The agent successfully assembled a comprehensive explanation of Zhipu AI's business model, pricing, and competitive positioning, but had to work around an inaccessible primary domain (zhipuai.com) and rely entirely on web search and prior knowledge rather than primary site content. The site itself was not directly navigable; the agent reconstructed the answer from third-party news articles, benchmarks, and pricing aggregators.

• ›The primary domain zhipuai.com failed to load (steps [1–2]), forcing the agent to pivot to web search. The actual platform operates under zhipuai.cn, but even that fetch (step [4]) returned only a Next.js template without substantive content—no pricing, product details, or positioning were machine-readable from the official site.
• ›All substantive answers came from web search results (steps [3], [5], [6]): news articles about GLM-5 pricing increases, Medium comparisons with Claude/OpenAI, GitHub API documentation, and third-party pricing aggregators. The agent cited 6 external sources and zero pages from zhipuai.com or zhipuai.cn itself, indicating the site publishes almost no discoverable product information.
• ›The agent had to infer or synthesize positioning, use cases, and technical capabilities from fragmented external sources rather than from a cohesive site narrative. Pricing tiers came from a third-party pricing tracker (UsagePricing), not official documentation. The comparison table and performance claims were assembled from news articles and benchmarks, not product pages.