The agent successfully understood Clerk's core offering, pricing model, and competitive differentiation, but relied almost entirely on web search results and prior knowledge rather than Clerk's own site content. The site itself was largely untraversable via direct HTML fetching—the agent could not extract meaningful structured information from the homepage, pricing page, or documentation, forcing it to reconstruct the answer from third-party comparison articles and pre-existing knowledge.

• ›Steps [1], [3], [8], [9] returned only truncated HTML boilerplate with no extractable text content—the site is a heavy Next.js SPA that does not render meaningful content in static fetch responses, making it opaque to the agent.
• ›The agent's actual sources were web search results [6], [11], [12], which surfaced third-party comparison guides (SuperTokens, DevToolReviews, AppStackBuilder) rather than Clerk's own marketing or docs. The agent cited these third-party sources in its final response, not Clerk's own pages.
• ›Step [4] returned a 404 for /features, indicating no traditional features page exists on the site, further limiting direct discovery of what Clerk does.
• ›The agent's comprehensive answer on pricing (MRUs, Free tier at 50k, Pro at $100/month, 2026 changes) came from search snippets and prior knowledge, not from fetching clerk.com/pricing directly, suggesting Clerk's pricing page is either not crawlable or not machine-readable in the way the agent expected.
• ›The agent explicitly noted friction: 'the site's complexity… is a heavy Next.js app that doesn't fully load in plain HTML fetching' and 'I couldn't directly see their homepage messaging,' acknowledging the site failed to surface its own value proposition.