisation should take is to ensure that robust governance around the use of AI and accessibility is in place. This means having well-defined long-term accessibility goals, rules around the responsible use of AI, and clear accountability.

Accessibility strategy

A long-term accessibility strategy will help teams within your organisation understand its priorities in terms of accessibility, and how they can contribute to achieving them. This is particularly important in times of rapid change; it keeps teams focused on what really matters.

Many organisations may have compliance with the Web Content Accessibility Guidelines (WCAG) 2.2 Level AA as one of their accessibility goals, for example.

Your organisation's accessibility strategy should be outlined in a high-level policy and a practical roadmap, and all teams impacted by it should be familiar with their contents. Some organisations may also decide to make it public to show their commitment to accessibility, like the BBC did with their Digital Product Accessibility Policy.

AI policy

Alongside a well-defined accessibility strategy, your organisation should have clear rules around how AI can and cannot be used to support accessibility tasks. For example, an organisation may decide that AI can be used to assist with testing and remediating accessibility issues but should never release fixes to production without human approval.

These rules could be outlined in a separate AI policy, referenced by the accessibility policy, or incorporated into the accessibility policy itself. Regardless of where they are placed, it's essential that everyone in your organisation is aware of them and understand their role in applying them.

Accountability

A key aspect of the responsible use of AI is to acknowledge that ultimately, AI is just a tool and cannot be held accountable for its actions and output. Organisations, teams, and people remain fully responsible for the accessibility of products and services.

Clear accountability helps ensure that everyone in your organisation follow the correct procedures, and that AI is used safely.

Value-led approach

With so many AI tools promising to take care of accessibility for you, it can be tempting to start using them without properly considering whether they can truly bring value to your organisation.

A value-led approach, rather than a tool-led approach, aims to identify where and how, within your organisation's processes and workflows, AI can have the best outcomes, before choosing the tools to use. It also focuses on monitoring the impact of using AI over time.

High-value opportunities

Start by reviewing your team's ways of working and look for tasks that could be safely automated. These could be writing accessible code for simple components, conducting basic accessibility tests, providing suggestions for remediation work, and so on. With guardrails and human oversight in place, integrating AI into some of these workflows can help improve efficiency and reduce accessibility backlogs, allowing the team to focus on more complex tasks.

A value-led approach will prove much more effective than using as many AI tools and capabilities as possible, without the proper planning and consideration. This may seem to improve speed and efficiency at first but it's likely to result in more retrofitting and remediation work in the long term.

Regular monitoring

With AI integrated into some of your team's workflows, it's important to regularly review its impact on the accessibility of your products.

As mentioned earlier, the last WebAIM Million survey has highlighted a decrease in the accessibility of many websites worldwide. Ensure the use of AI in your organisation is truly helping achieve its accessibility goals rather than producing unexpected negative results.

Regular assessments of your products can help evaluate any increase or decrease in the number of accessibility issues and their severity. The amount of retrofitting or remediation work needed is another useful measure when it comes to evaluating its effectiveness.

If negative outcomes outweigh the benefits of using AI, you may need to rethink how to make the best use of it. As with any new technology, finding the right balance may require a few iterations.

Usability testing

Compliance with technical standards like WCAG is only one aspect of accessibility. Truly accessible products also provide an enjoyable user experience for everyone. Conducting user research with people with a wide range of disabilities is the most effective way to ensure this is done.

With AI speeding up the product development process, it may be tempting to skip user research in favour of faster delivery. However, this is likely to harm your products in the long term. Instead, research teams need to find new ways to conduct meaningful research that fits the faster pace of delivery. Setting up regular testing days, focusing each study on a limited amount of content, and sharing findings in a more practical and efficient way are all useful strategies to deliver research faster.

AI itself can also be used to speed up some research tasks; for example, it can be used to draft moderation guides or to help write findings reports. However, it should not be used to replace testing with real people. We human beings are complex; we all have our own unique mix of expectations, needs, and preferences, and we all use technology and products differently, often in unexpected ways. Synthetic users can never capture the beautiful variety and richness of people.

Human guidance

AI is just a tool - the quality of its output depends on how well, or poorly it's used.

To get the best results, you need to feed it high-quality data, give it clear and accurate instructions, and consistently review its output.

Trustworthy sources

One of the most important aspects of using AI responsibly is ensuring it's trained on reliable sources; otherwise, it's likely to produce inaccurate or poor-quality results.

For example, if your team is planning to use AI to help with code generation, ensure the code AI is trained on is fully accessible, or you may end up with accessibility issues replicated from poor code examples. Similarly, if AI is going to be used to review content for accessibility or provide advice, ensure it draws on reliable sources such as the World Wide Web Consortium (W3C) Web Accessibility Initiative (WAI) website⁠, otherwise it may repeat inaccurate or misleading accessibility guidance.

If your organisation has an accessible design system, component library, or accessibility guidelines, these should also be fed into the AI to ensure consistency, as well as accessibility compliance.

Accurate instructions

Clear and detailed instructions, including prompts and skills, reduce the chances of AI producing unexpected results.

When it comes to accessibility, this means including detailed and accurate information on the standards the product needs to meet, any assistive technology and accessibility settings it needs to support, any customisation option it needs to offer, and usability requirements. Another important piece of information to include is when an AI agent should pause and request human review; for example, when it can't find a definitive answer in the training data it's been provided. This is critical to prevent the agent from hallucinating.

Instructions can be included in the prompts given to AI, as well as in the skills provided to AI agents. For better consistency and greater efficiency, your organisation may even decide to build a library of accessibility-focused prompts and skills that various teams can use and customise.

Output review

Using reliable sources and writing detailed instructions will reduce the chances of AI producing inaccurate or incorrect results. Nevertheless, reviewing its output remains a priority.

All teams within your organisation that use AI to help with the production of accessible products should build in accessibility checks at key points - not just at the end. They can rectify any issues with AI-generated designs or code early on and use that to iterate and improve on the process of producing design and code.

If necessary, role-specific training can be provided to fill in any gaps in knowledge or skills. This is especially important now, when vibe coding allows people with no coding skills to produce fully-coded products.

It's unlikely that your first attempt of using AI will deliver the desired outcome. You'll probably have to go through several iterations of instruction refinement to get the result you aim for. With every iteration, you'll learn where AI is more likely to fail, what types of instructions work better for it, and will also understand its strengths and limitations.

In turn, this can feed into the analysis of where, within your team's workflows, AI can be most useful, as discussed under High-value opportunities above.

Accessibility expertise

It's important to acknowledge that while AI can automate some accessibility-related tasks, it cannot replace accessibility specialists. Accessibility specialists, however, can help your team make the best use of AI.

For many of the activities described above, accessibility expertise is required. For example, specialised accessibility knowledge is needed when it comes to writing and refining accessibility-focused prompts and skills, reviewing AI-generated code (especially for complex components), selecting sources on the topic, and so on.

You may have expertise within your team or may have an in-house accessibility team you can reach out to. If not, you may need to request the assistance of external accessibility specialists.

AI and ethical concerns

The use of AI comes with a range of ethical concerns - from environmental impact to copyright and data protection. At the same time, if used to improve the accessibility of products and services, AI has the potential to improve the lives of many people.

Following the advice in this post will not only help your organisation achieve good results around accessibility, it will also help it use AI responsibly and safely. Ensuring that human oversight is in place at all times, and that AI is only used when it can really adds value, are two examples of a responsible and safe use of AI.

Next steps

To learn more about how to implement sustainable accessibility, you can read our 3-part series Sustainable accessibility in complex organisations. Or head to our Services to find out how can can help with Training Courses and Agile Usability Testing.