The agent assembled a partial but usable understanding of AI4Scholar by combining web search results with inaccessible homepage fetches. It successfully identified what the platform does, who it serves, and how it differs from competitors, but could not retrieve concrete pricing data or detailed feature documentation because the site relies on client-side rendering. The site's content was not directly machine-readable from static HTML; the agent recovered information almost entirely through web search discovery and prior knowledge.

• ›Steps [1], [3], [5], [7], [8], [11], [12], [13] all returned raw HTML with no rendered content — the Next.js frontend requires JavaScript execution. The agent could not extract answers from these fetches despite reaching 200 status codes.
• ›Web search steps [9], [15], [16], [18], [19] provided the only actionable content: search result titles and snippets mentioning 'AI-Powered Academic Research Platform', MCP integration, API capabilities, and comparisons to Semantic Scholar and Research Rabbit. The agent built its entire substantive answer from these search-returned snippets and inferred details.
• ›The agent cited six URLs in its final Sources section ([ai4scholar.net/], [pricing], [docs], [faq], [about], and GitHub/forum links), but none of those first-party URLs actually returned usable content in the trajectory. The citations were aspirational — pointing to where information *should* be — rather than grounded in what the agent retrieved.
• ›Pricing information remained completely unavailable: the agent explicitly called this out as a gap and acknowledged it could not fetch the /pricing page's rendered content. Web search also returned results for other tools (scholarai.io, scholarly.so) instead of ai4scholar's own pricing.
• ›The agent successfully inferred the product's positioning (multi-source search, nano-draw illustration, MCP/API integration, researcher-built) from scattered search snippets and GitHub plugin references, demonstrating reasonable synthesis of sparse signals. However, feature details like nano-draw capabilities, exact API endpoint count, and subscription mechanics remain speculative or incomplete.
• ›The site's Chinese-primary language and client-side rendering created a double friction: the agent could not execute JavaScript to render English content, and web search results sometimes returned Chinese docs (e.g., '平台介绍' links) that would also require rendering.