The agent gathered a comprehensive explanation of Pangram's business model, pricing, and competitive positioning, but had to rely heavily on external sources (web search results and Wikipedia) rather than the site's own published content. The site's JavaScript-heavy rendering made direct HTML parsing ineffective, forcing the agent to assemble most factual claims from third-party reviews, news articles, and search snippets rather than authoritative pages on pangram.com itself.

• ›The homepage and pricing pages returned valid 200 responses but were heavily JavaScript-rendered with minimal extractable text in raw HTML—the agent received truncated payloads that contained no meaningful content about features, pricing tiers, or company background.
• ›Six content-bearing steps were all web searches or external URLs (search results, Wikipedia, Tom's Guide, AI Insider news, research aggregators)—zero authoritative information came directly from pangram.com's own pages despite 4 direct fetches to the domain.
• ›The agent explicitly cited external sources in its final answer (Phrasly review, eesel.ai pricing blog, Tom's Guide, theaiinsider.tech, Wikipedia) because those sources contained concrete pricing numbers, accuracy metrics, and company history that the main site did not expose in accessible form.
• ›The site failed basic discoverability for an AI agent: /about returned 404, /features returned 404, and the homepage and /pricing pages were not machine-readable—forcing reliance on third-party aggregators and news coverage rather than primary sources.
• ›The most detailed information (Pangram 4 specs, 0.0041% false-positive rate, August 2026 release, $9M Series A, founder names) came from AI Insider and research aggregators citing press releases, not from pangram.com directly.