The integration of artificial intelligence into web development has sparked a significant debate regarding digital inclusivity and the future of web accessibility. While generative AI and large language models (LLMs) offer unprecedented capabilities for automating content creation and code generation, they also present substantial risks for users with disabilities. During the recent DECODE 2024 developer conference hosted by WP Engine, industry experts explored the intersection of these emerging technologies and the persistent challenge of making the internet navigable for all users. Amber Hinds, CEO of Equalize Digital and a prominent voice in the accessibility advocacy space, highlighted that while AI holds promise, it currently functions as a double-edged sword that requires rigorous human oversight.
The core tension lies in the quality of the training data used to build modern AI models. Because these systems are trained on vast swaths of the existing web—a digital environment where, according to the annual WebAIM Million report, 96% of the top one million homepages contain detectable accessibility failures—the models inherently learn to replicate inaccessible code patterns. This phenomenon, often referred to as "garbage in, garbage out," means that developers relying on tools like GitHub Copilot or other AI coding assistants risk deploying interfaces that do not comply with the Web Content Accessibility Guidelines (WCAG).
Semantic HTML and the Risk of Automation
One of the most frequent technical errors identified in AI-generated code is the improper use of semantic HTML. Semantic elements, such as the <button> tag, are essential for screen reader users because they communicate the purpose and functionality of an element to the browser and assistive technologies. A <button> element is natively keyboard-accessible, meaning users can trigger actions using the "Enter" or "Space" keys.
In contrast, developers often use generic <div> or <span> elements, which lack inherent semantic meaning and keyboard focusability. While AI models frequently suggest these non-semantic structures—likely because they appear with high frequency in the flawed code they were trained on—they create significant barriers for users who rely on keyboards rather than mice. The reliance on such code represents a regression in web standards. As Hinds noted during her keynote, the danger is that developers may accept AI suggestions without the necessary technical expertise to verify whether the output adheres to accessibility best practices.
The Complexity of Automated Alternative Text
Alternative text (alt text) for images is another area where AI is currently being deployed with mixed results. While several WordPress plugins now offer automated alt text generation, the quality of these descriptions remains inconsistent. Effective alt text is highly contextual; it must convey the meaning of an image based on its specific role within a document.
Current AI tools often struggle with nuance. For instance, an AI might describe an image of a person as "a woman," whereas a human-authored description might identify the individual by name and describe their specific action or relevance to the content. Furthermore, AI often fails to accurately transcribe text embedded within images, such as marketing slogans on clothing. This deficiency poses a real-world problem for users with visual impairments who rely on screen readers to interpret product images to make informed purchasing decisions. While AI can provide a "head start" by generating initial descriptions, the consensus among experts is that human review remains an essential final step.
Chronology of AI in Accessibility
The discourse around AI and accessibility has accelerated rapidly over the last 18 months, coinciding with the mainstream adoption of ChatGPT and similar generative models.
- Early 2023: The launch of advanced LLMs triggered a surge in interest regarding AI’s ability to "fix" web accessibility. Early marketing claims often mirrored the misleading promises of accessibility overlays—tools that claimed to automatically make a site compliant without manual code remediation.
- Late 2023: The accessibility community, led by organizations like WebAIM and advocates like Hinds, began publishing research and hosting workshops to caution against the "quick fix" mentality. This period marked a shift toward understanding that AI should be viewed as an assistive tool for developers, not a substitute for human-led accessibility audits.
- February 2024: During the DECODE 2024 conference, the focus shifted to the potential for AI to act as a supportive agent in complex user tasks, rather than just a code generator.
- Present Day: The industry is now navigating the legal and ethical implications of AI agents, particularly regarding business accountability when AI systems provide incorrect information to customers.
Legal and Ethical Implications
The question of accountability is increasingly being tested in the courts. A recent legal precedent suggests that companies are legally responsible for the actions of their AI agents. If a customer-facing chatbot provides inaccurate information regarding terms of service or refund policies, businesses cannot simply blame the AI’s "hallucination." As the technology becomes more prevalent, companies must adopt a policy of transparency, ensuring users are aware when they are interacting with an AI rather than a human representative.
From an ethical standpoint, the inclusion of users with disabilities in the feedback loop is non-negotiable. Many businesses currently roll out AI tools based on performance metrics that track efficiency but ignore the usability experiences of the 15-20% of the population living with some form of disability. Failing to test these tools with the disability community results in "solutions" that may improve speed for some users while completely excluding others.
Future Potential: From Automation to Assistance
Despite the current challenges, there is a legitimate case for optimism regarding the future of AI in web accessibility. Emerging research, such as projects currently being explored at Ohio State University, focuses on using AI agents to assist users with complex, multi-step digital processes. For a person with a motor impairment or a cognitive disability, navigating the multiple stages of booking an international flight or a conference registration can be exhausting. An AI agent that could handle the form-filling and data management for a user could be transformative.
This vision aligns with the universal design principle: when the web is made more accessible for people with disabilities, the user experience improves for everyone. Features that simplify navigation, provide text-based summaries of complex PDFs, or offer voice-controlled interactions are not just "accessibility features"—they are superior user experience features that benefit all web users.
Moving Forward with Best Practices
For organizations and developers looking to integrate AI into their WordPress workflows, the following practices are recommended:
- Prioritize Transparency: Clearly disclose the use of AI in any customer-facing capacity, whether it is a chatbot, automated content generation, or image alt-text.
- Maintain Human Accountability: Do not view AI as an autonomous decision-maker. Every piece of code or content generated by AI must undergo a manual review process to ensure it meets accessibility standards.
- Engage the Community: Accessibility is not a check-box exercise. Developers should actively seek input from people with disabilities to understand how AI-driven interfaces impact their daily lives.
- Use Specialized Testing Tools: Rather than relying on AI to "fix" sites, developers should use dedicated testing tools, such as the Accessibility Checker plugin, to identify common errors like missing headings or non-semantic HTML during the development phase.
As the industry moves into the next year, the goal remains clear: AI should be leveraged to remove barriers, not to automate the creation of new ones. By maintaining a focus on semantic standards, human oversight, and inclusive design, the WordPress ecosystem can lead the way in ensuring that the future of the internet is one that is accessible to all, regardless of ability. The transition from early-stage experimental use to mature, accessible integration will require a culture shift that values digital equity as much as technical innovation.
