Of 719 Charleston businesses with completed audits, only 81 were ever named by an AI model when a buyer asked for a recommendation. The other 638 got zero. That is roughly 9 in 10 local businesses that, as far as the AI is concerned, do not exist.
Ask ChatGPT for the best restaurant in Charleston. It hands you five names. Ask again tomorrow. Mostly the same five.
That is not a phone book. A phone book lists everyone. This is a shortlist, and the shortlist is short on purpose. When AI becomes the way people find a plumber, a realtor, a place for dinner, the businesses on that shortlist win by default and everyone else disappears. Not because they are worse. Because they were never named.
I wanted to know who is on Charleston's shortlist and who is not. So I stopped guessing and measured it.
There are two different questions people confuse constantly.
The first is "how often does AI mention my name." That is a mention count. Plenty of tools sell it. A business can rack up mentions and still never actually get recommended to a buyer.
The second is "when a real person asks AI for a recommendation, does it pick you." That is selection. That is the only thing that pays a Charleston business's rent.
ARO Index measures selection. I ask each model the questions a real customer would type, with no business name planted in the prompt, and I record who the model chooses on its own. Cold. If the model names you unprompted, that counts. If it does not, it does not. I never sell you a mention percentage and call it visibility. Selection is the number. Everything else is noise wearing a lab coat.
I ran three query variants per category, across four models (ChatGPT, Claude, Gemini, and Perplexity), for 43 service categories in Charleston. That is up to twelve independent chances for any single business to get named.
1,621 clean recommendations came back. They pointed at 1,052 distinct businesses and brands. But of the 719 audited Charleston businesses in this study, only 81 ever surfaced. That is a selection rate of about 11 percent.
The rest of the named slots went to national chains, aggregator sites, and a handful of local names the models already trust. If you are a small business owner in Charleston reading this, the odds are you are in the 638. That is not an insult. It is a starting line.
Selection is not random. In the categories where the models agree, they agree hard. Here is who the AI kept choosing, and how many of the four models named them.
Seven restaurants were each named by 3 of 4 models:
One business in the entire study was named by all four models: Roper St. Francis. Everyone else topped out at 3 of 4. The AI has favorites, but the favorites are almost never unanimous. That gap is where the next winners get made.
I am going to say this plainly, because research that hides its limits is just marketing with footnotes.
Observed selection tells you what the AI does. It does not tell you every reason why. A business can get selected because it has a strong Reddit presence, glowing reviews, press coverage, or a site the models can actually read. My cold queries capture the outcome, not the full mechanism.
The audited set is also a sample, not a census. 719 businesses completed the current audit methodology in time for this report. Charleston has more businesses than that, and I say so here instead of letting you find out later. The corpus of what the models actually recommended, all 1,621 selections, is complete and unfiltered.
One clean line to hold onto: observed selection measures what AI does. Site diagnostics measure one of the many reasons why. I never conflate the two.
If 638 businesses got left off, the obvious question is what the 81 did differently.
Short version: the models recommend what they can find, trust, and read. A business that is easy to verify, referenced across the open web, and structured so a machine can understand it gets picked. A business that is invisible to a crawler, thin on third-party signals, or impossible to parse gets skipped, no matter how good it actually is.
The uncomfortable part is that quality and selection are not the same thing. Some of Charleston's best businesses are in the 638. The AI is not judging the food. It is judging the footprint.
For anyone who wants to cite this, here is exactly what I did.
Every number in this report was verified directly against the source database immediately before publication. These are the audited figures, not estimates.
This is one market and one month. The point of ARO Index is that it does not stay a snapshot.
I re-run the market on a monthly cadence, so selection stability becomes measurable over time. My working expectation, to be tested against the data, is that most of any category's top five holds month over month, with the churn concentrated among emerging businesses fighting their way onto the list. That churn is the story worth watching.
Charleston is first. The same engine replicates to any market with locked query banks per category. If AI is becoming the front door to local discovery, someone has to publish the standard for who gets let in. That is the job I gave ARO Index.
If your business is in the 638, that is not a verdict. It is a to-do list.