SEO vs GEO, Part 2: What Actually Gets You Cited

Two collision shops, ten miles apart. Same size. Both have a clean modern website, both load fast, both have a services page and a contact form. Ask ChatGPT who does quality frame work in their area and one of them gets named. The other does not exist.

Their websites are not meaningfully different. So what is?

Part 1 covered what changed. This one is about the machinery — and about which parts of the standard AI visibility checklist actually hold up when you go looking for evidence.

Fair warning: some of what follows is inconvenient for people who sell this work. Us included.

First, how the answer gets built

A generative engine does not keep a ranked list of shops waiting to be served. When someone asks a question, it goes and gets source material, then writes an answer from what it found.

Three things have to happen for you to be in that answer:

1. You have to be found during retrieval — the engine pulls candidate sources
2. You have to be understood — the model has to know what you are, where you are, what you do
3. You have to be trusted enough to name — it has to be willing to put your name in an answer it is accountable for

Most of the technical work sold as "AI optimization" targets step two. The evidence says steps one and three are where the results actually live.

The uncomfortable part: schema markup

Structured data — JSON-LD, LocalBusiness schema, the code that tells a machine "this is a body shop, here is the address, here are the hours" — is the centerpiece of nearly every AI visibility package on the market. Ours included.

Ahrefs ran the largest test of this so far. They tracked 1,885 pages that added JSON-LD schema between August 2025 and March 2026, matched them against 4,000 control pages, and measured citations on Google AI Overviews, Google AI Mode and ChatGPT before and after.

The result:

• Google AI Overviews: −4.6% — a small but real decline
• Google AI Mode: +2.4% — statistically indistinguishable from zero
• ChatGPT: +2.2% — statistically indistinguishable from zero

Their conclusion, plainly: adding schema produced no major uplift in citations on any platform. Read the Ahrefs study

Here is the trap that keeps this myth alive. Pages that do get cited by AI are roughly three times more likely to carry JSON-LD. Every vendor quotes that number. But it is a correlation, not a cause — well-run sites tend to have schema and tend to get cited, because both come from being a serious operation. Adding the markup to a page that nobody trusts does not import the trust.

Does that mean skip it? No. Schema still drives rich results in ordinary search, still feeds voice assistants, still helps knowledge graphs understand what entity you are. It is cheap, it is permanent, and it is table stakes. We still ship it on every build.

It is just not the lever. Anyone selling schema as the AI visibility solution is selling you a seatbelt as an engine.

The other uncomfortable part: llms.txt

An llms.txt file is a plain-text summary at the root of your site, written for AI models. The idea is sound and the format is well designed.

In June 2025, Google's John Mueller said this:

"FWIW no AI system currently uses llms.txt."

He added that you can see it in server logs — the models fetch your actual pages, and they do not fetch that file. As of today, no major AI platform has confirmed using it.

We publish one on our own site anyway. We are telling you that plainly because you deserve to know which parts of this are proven and which are a bet. It costs an hour, it breaks nothing, and if adoption comes it is already there. That is the honest case for it. The dishonest case — the one you will hear on a sales call — is that it is how you get into ChatGPT. It is not.

So what actually correlates with getting cited

Here is where it gets interesting for a local operator, because the real drivers are things you can influence without a developer.

Being mentioned on pages you do not own. Ahrefs looked at 75,000 brands and found brand mentions correlated at 0.664 with AI Overview visibility. Backlinks — the thing SEO spent twenty years chasing — correlated at 0.218. Mentions beat links by roughly three to one.

Earned sources dominate the citation pool. Muck Rack analyzed over one million cited links in December 2025 and found the bulk came from earned media, journalism, academic and government sources, and third-party content. Press releases accounted for just 6%.

Distribution multiplies your own content. In one controlled test, brand-owned articles were cited 8% of the time on their own. The same articles, once distributed across third-party outlets, were cited 34% of the time.

A research team studying this in September 2025 put it bluntly: AI search shows a systematic and overwhelming bias toward earned media over brand-owned content — a sharper skew than traditional Google results.

All of these are correlational, and the researchers say so. Treat them as a direction, not a formula.

Which brings us back to the two shops

The one that gets named is almost never the one with better code. It is the one the internet has more to say about.

Their Google Business Profile is complete and matches their website exactly — same name, same address formatting, same phone, same hours. They have real reviews with real detail in them. They show up in a local news piece, a chamber listing, a supplier's certified-shop directory, a Facebook group where somebody asked for a recommendation and four people answered.

None of that is on their website. All of it is what the machine reads to decide whether naming them is safe.

That is entity clarity — the machine's confidence that it knows who you are, where you are, and that independent sources agree. When your name, address and phone are formatted three different ways across six directories, you have not just made a tidiness error. You have made yourself statistically ambiguous, and an engine writing an answer it has to stand behind will quietly choose the shop it is more certain about.

The honest summary

• Technical setup is the floor. Schema, fast pages, clean structure, mobile. Do them, expect nothing dramatic, move on.
• Entity clarity is the multiplier. One name, one address format, one phone, everywhere. Boring, unglamorous, and it compounds.
• Third-party presence is the lever. What other sites say about you outweighs what your site says about itself — by a lot.

The shop owner reading this has better odds here than the agency does, because the winning inputs are reputation, consistency and community presence. That is the stuff a good operator already has. It just has not been made legible to a machine yet.

Coming in Part 3

The playbook. Exact sequence for a local service business: what to fix in week one, what to ignore entirely, how to check whether your visibility actually moved, and how to spot a vendor inventing an AI problem to sell you a solution.

"The best thing about this is also the most annoying thing about it. You cannot buy your way in with a code snippet. You get cited for being genuinely well known in your area — which is exactly what you have been working on for twenty years anyway." — Jamie Weaver, Founder

Curious where you actually stand? Our free AI visibility audit runs your business through ChatGPT and Google and tells you what comes back. No cost, no pitch, yours either way.