For thirty years, gallery websites have competed for the same thing: a position on a Google results page. That fight isn’t over, but a second one has opened up alongside it, and most galleries haven’t noticed yet.
When someone asks ChatGPT, Google’s AI Mode, or Perplexity a question like “which London galleries deal in 19th century marine art,” the system doesn’t return ten blue links. It returns an answer, built from sources it has decided to trust. Whether your gallery is one of those sources has nothing to do with your domain authority or your backlink count. It has everything to do with citations — and most galleries are losing this fight without knowing the fight exists.
A test case: Bond Street versus the algorithm
We ran a UK gallery through a tool that audits how AI systems answer real buyer questions, and the result was uncomfortable reading for anyone who assumes prestige is the same as visibility.
The gallery in question has been in business for a long time. It has placed work in museum collections worldwide and exhibits at TEFAF and Frieze Masters. Across sixteen UK-focused prompts that buyers would plausibly ask an AI system, it appeared in the answer twice. Brand coverage: 3.2%. On fourteen of the sixteen prompts, it didn’t exist at all.
This isn’t a story about a struggling gallery. It’s a story about a gallery doing everything right by the old rules and discovering that the old rules no longer decide who gets seen.
What was actually getting cited
The interesting part isn’t that the gallery scored badly. It’s where it scored at all. Both of its mentions came on prompts about provenance and trade accreditation — questions like which UK galleries are members of LAPADA or BADA. Everywhere else: silence.
Look at the domains the AI was actually citing across the full set of prompts, and a pattern appears. BADA’s own site and LAPADA’s own site ranked third and fourth among all cited domains — ahead of Sotheby’s. Not because either organisation is more famous than Sotheby’s. Because their websites contain exactly the kind of structured, verifiable information AI systems are built to trust: membership lists, accreditation criteria, clear statements of who is and isn’t vetted.
That’s the whole lesson in miniature. AI systems aren’t rewarding fame. They’re rewarding citability — and trade bodies, archives, and reference sources are often more citable than the galleries themselves, simply because their content is built to be looked up rather than admired.
Citations are not a vanity metric, but they’re not the whole story either
It’s worth being precise about what a citation actually means. Getting cited means an AI system pointed to your page as a source for some part of its answer. That’s valuable, but it isn’t the same as being recommended. An AI Overview can cite your “About” page for a fact about your founding year while recommending a competitor as the actual answer to “best gallery for Impressionist works.” Citation and recommendation are related but separate outcomes, and a gallery that only tracks one is only seeing half the picture.
This matters more than ever as more publishers chase citations directly. SEO researcher Lily Ray recently published findings from a study of 100 B2B “best [category]” queries in Google’s AI Overviews, and the pattern she found is sobering (more from Lily in the next post). When a business’s own self-promotional “best of” article got cited as a source, that business was left out of the actual recommendation a majority of the time — and the recommendation went instead to the established competitors the article had named. In other words, plenty of businesses are writing the case for their rivals and having Google quote them doing it. We’ll return to this in more detail in a separate piece, because it has direct implications for how galleries should and shouldn’t write comparison content. But the headline point stands on its own: chasing citations as an end in themselves can backfire if the underlying signal — who the rest of the web treats as the real authority — doesn’t support it.
What this means for a gallery, practically
None of this is abstract. It points to specific, doable work.
First, trade body membership is no longer just a credibility marker for human buyers — it’s active citation infrastructure for AI systems. If you’re a LAPADA or BADA member and that isn’t clearly, verifiably stated across your own site and corroborated elsewhere on the web, you’re leaving a citation opportunity unclaimed.
Second, the things AI systems cite tend to be specific, structured, and checkable: provenance statements, exhibition histories, artist biographies, accreditation details. A beautiful homepage with three lines of copy gives a human a feeling. It gives an AI system nothing to cite.
Third, consistency across the web matters as much as content on your own site. AI systems corroborate facts across multiple sources before trusting them. If your gallery’s founding date, specialisms, and artist roster are stated one way on your site and differently — or not at all — on Wikidata, Artsy, or in press coverage, you’re making it harder to be cited even when you have the right content.
Fourth, being cited once is not the same as staying cited. AI systems favour content that reads as current and original over content that’s stale or duplicated elsewhere on the web. A page describing an exhibition that closed two years ago, or artist copy lifted near-word-for-word from an auction house listing, is exactly the kind of source an AI system learns to stop trusting. Provenance statements and artist biographies should be revisited and refreshed, not written once and left. The gallery that updates its content regularly, in its own words, gives AI systems a reason to keep citing it instead of quietly moving on to a fresher source.
The galleries that work this out first will have a real, structural advantage over the ones who assume that being well known is the same as being well cited. Based on what we’re seeing across the UK gallery sector right now, very few have made that connection yet.
