Over the past year, it feels like almost every experience conversation comes back to AI. Across verticals and industries, everyone is looking to jump on this exploding trend and incorporate AI into their brand. However, as brands race each other to implement agentic AI, inclusive design falls by the wayside and people are being left behind.
This is not a new concept. From mobile experience and responsive design to e-commerce sites, accessibility has been a late consideration. Even now, 95%+ of homepages have detectable WCAG failures. We have carried accessibility debt through every major digital wave.
With agentic AI, the stakes are higher. AI is becoming the experience itself: the thing that surfaces content, interprets intent, and shapes the entire journey. When accessibility fails in that context, it is not as simple as a broken button or an unreadable page. It is being locked out of the experience entirely.
The Chatbot Failures We Know
Imagine you rely on a screen reader to navigate the web. You land on a site with an integrated AI assistant. You activate it, ask a question, and wait. The response loads, but your screen reader says nothing. The content exists visually, but your assistive technology cannot read it at all. This is not just a hypothetical. It is a routine reality for screen reader users interacting with AI chat interfaces that have not been built accessibly.
- Keyboard Navigation
Is your chat modal actually a trap? Conversational interfaces create focus traps for keyboard-only navigation. Users get stuck in the input field after getting a response. When the chat closes, they can't return focus to the page. And some can't reach the chatbot in the first place. All chatbots, AI or otherwise, must be accessible through keyboard-only navigation, with intuitive controls that make it easy to interact with. - Screen Reader Compatibility
Code needs to be correctly marked up with accessibility roles and ARIA implementations to allow screen readers to read the content within them. The dynamic nature of AI chatbots can pose difficulties with streaming responses where text appears word by word to form a response. This can make the output from the screen reader sound fragmented or nonsensical, creating confusion for the user. This issue is solvable through technical fixes, allowing the AI response to be announced in chunks rather than a word at a time. - Cognitive Accessibility
When it comes to AI responses, complex answers can hinder rather than help, particularly for users with dyslexia or ADHD. Walls of text or excessive jargon add significant cognitive load. But this is where AI's potential is most underused. An AI interface can adapt to the individual with plainer language, shorter responses, and summaries instead of dense prose. That adaptability could be one of the most powerful accessibility tools available, but it can only happen with deliberate design.
Integrated AI: The Harder Problem
Integrated or ambient AI is woven into the fabric of a digital experience rather than sitting as a distinct component. If we think about a product like Amazon's Rufus, the AI experience is not just a chatbot, but shapes the process of product discovery, search, and access to information. With this shift, accessibility issues become more widespread across the experience, making them harder to spot, attribute, and fix.
With a chatbot, an accessibility failure is contained and locatable. With integrated AI, the failure is architectural, and easy to miss with standard reviews. A page restructured by AI might not follow a logical reading order for screen reader users. An AI-generated product description may be at a reading level which excludes users with cognitive disabilities. AI search results might not support keyboard-only navigation.
These issues are grounded in AI being layered on top of foundations where accessibility was already overlooked. But that makes the solution clearer than it seems. Fixing the foundations means incorporating accessibility best practices, which are what AI needs to function at its best.
The Overlap: Why Good Accessibility Is Good AI
The foundations of good AI and good accessibility overlap heavily. Both require structure and clarity to function. Accessible content is clear and unambiguous. AI needs that same clarity to interpret content and shape the user experience correctly. Which means fixing accessibility doesn't just make experiences more inclusive, it makes your AI work better.
- Semantic HTML and Content Structure
The best written content cannot stand up to accessibility requirements if it is not structured and tagged correctly. Headings, paragraphs, and buttons must all be marked up to carry their meaning in the code. Semantic HTML markup is vital for screen readers to understand page structure and provide accurate output. Similarly, an integrated on-site AI is dependent on that same semantic markup to understand what the content on the page is and how to process it. When those foundations are weak, both users and AI lose the context they need. - Alt Text for Visual Content
AI cannot interpret images without the correct supporting data. The alt text that accessibility standards require for screen readers serves the same purpose for on-site AI: providing the context needed to understand an image. As image-driven product discovery grows within agentic commerce, getting this right is a commercial imperative, not just an accessibility requirement. - Structured Product Data
For integrated AI to make informed product recommendations, it must know the product details. Agentic commerce flows hinge on clear, structured product data. That's the same foundation assistive technologies need to interpret product information. Brands that have considered their accessibility and locked in strong foundations have set themselves up for success with agentic.
Breaking the Cycle of Accessibility Debt
As agentic storms ahead, accessibility needs to be part of the conversation in the early stages. New WCAG 3.0 guidelines are underway and explicitly include AI considerations to remove bias. Rather than waiting for the guidelines to finalise and treating them as a compliance target, there is an opportunity to lead. Brands that build accessibly today won't need to retrofit to meet future standards. Those that don't will face a far more difficult, resource-intensive challenge trying to adapt inaccessible foundations for AI-first experiences.
Meanwhile, digital experiences that ignore accessibility shut out a large share of potential users. According to the WHO, 1.3 billion people globally have a disability, 1 in 6 of the population. To create an experience that shuts out these users is to exclude a market almost the size of China.
All of this points toward brands needing to rethink their approach to accessibility, breaking the pattern of seeing it as a last-minute scramble for compliance. Instead, accessibility is an integral foundation that makes the AI built on top of it more effective. Good for people and good for business.
At Merkle, accessibility is not bolted on but baked in from the start. It is embedded in our design systems, our component libraries, and our approach to content, so that experiences have inclusive foundations by default. We believe that effective agentic experiences and inclusive experiences are built the same way, so we make sure our clients never have to choose between inclusivity and innovation.
