The agent assembled a comprehensive explanation of Middesk by combining sparse, high-level content from the official site (homepage, platform, about pages) with substantive information from web search results, help documentation, and third-party comparison/pricing aggregators. The official website proved minimally agent-readable—most pages returned truncated Next.js bundles with limited visible text, and critical pages like /pricing returned 404s—forcing heavy reliance on external sources (50% prior knowledge, 20% web search, 30% previous artifacts per metadata). Despite this friction, the agent delivered a satisfactory answer across all four requested dimensions (what, who, pricing, differentiation), though with an explicit caveat that pricing details remain hidden behind a sales gate.

• ›The official Middesk site (steps 1, 3, 6) returned minimally useful HTML—Next.js bundles with no extractable text snippets—and critical navigation URLs (e.g., /pricing, /solutions, /industries) returned 404s, making direct on-site discovery impossible. The agent had to route around a broken information architecture.
• ›Pricing information came entirely from external sources (TrustRadius, Vendr, xpay, help.middesk.com)—not from middesk.com itself. The agent explicitly flagged this as 'confusing' and noted the friction (no public pricing page, sales-gated model). This is a notable limitation in site design for an AI trying to understand the product.
• ›Core product positioning ('business identity platform,' KYB/fraud detection, AI agents) was extracted primarily from web search results and press releases (steps 13, 14, 15, 16), not from the official homepage or platform page. The site does not clearly articulate what the product does in a machine-readable way.
• ›Competitor differentiation came from third-party comparison articles (step 11: Signzy, Global Database) and prior knowledge, not from Middesk's own positioning. The official site does not publish a competitive landscape or feature matrix.
• ›The agent's final answer cites 17 sources; only 4 are from middesk.com proper (1 homepage, 1 platform page, 1 about page, 1 product release page). The remaining 13 are external sources (TrustRadius, Vendr, Signzy, Abrigo, Finovate, PR Newswire, etc.). This 76% external dependency ratio reflects low on-site discoverability.
• ›The agent explicitly called out missing information: no public pricing, limited feature docs, unclear bundling, no obvious free trial, and no integration complexity guidance. These are all gaps the official site should have filled but did not expose in a machine-readable form.