The agent assembled a comprehensive overview of Goldman Sachs by combining sparse on-site content with web search and prior knowledge. The website itself was poorly navigable for this task—multiple intuitive URLs returned 404s, business descriptions were fragmented across pages, and pricing/differentiation were absent. The agent worked around these gaps by using search to reach SEC filings, competitor comparisons, and third-party sources, ultimately delivering a credible answer that was not fully grounded in the site's own published structure.

• ›The homepage [1] and Careers [3] pages were fetched but returned truncated HTML that did not visibly contain substantive business descriptions, forcing the agent to rely on search and prior knowledge for core business segment information.
• ›Multiple logical navigation URLs (/what-we-do, /about, /services, /businesses, /our-firm) returned 404s [2, 4, 6, 7, 8], indicating poor information architecture for a site tasked with explaining what Goldman Sachs does. The agent had to guess or search rather than follow the site's own structure.
• ›The Investor Relations page [9] and 2025 Annual Report [12] were fetched but their full content was truncated in the responses; the agent cited the Annual Report URL in its sources but could not demonstrate extracting specific numbers or details from it.
• ›Web searches [10, 11, 13, 14, 15] became the primary discovery method for business segments, fee structures, competitors, and Marcus offerings—information that should have been native to the corporate site but was instead pieced together from external sources (SEC filings, third-party reviews, industry analysts).
• ›The final response explicitly flags that pricing is opaque and that Marcus is 'hidden' on the corporate site, suggesting the agent recognized the site's poor self-disclosure even though it successfully synthesized an answer from non-site sources.
• ›No fetch or search returned machine-readable structured data (schema.org, JSON-LD) about business segments, pricing models, or target audiences, forcing the agent to construct the narrative from fragments and external sources rather than from authoritative, site-native metadata.