The agent successfully gathered comprehensive information about Exa.ai by fetching the homepage, pricing page, docs, and about/blog sections, then supplemented with web search results and third-party explainers. The site itself was navigable and direct for basics (what it does, who it's for), but the agent had to rely heavily on external sources (fastCRW blog, comparison pages, Swellpulse) to construct detailed pricing tiers, competitive differentiation, and use-case clarity—suggesting Exa's own pages contain the information but not always in a single, machine-readable artifact.

• ›Steps [1], [2], [3], [4] returned valid HTML but were truncated in the trajectory; the agent cited [1], [2], [4] as sources, indicating it extracted core positioning, pricing structure, and about/mission info from these pages directly.
• ›Step [10] (fastCRW pricing explainer) and step [11] (Exa's vs Perplexity page) were critical for answering the pricing breakdown and differentiation questions—Exa's own site had this info but the agent relied on third-party synthesis and its own comparison page to make it clear and structured.
• ›The agent noted that step [11] (versus/perplexity) was 'mostly JavaScript-rendered,' indicating the site's comparison content may not have been fully extractable from the raw HTML—a friction point for machine readability even though the site published the comparison.
• ›Pricing specifics (per-search-type costs, free tier, enterprise options) were well-documented on the pricing page [2] and cited in the final answer, but the agent had to search externally [8] and fetch [10] to confirm March 2026 pricing changes and construct realistic cost scenarios.
• ›The agent successfully assembled a coherent answer on what Exa does (neural search API for AI), who uses it (AI engineers, startups, Fortune 500 like HubSpot), how it's priced (pay-as-you-go, $7–15 per 1k queries), and how it differs (semantic search, LLM-optimized, massive domain filtering)—all tasks satisfied, but gaps remained in rate limits, search type tradeoffs, and data freshness specifics, which the agent flagged as not fully discoverable.
• ›The site surfaces information hierarchically (homepage → pricing page → docs → blog/comparison pages) and provides direct comparison pages, but lacks a single unified resource document; the agent had to navigate multiple layers and resort to external sources to build a complete picture.