The agent successfully assembled a comprehensive business profile of Practical AI by relying almost entirely on web search results, since the target website itself returned only header content. The site proved poorly navigable for automated retrieval, forcing the agent to reconstruct the answer from external search snippets and sources.

• ›The target website (practical-ai-for-business.co.uk) failed to surface its own content via direct fetches—all four attempts [0,1,3,4] returned only the header 'Practical AI · Hands-on AI training for senior teams' with no details about pricing, methodology, or differentiation. The agent had to abandon the site as a source.
• ›Web search step [2] and supplementary search step [5] were the only sources that actually yielded substantive, domain-specific facts: pricing (£3,500–£7,500), service types (half-day/full-day intensives), tools used (Claude, ChatGPT), and founder background (Ollie Ayton, London Business School, 350+ executives coached).
• ›The agent correctly surfaced that critical information—client names, pricing, methodology, and differentiation—came from search results, not the target site itself. It cited two sources in its final response, both external: the homepage URL (which provided no detail when fetched) and Ollie Ayton's LinkedIn profile (which was discovered via search, not navigated from the site).
• ›No internal site navigation was discovered. The agent never found a pricing page, about page, services breakdown, testimonials, or contact form—all standard elements for a B2B services website. These might exist but are not discoverable or crawlable.
• ›The agent filled knowledge gaps (e.g., 'what makes it different from alternatives') with cross-domain reasoning about AI training market differentiation from search results [5], rather than site-provided competitive analysis. This represents model inference rather than site content.