Does AI favor big brands over small businesses?
Updated August 11, 2026: I corrected the evidence grades in this piece and added the counter-evidence that has been published since. The original overstated two things, and I explain exactly what changed in will AI search stop citing small websites. The mechanism below stands; the certainty I attached to some numbers did not.
Less than most owners assume, and not in the way either the hype or the panic claims. Researchers who pulled and counted actual citations from live AI answers found that generative search structurally redistributes citation away from popular incumbents and toward niche, topically matched sources. The big brand's name recognition, the thing that wins a Google results page, buys noticeably less in an AI answer. I want to be precise about what that means and what it doesn't, because the honest version of this claim is more useful than the flattering one, and I'd rather you act on the real shape of it than on a story I made simpler than it is.
Start with the single most valuable position in any kind of search result: the top citation slot. In classic organic search, the top 1,000 domains by popularity take that slot 52.7% of the time. That's the world you already know. Authority compounds, the biggest sites have been at this the longest, and Google's ranking system rewards exactly that accumulated trust. Now look at AI Overviews. The same top-1,000 domains take the top slot only 40.0% of the time. Move to Gemini and it drops again, to 32.6%. The most contested, most valuable position in the whole system is the one that has opened up the most. Fifteen years of domain authority buys a national chain less in an AI answer than it buys on a results page.
It gets more specific than that. Across roughly 19,000 citations pulled from live AI answers, the top 25 domains, the household names with the biggest budgets and the oldest backlink profiles, account for only 23.8% of all citations. And 59.1% of the domains that get cited at all get cited exactly once. Read that twice. Most sources an AI answer pulls from aren't repeat winners with permanent standing. They're one-time, topically precise matches for one specific question. That's not a popularity contest, and it never was one under the hood. It's a matching problem, and a small, focused business with the right passage on the right question wins that match as often as anyone with a bigger name.
I want to grade this the way I'd want it graded to me, and this is the paragraph I got wrong the first time. The redistribution numbers, the 52.7 versus 40.0 versus 32.6, the 23.8%, the 59.1%, come from real citation pulls, but the studies behind them are preprints, so the right grade is directional, not hard. And the direction itself is now contested: a larger 2026 study across 24,000 queries found AI search surfacing fewer long-tail sources than traditional search, not more. What survives both readings, and what I will stake a claim on, is the decoupling: the selection is no longer the ranking, and who benefits varies by engine, question, and period. Also unsettled is how long any of this holds. The researchers who found the asymmetry flagged it themselves as a credibility risk the engines are likely to correct over time. One engine, Yandex, is already grounding its answers on top organic results, which is exactly the kind of correction that would close the gap. So this is a window, not a law of how AI search works permanently. Compounding into it now is a good bet. Assuming it stays open forever is not one I'd make for you.
There's a mechanism worth understanding, because it explains why the redistribution is real rather than a fluke of one dataset. These systems aren't ranking pages the way a search engine ranks a results page. They're assembling evidence for an answer they're about to write, and they go looking for whatever passage sits closest in meaning to the specific question asked. A national brand's homepage doesn't win that contest just by existing, the way it might win a branded search. A specific paragraph, from a specific business, answering the specific sub-question a real customer asked, does. Popularity isn't a tiebreaker in that matching step. It's mostly just not part of the calculation. I wrote more about why that mechanism favors specificity over size in why AI recommends national chains by default, because the redistribution I'm describing here and the reason chains still win by default are two sides of the same evidence.
The clock matters as much as the mechanism. Organic authority moves in quarters. A national chain's rank on Google took years to build and takes months to meaningfully erode. AI citation doesn't move on that schedule. On one engine, the set of cited sources was measured turning over on a cadence of days. I will not hang the precise figure on the whole category, because it comes from one commercial dataset on one engine, but the shape is corroborated. Which means the usual barrier protecting an incumbent, the fact that it takes forever to dislodge them, mostly doesn't apply on this surface. A small business can show up in the cited set this week with a genuinely good answer, even if the same content couldn't touch page one of Google.
There's a wrinkle worth naming if you're not a true one-location operation but you're not a national chain either. The popularity penalty doesn't fall hardest on the smallest, least-known sites. It peaks somewhere in the middle, roughly the tier of businesses big enough to look established but not so dominant they get treated as the default authority. If you've spent years building real local recognition and assumed that puts you in a safer position than a brand-new competitor, the evidence points the other way. That mid tier is where AI systems seem to discount incumbency the hardest right now.
None of this means small automatically beats big. It means the size advantage that used to be close to decisive isn't doing the same work on this particular surface, today, measured honestly. What still decides the outcome is whether your business has built the specific, accurate, well-matched record an answer engine can actually use, which is a different kind of homework than buying the biggest marketing budget in your category. I go through what that record actually needs to contain in what proof does AI look for before recommending a local business.
So if you run a small or mid-sized business and you've been assuming the AI era just hands your biggest competitor one more weapon, the measured evidence says you have that backwards, at least for now. The moat that protects a big name in organic search barely exists on the answer layer. What replaces it isn't size. It's whether you've done the specific, provable work of being the clearest answer to the question someone actually asked. That's a fight a small business can win, and right now the data says it's winning it more often than the old rules would have predicted.
This essay is part of The Discovery Layer: How Businesses Get Found in the Age of AI, a book I'm writing in public. Each essay becomes a chapter, and the book expands them with the full evidence. Get on the list and I'll tell you when it ships.