The agent assembled a reasonably complete profile of JD Williams by combining prior knowledge (50%) with web searches (13%) and a single successful fetch of their credit product page. The site itself was poorly navigable—404s on help/about sections and a client-side React app that obscured content—forcing the agent to rely heavily on external sources (GOV.UK company filings, third-party reviews, IBM case studies, Thinkbox) rather than the retailer's own published information. Despite these friction points, the agent delivered a substantive answer about their positioning (45+ demographic, inclusive sizing), payment model (JDW Pay credit system), and competitive differentiation.

• ›Step [1] (homepage) returned only boilerplate HTML with no semantic content; the site's dynamic client-side rendering meant the agent could not extract product range, pricing, or positioning from raw fetches without JavaScript execution.
• ›Steps [11], [13], [14], [17] (web searches) were the primary content sources, yielding company filings (N Brown Group structure), YouGov data on plus-size offerings, pricing details (JDW Pay terms), and competitive positioning—none of which were accessible by crawling the retailer's own domain.
• ›Step [16] (the /spread-the-cost page) loaded successfully but returned only HTML boilerplate; the agent could not extract actual credit terms from it and had to cite external sources instead.
• ›The agent had to guess or route through 404s for standard information architecture (help, about-us, customer-services, delivery-info, payment-options, categories)—indicating either a broken site structure or aggressive client-side routing that breaks static discovery.
• ›50% of the final answer relied on prior knowledge (company size, revenue figures, brand positioning) rather than current site state, suggesting either outdated information or information not published by the retailer itself.