The Customer Is Bringing an Agent · Series Hub · The Customer Is Bringing an Agent: Automotive Retail After the Search Box
They moved between Google, OEM sites, dealer websites, marketplaces, reviews, payment calculators, vehicle detail pages and whatever collection of browser tabs was necessary to turn a vaguely defined need into a reasonably informed decision. Every system contributed a piece. The shopper carried the pieces from one place to the next.
We became so accustomed to that behavior that we called it the customer journey instead of what it also was: a considerable amount of unpaid research.
The internet made information abundant. It did not necessarily make decisions easy.
Artificial intelligence is beginning to change that division of labor.
The important development is not simply conversational search. Customers can now describe a problem in more natural language, certainly, but the larger shift begins when the interface retains the objective, researches against it, eliminates unsuitable alternatives and carries what it learned into the next step.
The customer journey is beginning to move from assisted discovery toward delegated work.
That is a more consequential transition than replacing ten blue links with a generated answer.
Search historically helped the customer find places to continue working. An agentic interface can increasingly perform portions of the work itself.
Automotive will not follow conventional ecommerce perfectly, and anyone promising a fully autonomous F-150 in the driveway before breakfast should probably be asked a few follow-up questions.
A vehicle is finite inventory. Geography matters. Condition matters. Pricing moves. Financing introduces additional parties and regulatory obligations. Trade equity complicates the economics. Incentives vary. Human judgment remains useful at several points in the transaction.
But complexity does not protect the traditional automotive funnel.
It gives intelligent software more work to help remove.
A search query has always been a compressed representation of something larger.
The shopper typing “best three-row SUV for towing” probably does not wake up with a deep emotional attachment to that keyword. They may have two children, a dog, baseball equipment, a 5,000-pound boat, a garage with an unforgiving wall and a spouse who would prefer not to parallel park a small aircraft.
Traditional search requires that problem to be broken into pieces. The customer researches vehicle classes, towing, seating, trim levels, pricing and local availability separately, carrying the accumulating context between interactions.
The customer is doing the orchestration.
Google’s current direction makes the alternative increasingly visible. Search agents are designed to operate against user-defined criteria in the background, and Google’s agentic calling capability can already gather information from supported local businesses on a user’s behalf. The relevant signal is not that either feature currently replaces an automotive shopping process. It is that the interface itself is taking responsibility for more of the work between question and outcome.
Google describes that transition directly in its 2026 Search-agent announcement.
If an intelligent system remembers that towing is non-negotiable, removes vehicles that cannot satisfy it and retains that requirement while the customer adds budget and geography, meaningful delegation has already occurred.
The software does not need to purchase the vehicle to change the funnel.
Agentic shopping begins to matter as soon as software performs meaningful work the customer previously had to perform manually.
That creates a new competitive question for dealerships. The business is no longer competing only for the shopper’s attention. It is increasingly competing to be considered by the intelligence helping allocate that attention.
This is where the topic stops being an abstract AI conversation and becomes unmistakably automotive.
Generic research is forgiving. Real inventory is not.
An AI system can explain the theoretical differences between two SUVs from widely available information. Once the shopper asks which actual vehicle within seventy-five miles meets a specific combination of towing, seating, price and equipment requirements, the quality of the underlying dealership data becomes part of the customer experience.
The exact VIN matters. So do the current price, equipment, availability, condition and the relationships among those facts.
Automotive has spent decades becoming very good at distributing inventory. We have been somewhat less disciplined about ensuring that every representation of the same vehicle remains coherent enough for software to reason from confidently.
People compensate for that mess remarkably well. They inspect photographs when an equipment list looks suspicious. They infer that one description is stale. They call when incentive language is unclear.
An agent attempting to narrow the decision has a different responsibility. Ambiguity becomes part of the recommendation problem.
OpenAI’s current commerce infrastructure illustrates the broader principle. Product feeds are expected to maintain structured identifiers, factual product information, pricing, availability and seller context because accurate discovery depends on trustworthy and fresh product data.
OpenAI’s current product-feed specification documents those requirements.
Vehicles are far more complicated than ordinary retail SKUs, which strengthens rather than weakens the argument.
The VIN identifies the vehicle. The intelligence surrounding it explains why that vehicle belongs in this customer’s answer.
That means inventory intelligence eventually has to include more than fields and feeds. It has to connect the authoritative vehicle record to market conditions, comparisons, common customer questions and the expertise of the people who actually sell and service the product.
The same shift changes how we should think about the dealership website.
For most of automotive’s digital history, the website was primarily a destination. Marketing happened elsewhere and succeeded when the customer arrived. Traffic therefore became a useful proxy for influence because much of the useful digital experience occurred after the visit began.
That relationship has been weakening for years. Business Profiles, rich search experiences, marketplaces, video and social platforms all answer portions of the customer’s question without requiring a dealer-site session.
AI search accelerates the separation between influence and visitation.
A dealership’s model comparison can inform an answer upstream. Its employee expertise can help establish authority. Its inventory can participate in a comparison before a VDP is opened. Its public business information can affect whether the store remains in consideration.
The website therefore acquires another responsibility: it becomes one of the dealership’s most important controlled expressions of what is publicly true.
Who works here?
What do they know?
What inventory actually exists?
Which markets does the store serve?
What services can it perform?
How do its people, vehicles, expertise and locations relate?
