Visibility Is Not Truth
A Dealer vs. the Recommendation Machine

Visibility
Is Not Truth.

A car dealer asked an AI system for honest paid search consultants. It produced a confident, ranked list. Three follow-up questions later, it admitted it had no verified way to answer at all.

This is a real conversation between a car dealer and an AI system.

I was looking for a new paid search and paid social consultant/team. That part is straightforward. Any dealer who has been through a few agencies knows the cycle: hire, hope, measure, get frustrated, start looking again.

At the same time, I had been listening to a growing number of agencies and consultants tell dealers that AI search is the future of customer discovery. AEO. GEO. Answer engines. AI visibility. A lot of certainty from people who, as far as I can tell, are studying the same AI outputs the rest of us can pull up in thirty seconds.

So before I started calling around, I wanted to see what would happen if I ran the search through an AI tool myself. Not a question about AI search theory. A normal hiring question. "Find me four to six competent, affordable, knowledgeable, and honest consultants who specialize in paid search and paid social for car dealerships."

The answer came back fast, confident, and organized. It looked like someone had done the work. But when I read through the names and the reasoning, I knew there were credibility issues. What I did not know yet was how broken the process behind those answers actually was. I assumed the methodology was weak. It turned out throwing darts at a board could have gotten me to the same place.

What follows is the full transcript of that conversation, with editor's notes pinned to the moments that matter. After that, the analysis of what it exposed and why it should change how dealers evaluate both AI recommendations and the experts claiming to understand them.

It's not pretty...

Part One — The Transcript

Verbatim, with editor's notes
  1. 01
    The Six Picks
    A simple ask for honest consultants. Claude returns a ranked list, fast.
  2. 02
    The Avoid List
    Author asks who not to call. Claude delivers the conversation's most useful answer — and buries it under more unverified material.
  3. 03
    Caught: CF Search
    Author points out that CF Search was cited as a source for recommending itself.
  4. 04
    Credibility Theater
    DealerRefresh. BBB. DealerRater. None of them belonged.
  5. 05
    The Admission
    Six messages in, the model says what it should have said first.
  6. 06
    The Consumer Parallel
    If the model fails a dealer principal this way, how does it answer a customer?
  7. 07
    The Mechanics
    No formula. Pattern matching from training data, plus whatever Google surfaces.

Part Two — The Analysis

Eight sections · ~10 min
§
Visibility Is Not Truth
Why an LLM told a car dealer how to find honest experts, then admitted it lied — and what it means for every dealer betting on GEO and AEO.

Appendix — Scott's Notes

§
Scott's Notes
Some brief thoughts on this experience — on UGC, AI scale, the trout farm fallacy, and the named vendors.