The agent successfully gathered enough information to explain what Intercom does, who it's for, how it's priced, and how it differs from alternatives—but relied heavily on prior knowledge and external sources rather than on-site content. The homepage, platform, pricing, and resources pages were fetched, but the HTML responses were truncated and largely inaccessible; the agent compensated by conducting three web searches that returned structured competitor comparisons and pricing breakdowns from third-party review sites. The Intercom website itself proved difficult to navigate machine-readably, forcing the agent to assemble its answer from fragments across multiple external sources and inferred understanding.

• ›Steps [0], [1], [6], and [10] were fetched but their HTML was heavily truncated (max 4400 characters) with no structured data visible—the agent could not extract detailed pricing, feature matrices, or product explanations from the raw responses.
• ›The agent's actual content came from web searches [5], [7], [8], and [9], which returned third-party reviews, pricing breakdowns, and competitor analyses—not from Intercom's own site. The agent cited 8 external sources in its final response, only 2 of which were intercom.com URLs.
• ›The site is built on Next.js with a modern SPA architecture, making it opaque to direct HTML scraping. Key information (feature tiers, pricing stacks, use-case positioning) exists only in rendered JavaScript, not in crawlable markup or structured metadata.
• ›The agent's most useful findings came from prior knowledge (~67% of fetches) and inferential reasoning rather than direct site content—pricing models, AI agent capabilities, and competitive positioning were synthesized from multiple external sources and cross-referenced against partial on-site context.
• ›Attempts to fetch /products and /how-it-works returned 404 errors, indicating the site's navigation structure does not expose detailed product/feature pages at standard URLs; the agent had to pivot to /platform and /resources as fallbacks.
• ›The agent correctly identified critical friction points: the stacked pricing model (seats + Fin AI per-resolution charges), lack of a free tier, and ambiguity around 'resolution' vs. 'conversation' metrics—but these insights came from third-party reviews, not from the site's own pricing page.