I asked ChatGPT where to get a custom acoustic guitar built in Montana. It named three luthiers. One of them was in the Flathead. The other two were on the other side of the state. Nothing about that answer was magic, and nothing about it was random. The AI was reading two things and making a choice.
Training data and live retrieval
Every AI answer engine uses a mix of two ingredients. Training data is what the AI learned when it was built, months or years ago. Live retrieval is what the AI reaches out and reads from the web right now, at the moment of the question. Different tools lean on these two ingredients differently. ChatGPT with browsing turned on leans heavily on live retrieval. Perplexity is almost entirely live. Google's AI Overview is a hybrid. Claude in 2026 can do both depending on the setup.
The practical effect for a small business is straightforward. If the AI is leaning on training data, an older, well-documented business has an edge. If the AI is leaning on live retrieval, whichever business has the cleanest, freshest, most structured information at this moment gets picked. Most answers are a mix. The business that wins both halves is the business that has been around AND keeps its information current.
Structured data, reviews, plain prose
Structured data first. When the AI looks at a business, it checks for a machine-readable layer on the website and the Google Business Profile. JSON-LD markup that says "I am a LocalBusiness, my hours are X, my services are Y, my address is Z." The AI treats this as authoritative because it is. Without it, the AI reads the page like a person, and its guesses are worse.
Reviews second. AI tools read reviews the way a person does. Five stars with a thin sentence is worth less than four and a half stars with twenty specific sentences about the work. The AI notices specifics. "Fixed my action perfectly" counts for more than "great service." Businesses with review patterns that read like real conversations get surfaced more often than businesses with suspiciously clean averages.
Plain prose third. If the page says what the business does in language a person would use, the AI can quote it. If the page is keyword-stuffed filler, the AI either skips it or quotes something vague. The prose discipline that makes a page trustworthy for a customer is the same discipline that makes it quotable for an AI.
Backlinks, keyword density, clever titles
The old SEO playbook mostly does not help. Backlinks, the classical metric that made an industry, barely register when an AI is picking a local business. Keyword density is worse than useless; the AI reads keyword-stuffed pages as unreliable. Clever title tags do not matter; the AI reads the page, not the metadata in the sense of the original Google ranking algorithm. None of this means the old work was wasted. It means the weights changed.
One thing worth retiring: the instinct to write a page that explains how good you are at something. The AI is not persuaded by self-description. It is persuaded by evidence: structured data that backs up the claim, reviews that confirm the claim, and plain prose that describes how the work actually gets done. A shop that says "We are the best luthier in the valley" loses to a shop that says "We build custom acoustics in the Flathead, turnaround six to eight weeks, prices start at two thousand dollars." The second one is citable.
Back to the luthier question
The three luthiers ChatGPT named had three things in common. Each had a complete Google Business Profile with categories, hours, photos, and recent activity. Each had a website with plain prose about the work (turnaround, materials, price ranges). Each had at least a dozen reviews with specific language. The luthiers the AI did not name had one or more of those pieces missing. Not bad builders; invisible ones.
If you want the AI to name your business the way it named those three, do the three habits in order. Put structured data on your website. Keep your Business Profile fresh. Write the pages like a person, about the specific work you actually do. The AI is not looking for perfection. It is looking for evidence.