How ChatGPT and Perplexity decide which car dealer to recommend

A shopper in Tampa opens ChatGPT and types: which Toyota dealer near me is honest about pricing and has a good service department? Eight seconds later she has three dealer names, a sentence on each, and a suggestion to check recent reviews. One of those three names is going to get a visit. The other twenty Toyota stores within driving distance did not lose the sale to a competitor. They lost it to not being in the answer. This piece is about how that answer gets built, engine by engine, and what a dealer controls.

What a dealer recommendation prompt looks like

The prompts are more specific than most dealers expect. Across the recommendation-style questions we log for automotive clients, very few are as blunt as best car dealer near me. The common shapes are narrower and carry a constraint: a brand, a body style, a financing situation, a reputation concern, or a service need. Which Ram dealer in Phoenix actually stocks 2500 diesels. Used truck dealers in Ohio that do not add dealer fees. Is there a Honda dealer in Denver with good reviews for the service department. Where should I buy a used Tacoma near Sacramento if my credit is average.

Every one of those has two halves. The first half is a filter: a brand, a location, an inventory type. The second half is a trust test: honest, no fees, good service, works with average credit. The engine has to satisfy both. It finds dealers that match the filter, then looks for evidence about the trust test, and names the dealers where the evidence is clearest. If your site and your public footprint answer the filter but say nothing about the trust test, you get skipped for a dealer whose reviews and pages do.

This is a different game from the one most dealer marketing teams are staffed for. It is not a ranking contest for a keyword. It is a legibility contest for a set of claims. The dealer that is easiest to describe accurately wins, and that dealer is usually the one that took the trouble to describe itself.

How each AI engine builds its answer

The three engines that matter for dealer recommendations behave differently, and the differences change what you should fix first. What follows is the pattern we see in logged answers, not a published spec from any of the companies.

Google has said plainly that there are no additional requirements to appear in AI Overviews or AI Mode beyond the practices that already govern Search, and that its AI features use query fan-out to pull a wider set of links than a standard results page. For a dealer that means the local signals you already work on for the map pack are the same signals AI Mode reads, and a weak Google Business Profile hurts you twice.

ChatGPT and Perplexity have no access to Business Profile data and have to reconstruct the same picture from the open web. That is why marketplace dealer pages carry so much weight with them: an Autotrader or Cars.com dealer page is a tidy, structured summary of your name, address, brands, inventory count, and review score. It is often the cleanest description of your store that exists online. That should bother you.

The four signals AI engines read before they name a dealer

The relative weight of those four moves by engine and by prompt, but the order of failures is consistent. Dealers almost never fail on entity clarity alone. They fail because the reputation corpus is thin or generic, and because the site has no sentences an engine can lift. Both are fixable in a quarter without a vendor ticket.

Those shares are from our own tracking and will move as the engines change, so treat them as a shape rather than a benchmark. The shape is the point: roughly a third of the evidence about you comes from marketplaces, and less than a fifth comes from pages you wrote. The way to change the ratio is not to fight the marketplaces. It is to make your own pages worth citing.

Why marketplaces show up in the answer and the dealer often does not

Open a marketplace's dealer page for any store and read it as an engine would. Name, address, phone, brands sold, number of vehicles listed, average review score, review count, a few review excerpts, hours, and a map. Every fact is labeled. Every fact is in text. Now open the same dealer's homepage: a hero slider, a search box, three OEM banners, a finance CTA, and a footer. The engine can extract eight facts from the first page and one from the second.

That is the whole explanation. Engines cite what they can parse, and marketplace dealer pages are built to be parsed. The fix is not exotic. Your site needs an about page and dealership location pages that state the same facts in the same plain form, plus the things the marketplace cannot say: your specialty, your pricing policy, your service department's strengths, the neighborhoods you draw from, the reasons a shopper picks you over the store across the street. This is the same legibility work that lets independent used car dealers get named in their metro despite having a fraction of a franchised store's footprint.

There is a second reason dealers fall out that has nothing to do with the site. When a shopper adds honest or no hidden fees to the prompt, the engine looks for evidence on that exact point. If your reviews never mention fees and your site never mentions pricing policy, there is no evidence either way, and the engine names a competitor whose reviews say straightforward pricing forty times. Silence on a trust question reads as a no.

How to become the named answer

The multi-location version of step two deserves its own reading, because dealer groups fail it in a specific way: one profile does the work of five, and four stores are invisible. We laid the structure out in our piece on Google Business Profiles for multi-location groups. Step six is easy to do badly. If the account reads like a marketing department, the engines and the humans both discount it.

The Cox figures are worth sitting with. A quarter of new-vehicle buyers already used an AI tool somewhere in their shopping process, and the ones who did were unusually satisfied. Satisfied buyers repeat the behavior and tell friends. The share of dealer decisions that route through an AI answer is going to grow, and the dealers who get named early build a reputation record inside the engines that latecomers have to overcome.

How to know whether it worked

You cannot manage this from a rank tracker. The measurement is a fixed set of prompts, run on a schedule across ChatGPT, Perplexity, and AI Mode, with the named dealers and cited sources logged each time. Twenty to forty prompts per dealership location is enough: brand plus metro, brand plus metro plus a trust test, model plus metro, service plus metro. Track three numbers per prompt: whether you were named, what position you were named in, and which sources the engine leaned on. That last number tells you where to work next. If the marketplace share of citations about you is not falling month over month, your own pages are still not worth citing.

We run this for every automotive client through our AI citation tracking setup, and the pattern that emerges after a quarter is consistent: the dealer gets named first on the prompts closest to its actual specialty, then the naming spreads outward to adjacent prompts as the reputation corpus fills in. The broader mechanics of getting an automotive business cited, across dealers, fleets, and parts sellers, are in our generative engine optimization overview for automotive.

Start with one dealership location and one prompt set. Rewrite the dealership page, fix the profile, run the prompts weekly, and watch which sources the engines cite. When your own pages start showing up in the citations, you have become describable. When your name shows up in the answer, you have become the recommendation. Those two events are usually about eight weeks apart, and the second one is the only marketing result that sells a car before the shopper has typed your name.

See where you are cited today

A free snapshot audit of your rankings and AI citations before we ever talk.