The agent assembled a substantive overview of Smart BTP's purpose, target market, and positioning by combining fragments from the site's homepage, pricing page, demo page, and blog/guide pages, along with competitive context from web search. The site proved difficult to navigate: key pages returned HTML but with truncated/SSR-obfuscated content; pricing information was explicitly custom-quote-only; and detailed feature breakdowns were either missing or inaccessible through the fetches. The agent worked around these gaps by inferring positioning from URL slugs, search results, and prior knowledge to deliver a coherent answer—but had to flag multiple gaps (pricing opacity, accuracy details, scope limits) as genuine product information holes, not just retrieval failures.

• ›Steps [1], [3], [6], [16], [19], [20] returned 200 status pages but with Next.js-rendered HTML that truncated in the response; the agent could not extract raw text content from these fetches, forcing reliance on prior knowledge (23% of data sourced) and web search (38%) to reconstruct the narrative.
• ›The pricing page ([3], [13]) returned a 200 but contained no visible pricing data—the site's explicit positioning ('Une solution adaptée à chaque entreprise BTP. Prenez rendez-vous pour un devis personnalisé') had to be inferred from search and prior knowledge rather than from the actual fetched HTML.
• ›The agent discovered Smart BTP's core use case (DCE/tender automation) primarily through URL semantics ([16] /comparatif-logiciel-appel-offre-btp, [19] /cctp-marche-public-btp, [20] /dpgf-decomposition-prix-global-forfaitaire) and web search results, not from readable homepage content—indicating the site publishes positioning via SEO/blog but not via machine-readable homepage structure.
• ›Competitive differentiation was sourced from web search ([11] KALAO blog, [12] AnalyseurDCE) rather than from Smart BTP's own comparison pages, suggesting the site lacks a native competitive positioning section or it was inaccessible due to rendering.
• ›The site's French-only language and custom-quote-only pricing model were stated facts the agent recovered, but customer testimonials, accuracy metrics, supported document types, and integration details were completely absent from all fetches—indicating genuine content gaps, not just retrieval friction.