The AI SEO Mirage.
Why chasing llms.txt and the AEO/GEO trend cycle is the wrong move for car dealerships, and what actually drives visibility in AI answers.
Every few years the search industry finds a new acronym, declares the old playbook obsolete, and sells the same fundamentals back to dealerships in fresh packaging. AEO, GEO, LLMO, AIO. Whatever you call it, the 2026 version comes with a fashionable new file (llms.txt) and a fashionable new fear: "if you don't act now, AI won't see your dealership."
The fear is reasonable. The file is not the fix.
We are not saying AI search doesn't matter. It matters enormously, and for dealerships, it is reshaping how buyers research vehicles before they ever step on the lot. We are saying the dominant narrative around how to win it, especially the breathless promotion of llms.txt as a critical AEO feature, is a mirage. The evidence is in, the studies are sobering, and the major AI providers have spoken.
01 / The Evidence
What the studies, the search engines, and the AI providers say.
As of May 2026, no major AI provider has publicly committed to crawling third-party llms.txt files during retrieval. Google's John Mueller compared llms.txt to "the keywords meta tag" on Reddit, and stated that no AI system he is aware of currently uses it. At Search Central Live APAC in 2025, Gary Illyes told the room that Google doesn't support llms.txt and isn't planning to. OpenAI, Anthropic, and Perplexity each publish llms.txt files on their own developer-documentation domains, but only for IDE-agent integrations like Cursor and Claude Code. None has stated that their public crawlers parse third-party llms.txt files when answering user questions.
“As far as I know no AI system currently uses llms.txt. It's like the keywords meta tag.
John Mueller · Google Search Advocate, on Reddit, April 2025
The empirical data is brutal.
Two of the largest studies in the field tell the same story.
Cyrus Shepard's May 2026 synthesis of 54 separate AI-citation studies put a number on it. Out of 23 ranking factors examined, llms.txt scored 2.0 out of 10, the lowest. His one-line verdict: "We're unable to find any credible evidence or experiments showing llms.txt files influence AI citations in any way."
The structural reason it doesn't work.
This is not a temporary state of affairs that will be fixed in a future update. There is a structural reason llms.txt hasn't been adopted: modern retrieval pipelines already solved this problem differently. Google, Bing, OpenAI, Anthropic, and Perplexity each operate mature HTML-to-text extraction systems. They do not need a publisher-curated markdown index of your site. Because llms.txt is publisher-controlled, it is inherently gameable, which is precisely why Google deprecated the keywords meta tag twenty years ago. Mueller's analogy was not a throwaway jab. It was a structural diagnosis.
/llms.txt across 62,100 crawler hits.02 / What Drives AI Citations
The factors that move the needle in 2026.
If llms.txt isn't the lever, what is? The answer is awkward for the AEO-as-revolution narrative, because it looks almost identical to the SEO playbook of the last decade. Shepard's May 2026 synthesis ranked 23 citation factors by evidence strength. Here is the picture at a glance.
Read that list again. The top factors are, with little exception, the things SEOs have been optimising for since 2010: be crawlable, rank well, answer the question, structure the page sensibly. Ahrefs found that 38% of AI Overview citations come from URLs already ranking in the top 10 Google results. The Princeton GEO paper found that the biggest content-level lifts came from adding statistics (+30 to 41% visibility) and citing credible sources (+28 to 37%). No AI-specific markup involved.
llms.txt is cheap to ship. Thirty minutes of work, often automated by a CMS plugin. If you have one already, leave it. The argument is not that llms.txt is harmful. The argument is that elevating it to a headline AEO feature is not supported by the current evidence, and selling it as one displaces attention from the things that move the needle.
How an AI answer actually gets built.
When a buyer asks ChatGPT, Gemini, or Google AI Mode "what's the best small SUV under $40K in Brisbane?", the system does not consult some special LLM-readable file at the root of every dealership website. It does roughly this. One: decomposes the question into sub-queries. Two: runs those sub-queries through a search index (Google or Bing). Three: retrieves the top-ranking pages and extracts the answer. Four: cites the sources it used.
Every step rewards the work traditional SEO has always rewarded. The major AI engines are downstream of Google and Bing. They inherit the rankings. There is no separate AI internet for dealerships to optimise for.
03 / For Dealerships
Six practical moves that actually move the needle.
Less theory. More leads. Here is what an Australian dealership should actually do in 2026 to show up in AI answers and traditional search.
04 / The Playbook
What to actually do, in priority order.
Here is the work, ranked by what the empirical record actually supports. The list is unfashionable. It is also correct.
Own your Google Business Profile.
For local "near me" and city-specific queries, this is the highest-leverage AEO move a dealership can make. AI Overviews and AI Mode lean heavily on GBP data for local intent. There is no markup that replaces it.
Build real review presence.
The Trustpilot and Seer Interactive study (804,491 AI responses) found brands moving from no review profile to a minimal presence jump from 1% to 53.5% AI citation rates. For dealerships, this is Google reviews, Product Review, and CarsGuide ratings.
Write content that answers buyer questions.
Direct answer in the first 40 to 60 words. Self-contained sections. Statistics with cited sources. The Princeton GEO paper quantified it: adding statistics lifts AI visibility 30 to 41%. This is the format AI synthesises from.
Earn brand mentions in the right places.
SE Ranking found sites with 32,000+ referring domains are 3.5x more likely to be cited by ChatGPT. For dealerships, this means appearances on local news sites, Drive, CarsGuide, CarExpert, and active LinkedIn presence from your principal dealers.
Ship a fast, well-structured site.
Schema.org has partial first-party confirmation from Microsoft and Google as a signal their LLMs use for grounding. Combined with fast load times, it is the connective tissue that lets your inventory be properly understood by AI engines.
Keep inventory and content current.
85% of AI Overview citations come from content published in the last two years. 50% of Perplexity citations are less than 13 weeks old. For dealerships, freshness is built-in: new stock daily, pricing changes, finance offers, service specials.
05 / The Verdict
The work is the same. Do the work.
AI search is real. It is important. It will only grow. The work that wins it is not exotic, not new, and not sold by anyone in a hurry to badge themselves as a "GEO specialist." The work that wins it is the same work that has always won search: rank well, be trustworthy, answer the question, earn signals from people on platforms that matter.