Visibility Is Not Truth

A lot of smart people are talking about how AI search is going to reshape the way customers find dealerships. They are probably right. The shift is real and worth taking seriously.

What caught my attention was how many of them were talking about it like they already had the formula. Agencies with new product lines. Consultants with new acronyms. A lot of confidence from people who, as far as I can tell, are studying the same outputs anyone else can pull up in thirty seconds.

So I asked.

If you run a dealership, someone has probably told you that AI search is going to change how customers find you. They're right. Customers are already asking AI tools, which dealer to trust, who has fair prices, and who is easy to work with. Not shocking.

What I have been inundated with is people talking about it like they had the formula. Agencies with new product lines. Consultants with new acronyms. A lot of certainty from people who, as far as I can tell, are studying the same outputs the rest of us can pull up in thirty seconds.

I didn't test that theory by asking about AI search. I asked a normal business question. I told an AI system I was a car dealer looking for four to six competent, affordable, knowledgeable, and honest paid search and paid social consultants. Nothing exotic. Just a dealer trying to hire well.

The answer looked great. Then I started asking where the confidence came from. The conversation that happened is what you read. Here's more detail if you need it.

I asked a straightforward question: find me four to six competent, affordable, knowledgeable, and honest consultants who specialize in paid search and paid social for car dealerships.

Fair question. The kind of question a dealer actually asks when trying to make a smart hire.

The answer came back fast. Organized. Specific. Confident. Six ranked names, each with a tidy paragraph explaining why they fit. It sounded like research had been done.

That is where the problem starts.

The answer looked good before it was questioned. A bad answer that looks bad is easy to throw away. A weak answer that looks polished can move real decisions. It can move attention, money, and trust.

When I asked where the confidence came from, the answer fell apart in stages.

Circular sourcing. The model used an agency's own marketing claims as part of the reason to recommend that agency. Sales copy treated as evidence. No dealer would accept that from a vendor in person. The model did it without blinking.

Credibility theater. The model cited DealerRefresh forum posts, BBB ratings, and DealerRater pages as supporting evidence. When pressed, it admitted forum posts are opinions that can be anecdotal, biased, anonymous, or planted. BBB is pay-to-play accreditation that proves nothing about campaign performance. DealerRater is a consumer car-buying review site with zero relevance to evaluating a marketing vendor. The model's own words: those sources "added credibility theater more than substance."

The admission. After three follow-up questions, the model said what it should have said first: "I don't have a verified, reliable way to answer your original question." It could not audit campaign results, confirm testimonials, verify cost-per-lead claims, or validate any of the credentials it had just presented as fact.

Then I asked the harder question. If this is how the model answered a dealer-principal with 30 years of pattern recognition, how is it answering the consumer searching for "trustworthy dealership near me"?

The model confirmed it works the same way.

When a consumer asks "best Kia dealer near Joliet," the model assembles fragments from training data and web results into something that reads like a recommendation. Confident. Structured. Built on the same unverifiable material it just admitted was unreliable.

The consumer doesn't know that. They don't have the experience to push back. They take the answer.

User-generated content was never meant to be the final word.

A review is one person's statement. A forum post is one person's opinion. A star rating is a rough signal. UGC can be useful. Real customers leave real feedback. Patterns can matter. A good dealer should pay attention.

But UGC is not an audit. It is not a controlled sample. It is not verified. It can be fake, bought, coached, planted by competitors, written by people with no expertise, or outdated by years. There is no penalty for being wrong, no method, no cross-examination.

AI changes the danger because it makes weak evidence look complete.

A sloppy pile of reviews, vendor claims, directory listings, and forum posts goes through the model and comes out as a clean, confident recommendation. The model does not improve the evidence. It improves the packaging. A raw Google review carries obvious limitations. An AI-generated recommendation built from that same review carries implicit authority that the raw review never had. The underlying material is the same. The presentation is not.

When I asked the model whether any technical formula drives its recommendations about dealerships, it said no.

It then described several factors it believed influenced its own answers: structured data and schema markup, consistent business listings, crawlable content, review volume and recency, and the specific language used in reviews. It also listed things it believed had no influence: paid search spend, DMS data, CRM data, CSI scores, customer retention, employee tenure, service absorption.

That is not the same as proof.

Those are self-reported descriptions from a system that had already demonstrated, in the same conversation, that it will produce confident answers from weak foundations. The model was describing its own behavior from the inside. It may be directionally accurate. It may be incomplete. It may be wrong in ways it cannot detect. I have no independent way to verify any of it, and neither does anyone else outside the companies that built these systems.

This is the part that should matter for dealers. When an expert tells you "AI systems prioritize schema markup" or "review language drives recommendations," ask where that claim comes from. If the answer is "we tested it by prompting the model and observing the output," that is the same method I just used. It produced a description, not a specification. It is observation, not proof.

Some of those observations may turn out to be correct. Some may be incomplete. Some may be artifacts of how the question was asked rather than how the system actually works. Dealers should treat these ideas as hypotheses worth testing, not formulas worth buying.

This brings us back to the agencies and consultants telling dealers how AI recommendations work.

Most of them are doing the same thing I did. They prompt the model, study the output, and build a theory. That can produce useful observations. It does not produce proof of how the system works internally.

They stuck their fishing pole in the same stocked trout pond as the customer and think they caught a different fish. The pond was stocked with trout. What makes anyone think the trout on their line is something else?

A person can say, "We are seeing that AI systems seem to favor clear entity information, crawlable content, consistent data, and review language that matches customer questions." That is a useful observation.

A person should be much slower to say, "This is how AI decides who to trust."

The first statement is about visibility. The second is about truth. Dealers should not let anyone blur those together.

This is not a reason to do nothing. The customer discovery layer is changing. Dealers who ignore it will get hurt.

But the answer is not to buy certainty from the first person selling it.

Make sure the public record reflects the real business. The website should explain what the store actually does well. Inventory should be readable by machines, not locked behind rendering that AI crawlers cannot access. Service pages should be useful, not decorative. Reviews should be earned, not scripted. Data should be consistent everywhere. The dealership's real strengths should be visible to both humans and machines.

Do that work. It probably matters. But be honest that "probably matters" is the current state of knowledge. Anyone selling guarantees is outrunning their headlights.

When a vendor tells you they know how to make your dealership win in AI search, ask how they know. Ask what they tested. Ask whether results held across markets. Ask whether they can separate cause from coincidence. Ask whether they are measuring actual customer behavior or just screenshots of AI answers. Ask whether they can tell the difference between normal SEO and something genuinely new.

The question dealers should keep asking is simple: how do you know?

Not what did the model say. Not what did the webinar say. Not what did the audit say. How do you know? Can you prove it? Can you repeat it? Can you show the method? Can you admit what you don't know?

The car business has always had people selling shortcuts. AI did not create that. It just gave the shortcuts better language.

A dealer who understands this has a chance to use the new tools without being used by them. A dealer who does not may end up paying someone to explain the pond while both of them are holding a trout but the vendor insists his is a bass..