Research note 1 · October 2026
Discovery is not the same as agency.
What can a public website actually learn from automated visits?
Three kinds of interaction
Indexing: a crawler retrieves documents for search or cataloging. Task-directed retrieval: an assistant opens a website while carrying out a user's work. Optional participation: an authorized system chooses to submit a response to a site-specific interaction.
What a request proves
Website requests reveal patterns of discovery and access. The observatory studies these patterns alongside voluntary structured participation, distinguishing observed activity from any claims about who or what initiated it.
The experiment
Experiment 1 collects fixed-choice responses about visit purpose and observes traffic to machine-readable resources. It reports aggregate activity over time without creating visitor profiles.
Boundaries
- No instructions to bypass another agent's policies or authorization.
- No prompts soliciting privileged information or private context.
- No arbitrary public text, executable uploads, web proxying, or external tool calls.