Courtyard Research

Original data on how AI reads the web

Jul 30, 2026Sam Pajot-PhippsServer-side analysis

How ChatGPT retrieves a rental catalogue

41 days and 2,001 IP-verified fetches from one agent, ChatGPT-User, with crawlers excluded so every request is a moment someone was mid-question. Retrieval started at 31 hours, went to neighbourhood pages over individual listings, ended after a single fetch two-thirds of the time, and constructed filter URLs we had never published.

Read the report
Jul 28, 2026Sam Pajot-PhippsServer-side analysis

An experiment to unlock better agentic commerce for theatres

Cineplex seat availability existed only inside a booking widget. We republished it for the whole country as structured pages and measured what AI did with it: 49 days and 311,048 IP-verified AI requests. Crawlers found the site in 50 minutes, AI read 81% of every URL it served, and demand went straight to the seat maps, while sending 48 people.

Read the report
Jul 15, 2026Sam Pajot-PhippsServer-side analysis

How AI reads a knowledge base

Thirty days of IP-verified server logs across twenty anonymized small-business subdomains. What we found: Anthropic trains on our fleet but barely retrieves from it, OpenAI does both, and live retrieval concentrates tightly on the buyer-intent pages a customer would open before booking.

Read the report
2026Benchmark

The State of AI Visibility 2026

A cross-vertical benchmark of how discoverable small businesses are to AI assistants, measured across ChatGPT, Claude, Gemini, and Perplexity.

Read the report