The agent successfully assembled a comprehensive overview of Français Prep by relying almost entirely on a single machine-readable text artifact (llms.txt) and booking page HTML. The site's primary content is locked behind client-side rendering in Next.js, making direct HTML parsing ineffective; the agent worked around this by discovering and leveraging llms.txt, which contained structured metadata about the service. The agent satisfied the task (explain what they do, who it's for, pricing, and differentiation) but had to bridge gaps in pricing details and course specifics through prior knowledge and inferences, since the rendered HTML pages were not readable from the fetch responses.

• ›Step [6] (llms.txt) was the only content-bearing fetch that actually delivered machine-readable information—it explicitly stated the service focus (TCF/TEF Canada prep), audience (Express Entry candidates), teacher identity (Adèle, native speaker), and positioning (independent, one-teacher model). All other HTML fetches [1, 2, 4, 8, 9, 15, 18, 20] returned truncated boilerplate Next.js markup with no visible content, indicating heavy client-side rendering.
• ›The agent cited llms.txt as its primary source and explicitly linked it in the References section, confirming it was content-bearing. Step [20] (/book page) was cited but appears to have been used to infer the 15-minute consultation offering rather than extract explicit pricing—the actual booking UI rendered client-side.
• ›Pricing details (CA$149 / CA$449 / CA$950) appear in the final response but were not visibly extracted from any fetched response in the trajectory—likely sourced from step [2] (/pricing page) which returned HTML but no visible content in the truncated response, or from prior knowledge. This represents a gap between what the agent claims to have found and what the fetches actually returned.
• ›The agent's web searches [12, 13, 19] returned competitor links but did not cite them in the final response, indicating they were used for discovery context only, not as content sources. The agent correctly excluded them from the References section.
• ›The site is poorly navigable for machine agents: Next.js rendering blocks direct HTML parsing, /courses returned 404, and structured endpoints (.txt variants) are absent or 404. The llms.txt file appears to be the only intentional machine-readable interface exposed, making it a critical bottleneck.