How AI Decides Who to Recommend

A customer sits in their truck outside your competitor’s office and asks ChatGPT, “who’s the best RV park near Batesville?” The model answers in three seconds. Somebody gets named. Somebody doesn’t.

That moment is replacing the first page of Google for a growing slice of your customers, and almost nobody in a small market is preparing for it. Here’s the part most “AI SEO” pitches skip: the model isn’t ranking you. It’s reading about you, everywhere it can, and deciding whether you’re a safe name to say out loud. Those are different problems with different fixes.

What the machine actually reads

When an AI assistant answers a local recommendation question, it works from a pile of signals most business owners have never looked at:

• Your structured data. The schema.org markup in your site’s code that says, in machine language, “this is a business, in this town, that does these things.” Most local sites have none, or a half-broken plugin’s version.
• Your own pages. Not your keywords — your answers. Models quote businesses whose pages plainly answer the questions customers ask. “Do you take walk-ins?” “What does an estimate cost?” If your site answers it in a sentence, the model can repeat the sentence.
• Everywhere else you’re mentioned. Your Google Business Profile, directories, reviews, local news. The model cross-checks. If your hours differ across three sites, you look unreliable — and unreliable names don’t get said.
• Whether it’s allowed to look. Plenty of sites accidentally block AI crawlers, or serve them a JavaScript wall with nothing behind it.

Why most local sites are invisible to it

The average small-business site was built for a human scanning a homepage. To a model it’s a thin, contradictory pamphlet: no structured data, services described in adjectives instead of facts, contact info in an image, and a copyright date from three years ago.

Then the owner wonders why ChatGPT recommends a competitor with an uglier site. The competitor’s site is uglier but legible — the model can extract who they are, where they are, and what they do without guessing. Fluency beats beauty every time in this fight. This is the same discipline behind real local SEO — say exactly what you do, where you do it, in a form machines can parse — pointed at a new reader.

What we actually do about it

This is now a service we sell, so let me be plain about what’s in it. It’s not “AI optimization” spray. It’s an evidence job:

1. Ask the models first. Before touching anything, we ask ChatGPT, Claude, and Gemini about your business and your competitors, and screenshot the answers. That’s the baseline.
2. Fix the entity. Structured data that states exactly who you are, what you do, and where — plus aligning your Google Business Profile so every source agrees.
3. Open the door. An llms.txt file and crawler access that tell AI systems what’s here and where the good stuff lives. (This site has one. Look at it.)
4. Write answers, not copy. Pages that answer the questions your customers actually ask, in plain sentences a model can quote verbatim.
5. Ask the models again. Same questions, after the work. You see the before and after, not a promise.

The honest caveat, because there always is one: nobody controls what a model says, and anyone who guarantees you a recommendation is lying. What you can control is whether the machine has clean, consistent, quotable facts about you. That’s the whole game — and right now, in most local markets, almost nobody is playing it.

The window where this is cheap and uncontested won’t stay open long. If you want the baseline audit — what the models say about you today, in writing — that’s where we start.