The agent successfully assembled a comprehensive profile of Amity Web by combining direct site exploration (homepage, services, about, portfolio, contact, FAQs, hosting pages) with web search results and LinkedIn data. The site itself provided structural navigation but lacked explicit pricing pages, detailed service descriptions, and case study content, forcing the agent to work around poor information architecture and rely heavily on search results and external sources (Trustindex reviews, LinkedIn, company directories) to fill gaps and validate claims.

• ›Steps [1–3, 5–6, 15–16] returned Next.js-rendered pages with truncated HTML bodies, making detailed content extraction difficult; the agent could infer section existence (services, about, portfolio, hosting) but not read specific claims or pricing directly from the site.
• ›Pricing information was entirely absent from the site (step [4] returned 404 for /pricing); the agent reconstructed partial pricing (£2,000 minimum, 50/50 payment split, monthly hosting) by synthesizing fragments from search results and prior knowledge, not from official documentation.
• ›Trustindex reviews (step [11]) and LinkedIn (step [21]) became primary content sources for differentiators (longevity, consultative approach, crisis response) because the website's own pages didn't surface customer testimonials or competitive positioning clearly.
• ›Search queries (steps [8–14, 18–19]) accounted for 50% of fetch sources and were essential to discovering FAQs, hosting details, and business background; the site's own navigation and structure did not lead the agent to these resources.
• ›The agent had to note what was *not* findable (lack of portfolio case studies, unclear support tiers, geographic limitations) as confusions, indicating the site publishes positioning but withholds concrete details that would aid discovery and decision-making.
• ›No structured metadata (schema.org, llms.txt, robots.txt hints) was present; the agent operated largely on page-title inference and search engine visibility rather than machine-readable documentation.