OpenAI’s current publisher guidance is quite practical on the technical baseline: public pages can appear in ChatGPT Search, and publishers that want their content discoverable, surfaced and clearly cited should allow OAI-SearchBot access.
OpenAI’s publisher guidance is available here.
Crawlability gets the dealership into the room. It does not create authority by itself.
The website remains a destination for customers while becoming an authoritative publishing surface for machines trying to understand the dealership.
That actually makes owned infrastructure more valuable as discovery fragments across surfaces the dealership does not control.
The most absurd failure in an increasingly intelligent customer journey may happen after everything upstream works correctly.
The customer’s agent understands the need. It narrows the vehicles. It identifies the dealership. The store has the right VIN. The content answers most of the questions. An employee inside the dealership possesses exactly the expertise the customer needs next.
Then the shopper raises a hand and receives:
First name. Last name. Email. Phone. How can we help you?
Automotive has invested heavily in shortening forms while remaining remarkably tolerant of the fact that the customer’s accumulated context often disappears immediately afterward.
That was frustrating when the customer carried the context entirely in their own head. It becomes harder to justify when an intelligent system may already have organized much of it.
The answer is not dumping every private AI conversation into the CRM. Customer consent and appropriate information boundaries matter enormously.
A better handoff is narrower and more useful.
The customer can choose to carry forward the selected VIN, relevant comparison, unresolved question and preferred next action. The dealership receives enough information to continue rather than restart.
A modern conversion should preserve useful customer progress, not merely capture customer identity.
That can make the human interaction better, not less important. The salesperson no longer has to rediscover the entire problem before contributing expertise. The person enters precisely where the machine stops being the most valuable participant.
For years, discoverability was the scarce digital resource.
AI search increases the value of being understandable enough to participate in an answer.
Agentic systems introduce a third requirement.
Eventually something useful has to happen.
The customer may want to confirm current availability, identify a knowledgeable employee, carry context into a dealer conversation, ask a dealership-specific question or initiate an appointment workflow.
None of those activities requires autonomous vehicle checkout. They are ordinary pieces of a complicated automotive journey that have historically required the shopper to cross another interface and manually pick up the work again.
Google’s Universal Commerce Protocol offers an instructive example of the broader architectural direction. UCP lets participating businesses define supported capabilities for AI environments and supports multiple communication mechanisms, including APIs, MCP and A2A, while the merchant retains its customer relationship and transaction responsibility.
Google’s current UCP documentation is here.
Automotive will expose different capabilities, under different rules, from conventional ecommerce. The principle is still highly relevant.
Discoverability asks whether an agent can find the dealership. Actionability asks whether the dealership has deliberately given authorized intelligence something useful it can do next.
The word authorized is doing important work there.
We ended Year One with The Dealer Still Decides for a reason. Connectivity is not authority. A callable capability is not permission. MCP is a protocol, not a security strategy.
The next dealership has to become easier to work with without becoming easier to exploit.
Much of our first year followed the dealership inward.
Content became infrastructure. Human expertise became signal. Organizational memory became an asset. Interoperability became an operating strategy. The intelligence layer gave the business somewhere durable to understand itself. The permission layer established what increasingly capable systems should be allowed to do with that understanding.
We ended with the dealer deciding.
Year Two begins by looking across the table.
The customer is acquiring leverage too.
Not because every car buyer suddenly cares about agent architecture. Good technology succeeds partly because customers do not have to understand any of the plumbing underneath it.
They simply notice that the annoying work gets easier.
That creates a durable standard for automotive retail.
The dealership should be understandable enough for machines to represent accurately, useful enough for intelligent systems to recommend, open enough for authorized agents to interact with and governed enough for the operator to remain in control.
The durable dealership will not optimize for one AI interface. It will build enough organizational clarity that people, machines and authorized agents can work from the same business truth.
The customer is not abandoning the dealership.
They are arriving with better help.
The dealership should be ready to meet both of them.
The Search Box Is Becoming a Task Box →
How the economics of customer research change when software performs portions of the work rather than simply returning destinations.
Your Inventory Has to Make Sense to Machines →
Why VIN-level accuracy, freshness, context and expertise become more important when intelligent systems begin narrowing real inventory.
The Website Is Becoming a Knowledge Endpoint →
How the dealership’s owned environment evolves from traffic destination into a durable public expression of business knowledge.
The Handoff Is the New Conversion →
Why preserving customer progress may become more valuable than simply making the lead form shorter.
Discoverable Is Not Enough. The Dealership Has to Be Actionable. →
Why the next stage of interoperability concerns governed capabilities as much as discoverable information.
Interoperate with Hrizn MCP →
Give authorized intelligent environments a governed road back to dealership intelligence and capabilities.
Explore Hrizn Creator →
Turn real employee expertise into durable dealership knowledge that can travel beyond the original conversation.
Explore the Hrizn Resource Library →
Go deeper on AI search, structured information, human signals and automotive interoperability.
Apply to Hrizn Banditworks →
Build and pressure-test the emerging automotive agentic ecosystem with operators working at the edge.
Let Dealers Plug In →
Support an automotive ecosystem where dealer choice remains part of the architecture.
Hrizn gives dealerships a governed intelligence layer for connecting inventory, expertise, content and authorized AI environments without rebuilding the business inside every new interface.
Build once. Let the intelligence travel. Keep the operator in control.
Free Around and Find Out.
We Rise Together.
