How a Honda Dealership Went From Invisible to #1 in AI Search

The problem: a 9.4% visibility score

A Larry H. Miller Honda dealership in the Salt Lake City market had an AI visibility score of 9.4%.

That number means something specific. When someone asked ChatGPT, Perplexity, Claude, or Google's AI Overviews a question about Honda vehicles or Honda dealerships in Salt Lake City, the dealership was cited in roughly one out of every eleven answers. In the other ten, it did not exist.

Not ranked low. Not buried on page two. Absent.

This is the part most businesses have not internalized yet. Traditional SEO has a floor. Even a badly optimized site shows up somewhere in the results for a branded query. AI search has no floor. You are either in the generated answer or you are not, and there is no page two to fall back to.

Why traditional SEO could not fix this

The dealership was not neglected. It had a functioning website on a major automotive platform, real inventory feeds, and the kind of local SEO baseline you would expect from a dealer group of that size.

The problem was that none of those signals were the ones AI engines use to decide who to cite.

Traditional SEO optimizes for a ranking algorithm that weighs backlinks, domain authority, and keyword relevance. AI citation selection works differently. It weighs whether a page contains a clean, extractable, structurally obvious answer to the specific question being asked, and whether the entity behind that page is machine-legible enough to trust.

What I actually did

The work broke into four tracks that ran in parallel. I have written about the underlying framework in how AI search engines choose sources; this is that framework applied to a real property with real revenue attached.

1. Query mapping before any content work

I started by building the actual question set. Not keyword volume exports, but the phrasing real buyers use when they ask an AI assistant about buying a Honda in Salt Lake City.

That distinction matters. Keyword tools give you "honda crv price salt lake city." Real users ask "how much should I expect to pay for a CR-V in Salt Lake City right now?" Those are different queries with different winning answers, and AI engines respond to the second phrasing.

Every subsequent decision traced back to that mapped query set. Content was not written to fill a calendar. It was written to answer a specific mapped question that had a measurable visibility gap.

2. Structured data rebuilt for entity clarity

The existing schema was platform-default markup: technically present, functionally useless for AI citation. It described the page. It did not describe the *entity*.

I rebuilt it around what an AI model actually needs to establish trust:

• AutoDealer and LocalBusiness types with complete address, geo coordinates, hours, and telephone
• Vehicle and Product schema on inventory with trim-level specificity
• FAQPage schema on every question-answer block, so extractors could lift answers cleanly
• Explicit entity relationships tying the dealership to its parent brand and its physical market

The goal was not "valid schema." The goal was a machine-readable identity that an LLM could resolve without ambiguity.

3. Answer-first content architecture

Every page was restructured so the direct answer appears in the first forty to sixty words. No preamble. No brand throat-clearing. The question restated, then answered.

Headings were rewritten as the actual questions people ask, because AI extractors treat question-formatted headings as discrete answer units. A page with an H2 reading "Inventory Overview" is invisible. The same page with "How much does a Honda CR-V cost in Salt Lake City?" is a citation candidate.

The content was not better because it was longer. It was better because it was structured so a machine could find the answer without guessing.

4. Measurement from day one

Everything was tracked in Profound, monitoring citation share across ChatGPT, Perplexity, Claude, Gemini, and Google AI Overviews against the mapped query set.

This mattered more than I expected going in. Citation share data showed which specific pages were being cited for which specific queries, which meant optimization decisions were driven by observed behavior instead of theory. The FAQ schema work, for example, produced an outsized share of early gains, and the data made that visible in time to double down on it.

The results

| Metric | Before | After |

|---|---|---|

| AI visibility score | 9.4% | 40.9% |

| Relative improvement | Baseline | 335% |

| Category ranking, Salt Lake City automotive | Not ranked | #2 overall |

| Citation position on target queries | Absent | #1 cited source |

The dealership became the #1 cited source on ChatGPT, Perplexity, Google AI Overviews, and Claude for a wide range of Honda-related queries in the Salt Lake City DMA.

The result I did not expect: on the top five most-searched Honda questions nationally, the dealership's content outranked Honda's own corporate website as the cited source.

Honda corporate has millions of backlinks, decades of domain history, and dedicated content teams. On raw authority, this should not be possible. It happened because corporate content is written broadly for a national audience, while these pages were written to be the specific, complete, citation-ready answer to a specific question. When a model needs to cite a source for a concrete question, structural fit beats domain authority.

What transferred to the rest of the network

This engagement became the pilot model for AEO rollout across the broader dealership network I support at Asbury Automotive Group, which spans more than 175 franchise locations.

That is the part that matters strategically. A single-location win is a case study. A repeatable framework that survives contact with 175 locations, multiple website platforms, and dozens of content contributors is an operating system. The structured data framework built here was standardized and deployed across the Dealer.com-hosted properties in the network, replacing legacy markup with AI-ready schema.

What I would do differently

Implement citation share tracking on day one as a leading indicator, not just the aggregate visibility score.

The visibility score is the headline number, but it is a lagging measure. Query-level citation data shows which content changes are working within days instead of weeks. I did not have granular query-level data until partway through, and the FAQ schema work that drove the largest early gains could have been expanded sooner with that signal in hand.

The transferable playbook

Strip out the automotive specifics and the sequence holds for any business:

1. Map the real questions your buyers ask an AI assistant, in their phrasing, not keyword-tool phrasing.
2. Rebuild structured data for entity clarity, not just technical validity. The model needs to know who you are without ambiguity.
3. Front-load the answer in the first forty to sixty words of every page, and format headings as the actual questions.
4. Measure citation share from day one so optimization follows observed behavior instead of assumption.

The window on this is still open. Most industries have almost nobody building for AI citation deliberately, which means the competitive bar is structural discipline rather than budget. That will not stay true indefinitely.

If you want the deeper technical breakdown of this engagement, the full LHM Honda Murray case study covers the implementation detail. If you are trying to figure out where your own site stands, the AEO readiness scanner will give you an honest starting read in about thirty seconds.