Answer engine optimization · Original data
How many local businesses does AI actually cite?
A number keeps making the rounds that sounds like good news for anyone running a local shop: across almost two million AI citations, 93% of domains referenced were standalone business websites rather than directories or aggregators. Real companies, named by name.
On the surface, that reads like the discoverability problem is handled. If AI overwhelmingly points to actual businesses, your chances look solid. So why does typing your name into ChatGPT still feel like nobody's listening?
I dug into the study generating that figure, then ran it against our own records. The 93% is genuine. It's also masking exactly why so many owners still feel invisible.
A study built on 1.97 million citations
The research comes from BrightLocal, and the scope deserves a moment: 60,970 live queries, covering 1,355 business locations across nine markets, yielding nearly 1.97 million citations spread over 116,670 distinct domains. They manually flagged 415 directory sites that appeared somewhere in that stack. Of all 116,670 domains referenced, 93% proved to be individual business sites rather than those 415 directories.
That holds up, and it's significant. AI models aren't inherently rigged toward directories the way plenty of GEO guidance suggests. Most citations genuinely do land on someone's own property.
Where the number stops telling the whole story
Here's where it breaks, and it's a split between two things both labeled "citations." Ninety-three percent of unique domains are businesses. That does not mean ninety-three percent of citation volume flows to businesses. Track where the actual mentions pile up instead of counting domain names, and the picture inverts. Google's own listings alone grabbed 28.63% of every citation in the entire dataset, one domain, nearly three in ten. Fold in Yelp at 9.53% and the next eight entries, and the top ten domains combined pull in roughly 44% of all 1.97 million citations, and that's from just ten sites.
So the real finding: thousands of individual businesses do get cited, each one barely, while a tight circle of directory-type domains gets cited relentlessly. Businesses hold ninety-three percent of the slots, and ten directories still take nearly half of everything that gets cited.
The same shape in our own smaller dataset
I needed to see if that same architecture appears in our own data, at far smaller scale, before I'd stake anything on it. We run a probe across ten local service categories and five cities that's logged 1,423 citations across 927 unique domains so far. Of those 927, only 118 are known directories, about 12.7%, a smaller ratio than BrightLocal's 415-out-of-116,670. Yet those 118 directory domains still command 452 of the 1,423 citations, which works out to roughly 32% of total volume. Reddit alone surfaces 63 times, more than any other single domain we've tracked. Our top ten domains combined claim 207 of the 1,423 citations, pushing 15% of everything, from ten sites out of 927.
The dataset is different and the scale is smaller, yet the shape holds. A small cluster of domains does disproportionately heavy lifting, however many individual businesses technically populate the list.
What this means for your own visibility
What this shifts in how you approach your own visibility is fairly precise. Landing among the 93% doesn't put you on equal ground with the businesses already surfacing. It makes you one name among thousands dividing the leftovers, while a handful of directories keep seizing the general queries regardless of your site work. We've written before about why that split tracks with question specificity: broad queries route to directories almost mechanically because no single business page answers "best med spa in the city," while narrow ones create direct openings for you. That mechanism explains precisely why volume pools the way it does here. The fix isn't trying to out-directory the directories. It's making sure the narrow version of the question, the one your actual customer types, with your specific service attached, finds a page on your site built to answer it.
The five-minute self-check: pose the broad category-plus-city question, then pose the narrow one your customer would actually type, service included. If you're in the 93%, you'll typically appear on the second before the first.