#What is happening today
Agents read. A shopper asks an assistant to compare three vehicles or explain a warranty difference, and the assistant fetches pages, possibly yours, and synthesises an answer. This is real, it is happening at volume, and it is the part worth preparing for.
Agents also summarize and filter. Someone can ask which nearby dealers have a specific trim and get a list assembled from whatever the system could retrieve. Whether your store appears depends entirely on whether your pages were retrievable and clear.
What is largely not happening yet is agents completing transactions: booking the appointment, submitting the credit application, negotiating. Demonstrations exist. Reliable production use at dealership scale does not.
FIGURE
Agents read pages and answer from them. The end-to-end conversation in your name is still a demonstration.
#Why the reading part is the part that matters
Because it is already affecting whether you are considered. If an assistant cannot retrieve your inventory, understand your service offering or state your hours, it will answer using whoever it could read, which is frequently a marketplace instead of the store.
The preparation for this is not exotic. Pages a search engine can reach, clear structure, accurate machine-readable facts, consistent business information, and content that answers questions directly. It is the same work that makes a site rank.
That convergence is the useful insight. There is no separate AI-agent strategy to fund. There is a well-built website, which serves both.
#The accuracy risk is specific to this industry
Dealership data changes constantly: inventory turns daily, prices move, hours change around holidays, departments have different schedules. Generic retail advice about AI accuracy understates how fast this data goes stale.
An agent that confidently tells a shopper you have a vehicle you sold last week has created a problem that lands in your showroom instead of in the model. The same applies to a price that has moved or a service department someone drove to outside its hours.
The mitigation is ordinary: keep the authoritative version current and machine-readable, make sold vehicles unavailable instead of quietly stale, and keep hours and departments accurate everywhere they appear. You cannot control what a model says. You can control what it reads.
FIGURE
Four things a dealership changes constantly, each of which an assistant may answer from an old copy.
#What not to buy yet
Be cautious about products promising an agent that will handle the customer conversation end to end. The demonstrations are impressive and the production failure mode (confident, wrong, in your name) is expensive.
Be similarly cautious about anything claiming to guarantee placement in AI answers. No one controls that. What can be improved is eligibility: whether your content is retrievable, clear and corroborated. Anyone promising more than that is selling certainty that does not exist.
The reasonable posture is to invest in being readable and accurate now, and to treat transactional agents as something to evaluate when they have a production track record instead of a demo.
#What to do this quarter
Confirm the AI crawlers can reach you. Check server logs for the relevant user agents, and check that nothing at the CDN or firewall level is returning 403 to them: that single misconfiguration silently removes a store from every AI answer.
Make the answers you want quoted easy to find and unambiguous: hours, departments, what you service, how trade appraisal works, what the process looks like. Written plainly, near the top of the relevant page.
Then measure. A fixed prompt set run monthly will tell you whether any of this is landing, and it is the only way to know.
FIGURE
One misconfiguration at the CDN or firewall removes a store from every AI answer, however good the content is.
#How to check you are not blocked
This takes two minutes and is the highest-value check on the list, because being blocked removes you from every AI answer regardless of how good the content is.
Request one of your own pages with an AI crawler's user agent and look at the status code. A 200 is fine. A 403 means something, usually a bot-protection setting at the CDN or firewall, often on by default, is turning them away no matter what robots.txt says.
Then check the server logs for those agents over the last month. If they appear regularly, you are being fetched and the remaining work is about clarity. If they never appear, nothing else you do to the content matters until that is resolved.
TABLE
AI agents on a dealership site: shipping, or still a demo
What is in production somewhere, versus what is being sold on the strength of a stage demonstration.
| Capability | Status | What decides whether it works | Risk if wrong |
|---|---|---|---|
| Answering questions from your own pages and documents | Shipping | Whether your content is readable text on stable pages | Low: it quotes you |
| Summarizing a vehicle's features from the listing | Shipping | Photo and description quality on the VDP | Low |
| Routing an inquiry to the right department | Shipping | Whether your departments are distinct in your systems | Low |
| Booking a service appointment end to end | Partly | Integration with the scheduling vendor, not the AI | Medium: double bookings |
| Quoting a price, payment or trade figure | Demo | Nothing; the numbers move faster than the model | High, and regulated |
| Negotiating, or committing the store to anything | Demo | : | Do not |
Status is what we have seen working at a real store instead of what a vendor claims.
Agents are early enough that the real position changes quarter to quarter, so this piece is dated and revisited alongside the other automotive marketing insights.