How AI assistants decide which businesses to recommend

Ask an AI assistant to recommend a web shop for a small business in Texas and it answers with names. Ask for a commercial cleaner, a bakery, an accountant: same thing. Somebody gets named and somebody does not, and the difference is not luck.

Nobody outside the model companies controls those answers. But the mechanics that shape them are knowable, and most of them are things a small business can actually do something about.

Where do AI answers come from?

Two places, and the difference matters.

The first is training data. The model read a snapshot of the public web months or years ago, and some businesses are simply in its memory. You cannot buy your way into that, and you cannot rush it.

The second is live retrieval, and this is the one that moves. When you ask ChatGPT with search enabled, Perplexity, or Google with AI Overviews, the assistant runs a search, reads a handful of pages in the moment, and builds its answer from what it just read, usually with citations. Which pages it reads and quotes is decided by ordinary, boring things: whether your pages can be crawled, whether they load fast, and whether they answer the question that was asked.

That second path is why this deserves a business owner’s attention. It behaves less like magic and more like search, because underneath, it is search.

An assistant about to put a company’s name in an answer needs to be confident about the basics. Confidence comes from a few unglamorous places:

Crawlable, fast pages. If your content only appears after a pile of scripts runs, some retrieval systems never see it. Plain, fast HTML gets read.
Consistent facts. The same business name, the same services, the same service area everywhere: your site, your Google Business Profile, the directories. Machines cross-check. Contradictions read as risk, and risky answers get left out.
Structured data. JSON-LD schema is a block of code on the page that states, in a format machines parse directly, what you do, where you do it, and what it costs. It does not push you up a ranking. What it does is remove ambiguity, and ambiguity is exactly what keeps a cautious machine from naming you.
Pages shaped like answers. Assistants quote pages that plainly answer questions. A pricing page with real numbers can be cited. A page that says “solutions tailored to your unique needs” gives a machine nothing to hold on to.
Mentions beyond your own site. Reviews, local coverage, directories, a supplier’s partner page. A claim carries more weight when someone other than you makes it.

If that list sounds like plain good SEO, that is the honest secret. AI discovery is mostly the same discipline with a new and much more literal reader.

Does this actually work? Check our receipts

We sell this work, so our own site has to run it. Open branchworksoftware.com, view the page source, and you will find JSON-LD describing the business, the services, and the published prices: the $850 site base, hosting from $35 a month, all of it stated in machine-readable form. Our pricing page exists for humans first, but it is also a page an assistant can quote with confidence, because it commits to numbers. (Reading a big quote is its own skill, and the same transparency helps there too.)

None of that is a trick. It is the site saying clearly what the business is, in every language its readers use, including the machine one.

What nobody controls, said plainly

No one can promise you a spot in an AI answer. The models change, the retrieval systems change, and there is no paid placement inside the organic answer today. Anyone guaranteeing “we will get you recommended by ChatGPT” is selling weather.

What you control is being easy to verify and easy to quote. That is real, compounding work, and it improves your ordinary search results while the AI side matures.

What we actually do for $100 a month

Our AI Discovery add-on is the working version of everything above: schema markup for your services and prices, restructuring key pages so they answer questions directly, entity cleanup so your facts agree everywhere, and a monthly report showing what the major assistants say when asked about businesses like yours. It is priced at $100 a month deliberately, because the field is young and we would rather build the case studies alongside our clients than wait for a market rate to form.

Quick answers

Can you pay to appear in ChatGPT or Perplexity answers? Not in the organic answer. Ads exist and are labeled as ads. The recommendation itself has to be earned the slow way.

Does schema markup guarantee AI visibility? No. It removes ambiguity about who you are and what you offer. Machines recommend what they can verify, and removing ambiguity is the whole point.

How do I find out what AI says about my business today? Ask the assistants the way a customer would: who does your kind of work in your town? Whatever comes back is your baseline, and that baseline is the first thing we measure.

If you want your business legible to the next generation of readers, the add-on is on the menu, and the first conversation is free.