Article illustration 1

Navigating online retail as a blind person is an exercise in frustration. As accessibility expert Léonie Watson describes, even ostensibly 'accessible' e-commerce sites fail to provide sufficient detail for confident purchases. Product descriptions like "Long sleeve Knitted jumper Crew neck... 100% Cotton" omit crucial details—texture, fit, length—forcing a cycle of ordering multiple items and returning most. While AI-generated image descriptions (via tools like ChatGPT or Claude) offer a partial solution, they remain cumbersome and unreliable.

"Crafted from soft pure merino wool, this jumper is a luxurious addition to your knitwear collection. It's crafted to a relaxed fit for endless layering options..."

Even this improved description, Watson notes, leaves ambiguity: What defines 'relaxed'? How deep is the V-neck? The core problem persists: retail sites prioritize visual information, leaving non-sighted users stranded.

Enter Agentic AI: Beyond Chatbots
Platforms like Innosearch signal a paradigm shift. Unlike passive chatbots, agentic AI possesses agency—it executes tasks. Innosearch's CoBrowse acts as a true digital shopping assistant:

  • Searches across 500,000 integrated stores using natural language queries (voice or text)
  • Filters results based on specific parameters (e.g., "exclude polyester," "show only petite sizes")
  • Extracts details trapped in images automatically
  • Manages cart interactions: "Add medium to cart," "Apply discount code SUMMER25," "Show my basket"

Crucially, it structures chaotic retail data into a consistent, accessible interface. Watson discovered a jumper was menswear only through Innosearch—a detail missing on the original site.

The Web's Impending Transformation
This evolution prompts a radical question: Why build traditional websites at all? If users can delegate tasks to AI agents, the pain points of complex navigation, inconsistent layouts, and inaccessible interfaces become obsolete. The data underscores the shift:

  • OpenAI reports 2.5 billion daily ChatGPT prompts (up from 1 billion in Dec 2024)
  • Google's AI search already reduces website traffic by up to 50% in some cases

As Watson observes, humans gravitate toward convenience. We embraced GUIs over command lines, touchscreens over mice—and now conversational agents minimize interaction effort entirely. This isn't about eliminating the web but redefining interaction: the agentic web prioritizes task completion over page rendering.

Accessibility at a Crossroads
While promising, this shift introduces new challenges:

  1. Hallucination Risks: Can unreliable AI descriptions meet legal accessibility standards?
  2. Testing Complexity: How do you audit dynamically generated, non-deterministic outputs?
  3. Exclusion: What happens when agents can't handle niche queries or complex tasks?

Watson highlights the urgent need for frameworks ensuring agentic systems provide reliably accessible experiences. As organizations pivot toward this model, the accessibility community must answer: How do we guarantee equity when every AI interaction is unique? The agentic web isn't coming—it's already rewriting the rules of digital engagement.

Source: Tetralogical - Accessibility and the Agentic Web