AI is moving fast – faster than standards, regulation, and often faster than organisations can keep up with. It is now an integral part of everyday life, powering chatbots, generating content, summarising search results, even generating code. It brings with it great promise, but also pitfalls, particularly for people with disabilities.
Why? Because, as we often see with a technological shift, people with disabilities are at risk of being excluded if their needs aren’t designed into it from the start. So, how can organisations make sure they avoid excluding millions of people as the AI revolution continues?
While many commentators focus on whether AI will change the world, what we are asking is whether it will change the world for everyone. Because, unfortunately, while AI promises progress, without considering accessibility for people with disabilities and older people, it could deepen the digital divide.
This is something we discussed in a recent webinar, where we took a closer look at some of the most recent developments in AI, and the opportunities, the risks, and the practical steps organisations can take to make sure their AI tools meet the needs of those with accessibility requirements.
Here are our top five takeaways:
1. Be aware that AI can help, but it can also harm
AI can be a lifeline for many people with disabilities: simplifying tasks, generating captions and summarising complex content. But it can also hallucinate, mislead, and amplify bias, which, for people with disabilities, can be extremely detrimental.
For example, AI often fails at some of the most basic prompts, such as showing a specific time on a generated analogue clock. So, it is important to always validate AI outputs. Don’t assume they’re correct – especially for critical or disability-related content.
2. Make sure you train AI on diverse data
Most AI models are trained on narrow datasets, which means they often ignore minority needs. If your AI doesn’t know what a wheelchair user looks like, or how someone with dyslexia phrases a question, it can’t serve them.
So, if you are procuring AI, make sure you ask vendors: how was it trained? Or, if you’re building AI, ensure disability representation is included in your training data, or your AI will fail for millions of potential users.
3. Understand that accessibility boosts discoverability (in AI and SEO)
Being discoverable in AI summaries is becoming increasingly important for companies, but did you know that accessibility can actually help your brand appear in AI summaries and search results?
Things like missing alt-text and poor structure can make your content harder to analyse and process by AI tools. So, using proper headings, making sure alt-text is included on images, using descriptive links and readable content, will make your website more visible to AI and search engines. Just ask CoPilot or Gemini, if you don’t believe me.
Increasingly, accessibility = visibility.
4. Ensure transparency to build trust
Chatbots are now part of everyday life and can be a great tool for answering simple customer questions without human intervention.
However, for people with additional accessibility needs, we know that that trust can be a big issue, so it is important that users know when they’re talking to AI, not a human. For example, calling a chatbot by a ‘human’ name may seem like a good way to build trust, but if you haven’t been honest that it is AI, this erodes trust, especially for vulnerable users.
So, be upfront when you use AI, and ensure people know how your content and responses have been created.
5. Keep humans in the loop
Linked to this is the ability for AI to recognise when human intervention is needed. For example, if you are implementing a chatbot or AI agent, have you thought about what would happen if someone is dyslexic and may struggle with spelling or phrasing? Does it immediately do a ‘computer says no’?
Your AI needs to be able to adapt to what people are asking, and how they are asking it, for example with responses such as ‘I think you’re trying to say this’, ‘can you try rephrasing?’, or ‘can I just confirm?’
However, for many people with disabilities, the option to talk to a human being is non-negotiable in good customer service. So, you need to decide how many failed attempts for an AI to understand the user should trigger a handover to a live agent, and make that option clear. AI can assist, but human oversight prevents harmful errors.
Making AI work for everyone
Accessibility in AI isn’t a nice to have, it’s a must have, as there is both a regulatory and business case for building accessibility into your AI. Under the EU AI Act, AI systems must be accessible to people with disabilities, and have been trained to ensure they prevent negative impact to vulnerable groups. And beyond compliance, inclusive AI earns trust, loyalty, and a positive reputation.
AI is a tool which isn’t going away. Our job is to make sure it works for everyone.
