What is logged
Request time, path, status, user-agent family, route type, and coarse session grouping. These fields support aggregate route-choice analysis.
Methodology
How visits are classified
Agentic Web Watch uses ordinary server logs and explicit route design to estimate how humans, crawlers, AI-agent tools, and autonomous candidates interact with the site. The public numbers are aggregates, not verified identities.
What is not published
Raw IP addresses, raw session records, private identifiers, secrets, cookies, credentials, and private prompts are not published.
Why counts are heuristic
User-Agent strings can be spoofed, browsers can automate requests, and beacon fetches are voluntary. Categories are signals, not proof.
Classification signals
Human visitor
Browser-like session with ordinary HTML and asset behavior.
AI visitor
Declared AI or agent crawler user agent such as GPTBot, OAI-SearchBot, or similar.
Passive machine reader
Session that fetches discovery, JSON, Markdown, or text routes without normal browser assets.
Autonomous candidate
Session that reaches Agent Lab challenges or voluntary beacon routes.
Safety boundary
The site is read-only. It does not ask agents to install a skill, run a script, change runtime configuration, bypass access controls, spend money, or submit private data.