A Page Went From 4 AI Citations to 49 After It Started Naming Competitors. Your Dealership's Pages Only Talk About Your Dealership.

Most of what your store has been told about AI search visibility is theory wearing a confident voice. Somebody ran a query, took a screenshot, and built a service line on it.

On September 14, Search Engine Land published something closer to evidence. Zeeshan Yaseen ran two structured GEO experiments back to back, tracked every query manually, and logged 775 citation events across both. The first was a paid engagement for an existing brand that ran several months. The second was a 30-day cold start for a SaaS agency with no measurable AI presence at all when the clock started.

Both tests tracked 15 commercial-intent keywords. The first covered ChatGPT, Claude, Gemini and Perplexity. The second added Google AI Mode and Grok.

Buried in the second experiment is one result that nobody in the automotive vendor market is talking about, and it is the only lever in the whole writeup that a dealership fully controls.

The number

In experiment two, the brand published its own listicle. For most of the run it did what every dealership page does: it talked about the company that owned it.

On June 23, that page was rewritten to include the brand's leading competitors instead of presenting the brand alone. Mentions of that source rose from four to 49, a 12.25-fold increase. Daily visibility across the whole experiment climbed from 22 on June 23 to a peak of 95 on June 27.

The domain did not change. The author did not change. The keyword target did not change. The scope of the content changed.

And this was not a one-off. The same effect showed up in the first experiment, running the other direction. One listicle had placed the brand alongside Lily Ray and Aleyda Solis, two names any search professional knows. After those names were removed, performance declined within days.

Two brands. Two verticals. One mechanism, observed in both directions. Yaseen calls it one of the findings he would prioritize for further testing, which is the right amount of confidence to have about it.

Why this cuts against instinct

Walk into any franchised store and propose a page that names the competing rooftops in the market by name and describes honestly what they are good at. You will not get past the first sentence.

The reflex makes sense inside a hundred years of retail advertising. You buy the billboard to say your name. You do not buy a billboard and put a competitor on it.

A language model is doing something else entirely. It reads the web looking for the shape of a category. A page that covers one company is a source about one company. A page that covers the field is a source about the field, and "best three-row SUV for a family in Tampa" is a question about the field.

This is the same structural reason your own website loses the citation game in the first place. Shero Commerce's July study logged 1,851 sources cited by Google AI Mode, ChatGPT and Perplexity and found 2.8% of them were a brand's own page, against roughly 59% that were publishers. AirOps reports the same pattern from its 2026 State of AI Search work: 85% of brand mentions came from third-party pages rather than owned domains. We wrote up the dealer version of that problem in August.

The usual conclusion drawn from those numbers is that your website does not matter for AI visibility. Yaseen's second experiment says something more useful. His owned listicle produced 14.0% of 437 source mentions while third-party listicles produced 85.8%, so the ratio holds. What moved his own page by an order of magnitude was a single editorial decision: widen the scope until the page stopped reading as an advertisement. Schema did not do it. Neither did an AI text file or a readability pass.

What the experiments do not prove

These are two tests, run by one person, tracked by hand.

Yaseen is an SEO and AI search consultant who founded ZeeKnows and co-founded a digital marketing agency, so he sells in this exact category. Search Engine Land is owned by Semrush, which also sells AI visibility software, and the piece states the contributor was not asked to mention Semrush. None of that makes the data wrong. It does mean it deserves the same scrutiny you would give a vendor deck.

Neither experiment involved a local business. Both were national commercial-intent keywords in B2B services. Nothing here was tested against "Ford dealer near me," and local queries resolve through a different retrieval path than category queries do.

The 12.25-fold jump was a single before-and-after on one page with no control. Other placements were going live during the same window.

And the author reversed his own headline conclusion between the two tests. After experiment one he recommended publishing listicles on your own site. After experiment two he demoted owned content to a foundation rather than a growth engine. Take that as a mark in his favor and as a warning about treating any of these numbers as settled.

