How AI Chooses Which Businesses to Recommend

There's no docket, no appeals process, no visible position you can check the way you can with a Google ranking. When someone asks ChatGPT "who's a good plumber near me," the model isn't running a scoring algorithm against a list of candidates. It's recalling what it's read, half-remembered specifics, general impressions, a rough sense of who gets talked about credibly versus who doesn't, and turning that into an answer in real time.

That distinction matters because it changes what you're actually trying to influence. You're not trying to win a ranking algorithm. You're trying to be the kind of business that gets mentioned in passing, by name, with specifics attached, across enough of the internet that a language model has actually encountered you before.

Large language models are trained on enormous amounts of text, and increasingly, tools like ChatGPT and Perplexity also do live retrieval, meaning they search the web in the moment and read pages to help answer a question. Either way, the mechanism is the same. Your business has to show up in text that a model can read and understand.

This sounds obvious, but most small businesses fail it in a specific way. Their website says who they are ("Sunbelt Dental Group") but never says what they actually do in plain sentences a model could lift and paraphrase. A page full of stock photography and a tagline like "Your Smile, Our Passion" gives an AI model almost nothing to work with. A page that says "We place porcelain veneers for patients in Dallas who want to fix chipped or discolored front teeth, usually in two visits over three weeks" gives it something concrete to repeat.

AI models seem to weigh what other sources say about you more heavily than what you say about yourself. That tracks with how humans work too. If a friend tells you a restaurant is great, that lands differently than the restaurant's own ad.

This is backed up by real research, not just a hunch. The original study on Generative Engine Optimization, from Princeton, Georgia Tech, and IIT Delhi, found that citing outside sources and adding quotations to content produced some of the largest visibility gains, up to 40 percent, versus content only tactics like keyword stuffing (arXiv 2311.09735, "GEO: Generative Engine Optimization").

In plain terms, reviews, directory listings, local news mentions, and forum threads where real people describe your business in their own words are doing more heavy lifting for AI visibility than another paragraph of self written marketing copy. This is also why review volume and freshness matter. BrightLocal's long running Local Consumer Review Survey has tracked for years how heavily consumers rely on reviews to judge local businesses, and that same trust signal appears to carry over into what AI models treat as credible (BrightLocal, Local Consumer Review Survey).

Google's own explanation of how it built its AI powered search features is a useful window into how these systems think, even though it's describing Google's product specifically. In its announcement of generative AI in Search, Google described the goal as helping people get to a useful answer faster, with links available if they want to dig deeper (Google, "Generative AI in Search: Let Google do the searching for you"). That word, faster, is the tell. The model is trying to compress information down to something clean and useful. Pages that are already clean and useful are easier to compress. Pages that hide the answer under a wall of branding are harder.

This means your FAQ page answering "how much do veneers cost in Dallas" in two direct sentences is more useful to an AI model than your homepage's hero section. It means a clearly labeled service page beats a beautifully designed one that never states the service in words.

If your business is described one way on your website, another way on Yelp, and a third way on your Google Business Profile, that's friction. Not because any one of those is wrong, but because a model synthesizing an answer works better when the signals agree. Name, location, services, and hours that match everywhere reduce the odds a model quietly skips you in favor of a competitor whose information is cleaner.

One of OMG's dental clients in the Dallas area wanted to know whether they were showing up when people asked ChatGPT questions like "who does veneers near me in Dallas." At the time, the honest answer was no. Their site had general cosmetic dentistry language but nothing specific enough for a model to grab onto, and their reviews existed but weren't reflected anywhere on their own pages. When we first pulled up the practice's veneers page to run our test prompts, the phone number listed at the top was actually a picture of the phone number, not text, so no AI crawler reading the page could extract it at all.

Working with OMG, the practice rewrote its veneers page to plainly state what they do, who it's for, and roughly what to expect (timeline, visit count, the kind of cases they handle), added FAQ content answering the exact phrasing patients actually type into AI tools, and cleaned up their schema markup and Google Business Profile so the same facts appeared consistently everywhere. The rewrite opened with: "We place porcelain veneers for adults who want straighter, whiter teeth without braces. Most patients need two visits, and a full set typically runs $800 to $2,500 per tooth depending on materials," replacing a paragraph that never stated a price, a timeline, or the actual procedure by name. About three weeks later, they started getting new patients booking veneer consultations who said, unprompted, that they'd found the practice through a ChatGPT answer. That's a small sample, one practice, one procedure, but it's a real result we could trace directly back to specific pages.

AI tools have some sensitivity to whether a business seems active versus dormant. A site with a blog last updated in 2019, no recent reviews, and no fresh mentions anywhere reads, to a model, a little like a business that might not exist anymore. Regular, real content (new FAQs, service updates, even LinkedIn posts) signals you're an operating, current business worth recommending today, not a relic.

If you're trying to become the kind of business AI tools mention, the priorities in order are: say clearly and specifically what you do, get real third party sources (reviews, mentions, local press) talking about you in matching detail, structure your own pages so the answer is stated plainly near the top, and keep your information consistent and current across every place it appears. None of that is a trick. It's closer to just being a well documented, clearly described, real business, which is exactly why it's been so easy for a lot of businesses to accidentally get this wrong.