What AI assistants ask car dealers is about one vehicle, not the dealership. One assistant writing for a shopper asked for the trim, exterior and interior colors, VIN or stock number, whether it was available now or arriving later, and whether it could be leased. None of that is on a homepage.
A dealership executive publicly shared a message their store had received from Instinct, a general AI personal assistant, writing for a customer who wanted a 2026 Volvo XC90. The customer had been specific: a B5 or B6, not the plug-in hybrid; six seats; Ultra trim preferred, Plus acceptable; Onyx Black or Vapour Grey. The assistant called the message an inventory inquiry, said it wasn't asking to hold a vehicle, and didn't share the customer's contact details.
That's one message, seen as a screenshot. It says nothing about how often assistants write to dealers, and we didn't verify the sender or the customer. What it asked for is still the clearest list we've seen of what an assistant needs from a store. For any matching vehicle:
The trim.
The exterior and interior colors.
The VIN or stock number.
Whether it was available now, or when it was expected to arrive.
Whether it could be leased.
Why none of it is on your website
Every one of those questions is about a single unit, and most of them are operational. Trim and colors are on the vehicle detail page, if the unit made it into the feed. Whether it's on the lot, in transit or still an allocation lives in the DMS and the OEM allocation portal. Lease eligibility depends on lender programs that change monthly. And some of it is only in the sales team's heads, like the car a salesperson has promised to someone but hasn't written up.
It changes by the day: a car in transit on Monday is sold on Wednesday. A homepage, a model page and an About page can't answer for this exact car right now, however good they are.
Cox Automotive makes a related point from the discovery side. In an August piece on AI and vehicle discovery, Tessa Nadik, an AVP of product management at Cox, wrote: “If vehicle information is incomplete, inconsistent, or difficult to interpret, inventory becomes harder for AI systems to surface confidently.” Her advice is fuller listings and consistent dealership details, and that's the right first step. The XC90 message is the step after it: once an assistant has found a likely car, it asks about status and terms that a fuller listing still can't answer.
What Cox's AI tracker shows
Cox Automotive's AI in Auto Retail Tracker puts numbers on both sides. Of in-market vehicle shoppers, 63% say they will definitely or probably use AI on their next vehicle purchase (61% in Q1 2026). Among dealers, 82% use AI in some form, flat from Q1 2026. But only 29% of dealers are actively adjusting for AI-powered search, down from 32% in Q1 2026. And 32% of dealers know they need to adjust to AI-powered search but haven't started, up from 26% in Q1 2026.
Read those the way Cox reports them. The tracker is a quarterly survey of U.S. franchise and independent dealers (504 in the first quarter of 2026, 483 in the second, fielded in March and May) and of shoppers planning to buy a vehicle within 12 months (1,505 and 1,502). Shoppers are stating an intention, not describing a purchase. Two waves are a baseline and one comparison, not a trend. And “adjusting for AI-powered search” is the dealers' own answer; the tracker doesn't say what the adjusting involves.
Unknown is a valid answer
The expensive mistake is a stale feed turning into a confident “no.” Say the feed lists a black six-seat XC90 B6 Ultra, but the record hasn't updated in four days. Or the car is allocated and hasn't shipped, so it isn't in the feed at all. The tempting answer, from a chat widget, a reply template or an assistant reading your site, is “we don't have that.” The right answer is “unknown”: the last confirmed status, when it was checked, and how to reach a person who can look.
A false no leaves no trace. The shopper's assistant moves on to the next store, and nothing in your analytics shows it happened.
So each status needs to carry where it came from (the DMS, the OEM portal, a salesperson) and when it was read. When sources disagree, say the website shows available and the DMS shows sold, keep both, trust the more authoritative one, and fix the other. Two values, available and sold, can't hold any of this. A unit needs at least six:
Available: on the lot and for sale.
Incoming: in transit, with an expected arrival.
Allocated: assigned to the store, not yet shipped.
Held: set aside for a customer or a pending deal.
Sold.
Unknown: no current reading. Say so, and give the last one.
Contact comes before act
The assistant drew the line itself: it asked, and it said it wasn't asking for a hold. That's the right boundary for now. An assistant can ask about a unit and send an inquiry that a person answers. Holding a car blocks a sale. A test drive takes a salesperson's hour. A lease or finance application touches credit and money. Those wait until the dealer explicitly allows them, with approval, identity checks and a record of who did what.
The dealer also can't verify who's asking. An emailed inquiry from an assistant is an unauthenticated message, so treat it like any other lead: answer the question, and don't commit the car.
What a dealership can do now
Pick one in-transit unit and trace it: what the DMS says, what the vehicle detail page says, what each marketplace listing says, and when each last changed.
Put the VIN and stock number in text on every vehicle detail page, next to the trim and both colors.
Show incoming units as incoming, with an arrival window you'd stand behind, where your OEM and your process allow it.
State lease availability with the date it was last confirmed, or say to ask.
Decide who answers an inventory email from an AI assistant, and route it into your CRM like any other lead. Reply with the five facts and the time you checked them.
Find every place that can say no on its own, such as the chat widget, auto-replies and site search, and make each one say “not confirmed” and offer a person.
Ask ChatGPT, Claude, Gemini and Perplexity about one specific unit on your lot, the way a shopper would, and compare what they say with the DMS.
Where Lehvel is on this
We study this in our automotive research: what dealers and parts sellers know, where it lives, and what an assistant needs to answer correctly. We haven't built a dealer system, and the brief says plainly what is and isn't built.
They can read what's published: your vehicle detail pages and your listings elsewhere. They can't read your DMS or the OEM allocation portal, so status and arrival dates are usually missing or out of date.
Should a dealer answer an email from an AI assistant?
Yes, the way you'd answer any inventory lead: the facts asked for, the time you checked them, and a person to contact. Don't hold a car or start a deal for a sender you can't verify.
What should an assistant say when it can't confirm a vehicle?
“Unknown,” with the last confirmed status, when it was checked and how to reach the store. Never “not available.”
Will publishing this data get my dealership recommended by AI?
Nobody can promise that. Current, complete facts make a correct answer possible. They don't decide which store an assistant picks.
