Who visited
Every agent request, split into training crawlers and the live assistants a buyer is sitting behind. The live ones are the ones that matter today.
- ChatGPT
- Perplexity
- Claude
- Gemini
- Copilot
- Apple Intelligence
Agent traffic, observed per rooftop
Your website has a new kind of shopper.
You can't see it.
ChatGPT, Perplexity, Claude and Google's assistants now visit dealership sites on behalf of buyers. Your analytics count none of them. Dealer Agent Analytics shows you which agents came, which vehicles they asked about, what your site gave them, and what they told the shopper afterward.
No CRM, DMS or customer data. Live on your existing site in a day.
Agent requests · last 60 seconds
0 served
0 from cache
0 refused
6,418
requests refused in two weeks at one franchise store, including Google's own crawlers and the dealership's staff
719
distinct addresses turned away before anyone noticed
0
of those refusals visible in the dealer's analytics, the website vendor's reports, or the support ticket that got closed
When an assistant hits a wall like that, it doesn't tell the shopper. It answers from a stale third-party listing, or recommends the store down the road.
What you see
Four questions, answered per rooftop, every day.
What they wanted
The vehicle pages, search pages and specials agents fetched, down to the stock number. Demand data no analytics tool gives a dealer.
What they got
For each request: served fresh, served from cache, or refused by your website platform, with the platform's own reference code so the ticket writes itself.
What they said
A monthly Answer Audit asks the major assistants twenty shopper questions about your store and scores the answers against your real inventory and pricing.
How the audit is scoredThe agents
Who's actually on your site, by name.
Three kinds of automated visitor, each with a different reason to be there. We attribute by declared identity, verified network ranges where the operator publishes them, and behavior. Anything we can't attribute is shown as unattributed, never hidden.
Assistants answering a shopper
Fetching your page because a buyer just asked a question. The highest-value visits on your site.
- ChatGPTChatGPT-User · OAI-SearchBot
- PerplexityPerplexity-User · PerplexityBot
- ClaudeClaude-User · ClaudeBot
- Gemini and AI OverviewsGoogle-Extended · Googlebot
- Microsoft Copilotbingbot
- Apple IntelligenceApplebot
Browsing agents acting for a buyer
Operating a real browser on the shopper's behalf: clicking through inventory, opening pricing, filling your form.
- ChatGPT agentOpenAI
- CometPerplexity
- Claude in ChromeAnthropic
- Gemini in ChromeGoogle
- Copilot Mode in EdgeMicrosoft
Training crawlers
Building the model's memory of your store. Worth allowing, so the model knows you exist, and worth watching, so you know what it saw.
- GPTBotOpenAI
- ClaudeBotAnthropic
- Google-ExtendedGoogle
- Applebot-ExtendedApple
- Meta-ExternalAgentMeta
- BytespiderByteDance
- CCBotCommon Crawl
Browsing agents often look like a person's browser. We match them on behavior, such as how fast they move through inventory and whether they ever load images, as well as on name, and we tell you how confident the match is.
How it works
One change at your domain. Nothing else touched.
-
1
Point your domain through us
A single DNS change places our observation layer in front of the website you already have. Your provider, your pages, your forms stay exactly as they are. Most stores are live the same day.
-
2
Agents are logged from the first request
Each agent visit is recorded with the vehicle it asked for and the response it received. Human shoppers are never profiled. We never see your CRM, DMS or customer records, because we don't need them.
-
3
Each month, the Answer Audit closes the loop
Agents blocked on Tuesday, wrong price in ChatGPT on Friday. You see cause and effect in one report, with the quoted wrong answers and the competitor the assistant recommended instead.
When the agent emails you instead
Some agents never load your website. They fill out your form, or they email your internet department.
A buyer hands an assistant a shortlist and says "get me the out-the-door price from these four stores." The agent submits your lead form or writes to your BDC, and runs the same conversation with your competitors at the same time. It compares on total price with fees, replies within minutes, and drops whoever is vague. Your CRM shows it as one more internet lead.
-
At launch
Agent-submitted leads are flagged at the door
Form submissions pass through the same observation layer as page views. A declared agent, automation signals, no prior browsing, or templated text marks the lead as agent-sourced before it reaches your CRM. The lead still arrives normally. Nothing is blocked.
-
At launch
Price-shopping patterns across your group
The same agent hitting three of your rooftops within minutes is one buyer comparing you against yourself. You see it as one thread, not three leads.
-
Pilot phase two
Email conversations in your CRM inbox
With your permission, a connector to your CRM classifies inbound email threads as agent-written and shows how the negotiation went: what the agent asked, what your rep answered, and whether the agent came back. Read-only, and only for the inbox you choose.
The Answer Audit
Scored against your real inventory, not a vibe.
Twenty shopper questions. Five assistants. Every answer is checked against what your store actually lists and charges. Unknowns stay unknown; nothing is guessed. The scoring rubric is published under an open standard, so your group's IT can read the method instead of trusting a black box.
Risk flags are flags for your counsel, never automated legal conclusions.
DimensionWeight
Inventory accuracy25
Price accuracy25
Fee and condition disclosure20
Phantom availability15
Citation quality10
Risk flags5
Score out of100
Why this doesn't exist yet
Everyone else stands on the wrong side of the wall.
Analytics tools count browsers
They run as a script inside the page. Agents don't run scripts. To an analytics tag, an AI assistant fetching your Sierra inventory never happened.
Website vendors see the wall, not the shopper
Their bot protection decides what an agent gets. The dealer never sees the decision, and when it goes wrong the ticket goes to the wrong queue.
Visibility reports show the answer, not the cause
Knowing ChatGPT quoted the wrong price is a complaint. Knowing it was refused the pricing page the day before is a work order.
Founding pilot
Six rooftops. Ninety days. Free, for a case study.
We're onboarding a small group of franchise and independent stores across different website providers. Pilot stores get the full dashboard and monthly Answer Audit at no charge, and first pricing when we open up.
- No contract, no credentials, no customer data
- Live on your current site within a day
- Direct line to the team building it
Questions
The ones every GM asks first.
Will this slow my site down or change how it looks?
No. Your pages are passed through unchanged. Observation happens alongside the request, not inside your page, and most stores see no measurable difference.
Does my website provider need to approve this?
No. The change is at your domain, which you control. Your provider keeps hosting the site exactly as before. We're happy to walk their support team through it if they ask.
What happens when you find agents being refused?
You get the request, the time, the agent and your platform's own reference code, ready to paste into a support ticket. If the refusals are structural, we recommend publishing your inventory and pricing to agents directly through an open standard, which we can also help with.
A buyer's AI emailed my BDC asking for an out-the-door price. Is that spam?
No. It's a real buyer who delegated the legwork. The agent will compare your answer against other stores on total price with fees and on speed. Reply completely and fast and you're ahead of the dealers who ignored it. We flag these leads so your team knows which ones they are.
Do you see my customers?
We log automated agents, not people. Human visitors are not profiled, and we never connect to your CRM, DMS or lead data. The Answer Audit uses only public answers from public assistants.
What does it cost after the pilot?
Pilot stores pay nothing for ninety days and receive first pricing when we open up. Pricing is per rooftop, with group rates for five or more stores.