One more caveat that matters more for us than for him. The channel mix did not transfer between his two tests at all. PR was 24.1% of citations in experiment one and 0.2% in experiment two. Platform share flipped too: ChatGPT led experiment one with 148 citations, while experiment two ran Gemini 104, Google AI Mode 95, Claude 59, ChatGPT 32, Grok 4 and Perplexity 4, with Gemini and AI Mode taking roughly two-thirds between them. If the mix will not survive a move from SEO consulting to SaaS link building, assume it will not survive a move to car retail either.

What this means for your dealership

Your comparison pages are an AI asset, and you probably have two of them. Most stores have a "why buy here" page and nothing else. Every honest cross-shopping question your sales floor answers ten times a day is a page you have never written. Which of these three midsize trucks tows more. What a lease costs against a 72-month note on the same unit. Which nearby franchise carries the trim you actually want. That content sells and it is the shape a model can quote.

Write about the category, not the rooftop. The version of this that works is a genuine buyer's resource that includes options you do not stock, written to be useful rather than to redirect. The version that fails is a hit piece with a competitor's name in the H1. Comparison is the lever. Attack copy is a liability, and in a franchised environment it is a compliance conversation with your OEM you do not want to have.

Stop paying for AI visibility packages that are schema and an llms.txt file. Google's own guidance already says there is nothing special to do for its generative features. This experiment adds the positive case: the on-page change that produced the largest measured effect was editorial scope. That is a content decision, not a technical one, and your vendor should not be billing it as a platform feature.

Assume every placement expires. Roughly half of the sources in experiment one stopped being cited within 30 days. One placement dropped from 29 mentions to 11 week over week. AirOps reports half of cited pages change every month and that pages without quarterly updates were 3x more likely to lose citations. We covered citation volatility in July and this is more of the same evidence. A page you wrote in March is not working for you in September unless somebody touched it.

Do not read citation counts as traffic. In experiment two, Indie Hackers generated more citation volume than any other source while its referral traffic stayed flat. TechBullion produced fewer citations and took sessions from one to 64. Your visibility dashboard is counting the first thing. Your GA4 is counting the second. They will disagree, and Search Console's AI report only shows you impressions anyway.

Be patient with your own pages and measure them late. Time to first citation ranged from one to 18 days across his placements, and the slowest source of all was the one he controlled, at 18 days. Four placements were live and still uncited when the measurement window closed. Checking a week after publish tells you nothing.

What to do about it

1. Run your five highest-value commercial queries through ChatGPT, Gemini, Google AI Mode, Claude and Perplexity yourself this week. Log every source cited. That list is your real competitive set, and it will not match your rank-tracking report.
2. Pick one cross-shopping question your sales floor answers constantly. Write a genuine category page on it that names the alternatives, including vehicles you do not carry.
3. Put the answer in the first 100 words. Yaseen found a key takeaway block near the top of the page produced a larger improvement than any other on-page change he made.
4. Un-collapse your FAQs. Content visible by default performed better than content hidden behind expandable sections, and question-based headings beat comparison-style headings.
5. Give any page you care about a calendar date for its next refresh. Not a wish, a date.
6. Re-run step one in 30 days against the same queries and compare. That is your baseline.

The part worth sitting with

For a year the AI visibility pitch sold to dealers has been about getting mentioned. Get cited. Get into the answer. Buy the dashboard that counts it.

These experiments point somewhere less flattering and more actionable. What moved the numbers was who the brand appeared next to, and how recently that relationship had been reinforced. A store that is genuinely useful about its market gets described as part of that market. A store that only talks about itself is a source about one dealership, and almost nobody is asking a question about one dealership.

We publish comparison pages about our own category for exactly this reason, including ones that name our largest competitors in the headline. It is uncomfortable the first time. It is also the only page type on this list that nobody can buy out from under you.

If you want a read on how your store's site currently looks to the systems doing the answering, book a walkthrough and we will go through it with you.