An AI agent may choose a car before the customer visits the dealer’s website
For years, we have built the digital car-buying journey around a website visit. Advertising was meant to attract attention, the model page to build interest, the configurator to help customers choose, and the enquiry form to connect them with a dealer.
Every step required the user to do something. Visit, read, compare, choose and leave their details.
What happens when an AI agent does some of that work for them?
The customer will still need a car. They will still have a budget, preferences and a purchase timeframe. But they may expect someone to compare the available offers and present three sensible options instead of another thirty links.
This is the direction agentic commerce is taking. AI helps customers understand an offer and, with the right integrations, can also take further steps on their behalf.
For the automotive industry, this changes the question we should ask when designing our digital presence.
Can an agent entrusted with finding a car understand and compare what we offer?
Who will influence the customer’s choice?
In the article “Who Gets Paid When AI Does the Shopping?”, Alex Immerman and Santiago Rodriguez highlight an important shift in online commerce. An AI Agent does not need to own warehouses or organise deliveries to influence how value is distributed across the process. It simply needs to become the place where the customer makes their purchasing decision.
A similar scenario is easy to imagine in automotive.
The importer supplies the car. The dealer prepares the offer, arranges the test drive and hands over the vehicle. The finance company provides financing.
But the shortlist of brands and offers may have been created earlier, in a conversation between the customer and their agent.
This is my interpretation of what that shift means for automotive: some influence over the customer’s choice may move towards whoever helps them build that shortlist.
This does not mean dealers or vehicle marketplaces will lose their relevance. It does change the terms on which they participate in the process.
Customers will describe their needs instead of setting filters
Imagine a request like this:
“I’m looking for a family SUV with an automatic transmission. Budget up to PLN 180,000. It needs to be available within a month, no more than 100 kilometres from Warsaw. Compare an outright purchase with leasing using the same assumptions.”
From the customer’s perspective, this is a simple request.
For the industry, answering it requires connecting multiple pieces of information: model specifications, the exact trim level, vehicle availability, location, price, promotional terms and financing parameters.
The agent also needs to distinguish a car at the dealership from one in transit. The list price from the price of a specific vehicle. A discount available to everyone from an incentive conditional on financing or a trade-in.
This is where the challenge begins.
The information may be on the website. It may even be correct. Yet it can still be presented in a way that commerce agents cannot interpret clearly, making it difficult for them to assemble a complete offer that can be compared with others.
A monthly payment means little without context
In automotive, we have long advertised prices “from” and monthly payments “from”.
A person can call the dealership and ask for details. An agent preparing a comparison needs those details too if its recommendation is to make sense.
Does the payment include VAT? What is the initial payment? How long is the agreement? What are the final purchase payment and mileage allowance? Does the offer apply to the vehicle shown? When does it expire?
Two identical monthly payments can represent very different financial commitments.
The same applies to availability. “Available” might mean a car ready for handover, a reserved vehicle, a demonstrator or a car that has yet to arrive in Poland.
If the organisation has not agreed on what this information means, the agent will be working with ambiguous data.
The business consequence is clear. An AI agent may leave the dealer’s offer out of its recommendations. Or the customer may reject a good proposal because it cannot be compared reliably.
Marketplaces can also become part of this shift
It is easy to conclude that AI agents will bypass vehicle marketplaces and contact dealers directly.
That is one possible scenario. But there is another.
A marketplace has structured inventory, historical data, comparison tools and relationships with thousands of sellers. It can use those assets to supply information to an AI Agent. Autotrader already identifies search within ChatGPT as part of its product development. CEO Nathan Coe and the company’s strategic report explicitly emphasise combining proprietary data and technology with generative AI and adapting to changing search habits, including external AI models and search tools such as ChatGPT. This strategy involves supplying AI Agents with reliable market, valuation and vehicle availability data directly from the Autotrader platform.
Importers and dealers should therefore take care of the quality of their own source data. This allows them to use different distribution channels while retaining control over the vehicle information and purchase terms they share.
Websites will serve both people and the systems helping them
A good user experience still matters. Customers will want to see the car, get to know the brand, check the dealer and build confidence before buying.
Alongside that journey, however, another is developing: an AI agent reading and evaluating the offer on the customer’s behalf.
Google and Shopify are developing the Universal Commerce Protocol to help commerce systems work with agents. OpenAI’s documentation, meanwhile, identifies structured product catalogues, prices and availability as the foundation for supplying offers to shopping experiences in ChatGPT.
These are examples of commerce infrastructure taking shape. They do not yet mean that autonomous car purchases are commonplace in Poland. Not yet.
But they show why preparing the data in advance matters.
Being able to read an offer is the first step. Checking current availability, reserving a car or initiating financing requires further connections to the organisation’s systems.
Automotive Golden Context starts before publication
In my article “AI Will Not Fix Disconnected Systems. First, a Common Data Context Is Needed”, I wrote about the need for a shared data context before implementing AI.
Agentic commerce provides a very practical application of that principle.
Before publishing an offer, we need to know exactly what we are selling, on what terms, to whom and within what timeframe. We also need to establish which source takes precedence and who is responsible for keeping the information up to date.
This is the role of Automotive Golden Context: a purposefully designed contextual data foundation that connects data with its meaning, relationships and business rules.
For a vehicle offer, this means linking a specific car to its specifications, availability status, location, price and applicable promotion.
A promotion should have defined conditions, scope and validity dates. A financing offer must include the assumptions needed to understand it. Availability information should have an identifiable source and an update timestamp.
There is no need to integrate the entire organisation at once. You can start with one use case: reliably publishing inventory and promotions.
But the data must come together as a consistent offer that commerce agents can understand.
How we connect this with Automotive Digital Experience
At Datadise, we are developing Automotive Digital Experience as a shared digital platform for importers and their dealer networks.
We design our CMS so that vehicle inventory listings and promotional content can also be correctly interpreted by AI agents. This is supported by structured information about vehicles, offers and the conditions under which they apply.
In this approach, Automotive Golden Context provides the agreed context, while ADE draws on that data and makes the offer available across digital touchpoints.
These two layers need to work together.
A CMS can correctly publish the information it receives. But if a vehicle has the wrong status in the source system, or a promotion is assigned to the wrong trim level, that error will be passed on.
Preparing for agentic commerce therefore starts with data quality and meaning, followed by how that information is made available.
A clearly structured offer makes it easier to use the information correctly. The final recommendation also depends on the customer’s needs, the competitiveness of the offer and how the particular agent operates.
Where to start
I suggest a simple exercise.
Choose a few cars from your own inventory and try to answer the following questions using only the published information:
Can the customer buy this specific vehicle? What is the price, and what conditions must they meet to obtain it? What exactly does the monthly payment cover? Where is the car, and when can it be collected? What is the next step?
If answering requires calling three people and checking several files, you have identified a concrete area to improve.
That work will help customers and sales teams today. It will also prepare your offers for use by AI agents.
It is worth doing before more vehicle comparisons begin to take place outside our own websites.
Because the customer may not have reached us yet, but their agent may already be evaluating our offer.
Our philosophy at Datadise remains unchanged: “Everything works better when it is connected.”
