For the last couple of years, one of the simplest moves in AI search was to publish a “best [category]” article and name yourself first. Write “The Best Fine Art Galleries in London,” put your own gallery at the top, and there was a reasonable chance an AI system would pick up that page and use it. It worked. In plenty of cases, it still does.
But a recent piece of research suggests this tactic is starting to work against the people using it — and the mechanism behind why is worth every gallery understanding before they write their next comparison piece.
Citation and recommendation are not the same thing
SEO researcher Lily Ray recently studied 100 B2B “best [category]” queries answered by Google’s AI Overviews, tracking two outcomes separately: whether a business got cited as a source, and whether that business actually got recommended in the answer. Her finding was striking. When a business’s own self-promotional listicle was the source AI cited, that business was left out of the actual recommendation the large majority of the time. Worse, the recommendation regularly went to the established competitors the business itself had named in its own article.
Read that again, because it’s a genuinely strange outcome. A gallery writes “Top 10 Galleries for Modern British Art,” names itself first, names nine competitors below, and gets cited for the page — while the AI’s actual answer recommends the competitors. The gallery did the research, wrote the copy, and handed the win to everyone else on its own list.
Ray’s interpretation is that Google appears to have decoupled what it cites from who it recommends. Citation increasingly just means “this page contained relevant information.” Recommendation is anchored to something closer to reputation — what the rest of the web says about a business, independent of what that business says about itself. Calling yourself the best, repeatedly, in your own content, doesn’t move that second number much. If anything, a comparison article is an unusually efficient way of telling an AI system who your real competitors are.
Why this matters more for smaller, specialist businesses
There’s an important caveat in Ray’s findings: very large, well-known brands can sometimes get away with self-promotional listicles in a way that smaller, specialist businesses can’t. That’s not a comforting exception for most galleries. A handful of major auction houses and a handful of nationally known dealers might have enough independent reputation signal that a self-authored “best of” piece doesn’t dilute their position. Almost everyone else is, in effect, voting against themselves.
This is also not happening in isolation. Google made changes to how it ranks self-promotional comparison content in organic search earlier this year, with further drops following the May core update, and AI Overviews have started attaching disclaimers to categories it considers saturated with what it calls self-proclaimed experts. The direction of travel is consistent: the web’s most generic move — “let me tell you who’s best, starting with me” — is becoming less effective at exactly the moment everyone is reaching for it.
What this looks like for a gallery, specifically
Picture a gallery publishing “The Best Galleries for Impressionist Art in London,” naming itself alongside well-established rivals. If that article gets picked up as a citation, there’s a real chance the AI’s actual answer to a buyer’s question recommends the rivals on that list — the ones with more independent press coverage, more third-party mentions, more citations elsewhere on the web — rather than the gallery that wrote the piece.
Run the scenario forward. A prospective buyer asks an AI system which London gallery to approach for a 19th century landscape. The system has, somewhere in its training or retrieval, the gallery’s own article ranking itself first among ten names. But it also has years of press coverage, auction records, and trade directory listings pointing at three of the other nine galleries on that list as the more established names in the category. The AI system has no reason to trust the gallery’s self-ranking over everything else it knows. It cites the article as a useful page about the category, then recommends the names with the independent weight behind them. The gallery’s own research becomes free competitive intelligence for the AI system, and for the rivals it named.
That’s a particularly costly mistake for a gallery, because the whole point of being recognised by AI search is to be recognised as the authority on the artists and specialisms you represent — not to be a helpful directory pointing buyers toward someone else’s door.
None of this means comparison content is off the table. It means the comparison can’t be the whole strategy, and it can’t substitute for the harder work of building real, independent authority: the kind that comes from press coverage, trade body listings, third-party citations, and a body of original, specific content about the artists and work a gallery actually represents. That’s the authority an AI system recommends. A listicle, on its own, mostly just maps the competition for it.
If AI citation volume is the number a gallery is watching, it might be worth watching the wrong thing. Being recommended is the outcome that actually matters — and it’s earned somewhere other than a “best of” article about yourself.
Source: Lily Ray’s research was published on her Substack; the original data and full breakdown is available here.
