In a 2026 analysis by SOCi, ChatGPT recommended roughly 1.2 percent of business locations it could have recommended. Across the same set, about 35.9 percent appeared in Google's local three pack.

Read that gap twice. The traditional local result is roughly thirty times more inclusive than the AI one. Assistants do not produce a ranked list with a long tail you might appear in further down. They name one or two businesses and stop.

So the honest framing is not how to rank in AI results. It is whether you can be one of the extremely small number named at all, and for most local businesses in most categories the realistic answer today is no.

That does not make the work pointless. It changes what the work is for.

What the evidence actually supports

Start with the finding that holds up best. A Yext analysis of 6.8 million citations in late 2025 found that about 86 percent of AI citations came from sources the brand directly owns or manages.

Your own website. Your own listings. The places where you control what is written.

This is the most useful fact available on the subject, and it points somewhere unglamorous. The thing that most determines whether an assistant can describe you accurately is whether accurate information exists in places you control. Not clever structuring. Not a new discipline with an acronym. Information that is present, consistent, and specific.

One caveat matters and most coverage omits it. That figure shows where citations come from, not that owning those sources causes the recommendation. Businesses that maintain good listings may simply be the kind of businesses that get recommended for other reasons. The correlation is strong and the causation is assumed.

The sources everyone quotes are less stable than they sound

You have probably read that Reddit is the most cited domain in AI answers. A Peec AI analysis of 30 million sources across ChatGPT, Google AI Mode, Gemini, Perplexity, and AI Overviews found exactly that, with YouTube and LinkedIn behind it.

Now hold that against a Semrush study tracking citations over thirteen weeks, which reported ChatGPT's Reddit citation rate falling from roughly 60 percent of responses to roughly 10 percent within weeks.

Both can be true. They describe different moments. What they jointly demonstrate is that the citation landscape moves faster than anybody can build a strategy around, and that advice written confidently about which platforms matter was describing a snapshot.

This is the strongest argument against chasing specific platforms. A tactic built on Reddit citations was excellent and then was not, within a quarter, without warning.

How selective these systems are by vertical

The research coverage is thin in exactly the categories most small businesses occupy.

There is reasonable data on finance, hospitality, and legal. There is very little rigorous per-category data on dental, veterinary, fitness, contractors, home services, or individual real estate agents. The advice circulating about those categories is extrapolation.

Where vertical data exists, it shows concentration rather than breadth. In hospitality, one analysis found Booking.com accounted for 14.5 percent of cited URLs and appeared in 95.3 percent of queries. In national finance questions, two comparison sites absorbed over 70 percent of citations between them.

The pattern is that aggregators win the general question. Individual businesses appear when the question is specific and local, which is a meaningful qualification about which questions are worth caring about.

What this means for a first year business

Four things follow, and the order is deliberate.

Be accurate in the places you control. This is where 86 percent of citations come from and it is entirely within your power. Your site stating plainly what you do, where, for whom, and at what price. Your listings carrying identical details. Hours that are correct. Services named the way customers name them rather than the way your industry does.

Be specific rather than broad. An assistant assembling an answer needs something to match against. A business describing itself as providing quality service to the local area gives it nothing. A business stating it works on pre 1970 properties in three named neighborhoods gives it something concrete, and concreteness is what gets matched. This is also the thing most of your competitors are still not doing.

Earn reviews on the platforms that serve your category. Not because reviews are an AI tactic, but because they are third party corroboration, and corroboration is what these systems weight when deciding whether to name somebody. The useful side effect is that reviews work on humans regardless of what any model does.

Check what is actually said about you. Ask several assistants the questions your customers would ask, in your area, in a fresh session. See whether you appear and who does. This takes twenty minutes and is the only measurement available to you, since none of these platforms report impressions the way a search console does.

The measurement problem nobody solves

Here is the part that should make you cautious about anybody selling this as a service.

Research from SparkToro and Gumshoe suggests that producing a stable visibility figure requires something like 60 to 100 runs per platform per query, because responses vary between sessions. A single check tells you almost nothing.

Which means a report claiming your AI visibility score improved by some percentage, based on a handful of queries, is reporting noise with a decimal point attached. The underlying variance is larger than the effect being claimed.

The honest version of measurement here is directional. Do you appear sometimes, never, or usually. That is as precise as the available method allows, and anybody offering more precision is overstating what the tools can do.

What not to buy

Schema markup specifically designed for AI systems, sold as a technical fix. The structured data recommendations circulating in this space are largely untested by any published study. Standard structured data is worth having for ordinary reasons. Exotic schema bought as an AI tactic is speculative.

A single audit presented as a strategy. The landscape shifts quarterly, the variance is high, and a point in time reading has a short shelf life.

Any service whose pitch rests on a single platform. These systems differ substantially from one another, and a conclusion drawn from one does not generalize to the others.

Anything promising a ranking. There is no ranking. There is a very short list and a long tail of businesses that are not on it.

The uncomfortable summary

For most local businesses right now, appearing in AI recommendations is not a channel you can reliably work. It is an outcome that follows from being clear, consistent, well reviewed, and specific, which are the same things that have always made a business easy to recommend.

The work is worth doing. It just is not new work, and the businesses positioned best for whatever these systems become in two years are the ones doing the ordinary things properly today rather than buying a tactic for a landscape that changed last quarter.

If you want the related piece, the version of your business that machines read covers what these systems can actually see when they look at you. That is the input. This is about what happens to it.