The Detection Layer · Created by Chris Hudson and Ben Reuling · Operational May 2026
See the machines visiting you.
Hound is the first public-site agent detector. It scores every session that touches a dealership website, in real time, across five labels. Human. Uncertain. Likely Agent. Agent. Verified. It runs today at its first deployment site, Mark Miller Subaru.
The traffic was always there. Now it has a label.
Methodology
Real stores. Real numbers. Here is exactly how.
Over a 7-day window in May 2026, Hound scored 1,070 sessions across the two rooftops at its deployment site. 177 scored Agent or Likely Agent: 8.6% of everything that touched those sites. Labels are assigned in milliseconds from user-agent strings, request patterns, header signatures, navigation behavior, and fingerprint analysis. A session is tagged before the page finishes loading.
This is two rooftops, one brand, one market. This page is not claiming a national number. Nobody measuring honestly can, yet. What the data does say: if two stores in Salt Lake City see 8.6%, the only question that matters is what your store sees.
Hound scores sessions, not people. No personal information is collected or stored.
71% Human20% Uncertain 9% Agent / Likely Agent
How it works
Detection now. Verification next.
Behavioral scoring
Every session is scored in real time. User-agent strings, request patterns, header signatures, navigation behavior, and fingerprint analysis combine into a label within milliseconds. The session is tagged before the page finishes loading.
Cryptographic verification
When an agent presents a signed identity, Hound checks the signature against trusted operator registries. Claude, ChatGPT, Perplexity, and other known operators can be verified in real time. A verified agent is not a threat. It is a warm lead. The emerald label is reserved for them.
The dealer who sees the agents owns the relationships they bring. The dealer who cannot will pay someone else for access to their own customers.
The five labels
Not all agents are equal.
Human
A person browsing. Probably on a phone. Probably looking at inventory. Treat them the way you always have.
Uncertain
The signals do not add up to a confident label yet. Hound holds judgment until evidence accumulates.
Likely Agent
Multiple signals suggest agent activity, identity unconfirmed. This is where most of the action lives today. The agents are here. They have not announced themselves.
Agent
Confirmed AI agent traffic by behavioral and technical signature. Identity unknown, classification certain. An agent visited. Now there is a record.
Verified WARM LEAD
A named, cryptographically authenticated operator. A verified agent is not a threat. It is a warm lead, and the ARM can meet it directly through dmc-12.
Why this matters
You cannot manage what you cannot see.
The CRM worked because every lead announced itself: a name, a number, an inbox. The agent does not. It arrives without ceremony, reads inventory, sometimes asks for a quote, and leaves no trace in the systems most dealers run.
Detection comes first. Deployment comes second. Without detection, deployment is guessing.
Most dealerships have agent traffic right now and no idea. It shows up in analytics as “direct” or “unknown,” or gets flagged as a bot to block. It is not a bot to block. It is a buyer to greet.
Hound tells you which visitors are machines.
GEO tells you whether the machines can find you at all.
Run both. They are two sides of the same question.
Dealer who wants to see your own agent traffic? Operator who wants your agent Verified instead of Uncertain? Write.