Four AI assistants named 30 boutique hotels in Miami. Exactly one appeared on all four lists.
Four AI assistants named 30 boutique hotels in Miami. Exactly one appeared on all four lists. On a second query about independent hotels near a named landmark, 11 properties came back and none were unanimous. The concentration operators have been warned about did not appear. Fragmentation did, and it means there is no single set of AI results to optimise for.
We ran the same two queries through four AI assistants on 9 September 2026: "best boutique hotels in Miami" and "independent hotels near freedom tower". Across the first query the four assistants named 30 distinct properties between them, and only Mayfair House Hotel and Garden appeared on every list. The Betsy appeared on three.
This piece is for independent hoteliers who have read that a small share of hotels appears when AI recommends a place to stay. In one city, on one day, across four assistants, we found something more awkward than concentration, and the reason matters more than the numbers.
Agreement between assistants ran at one property in 30 on the broad query
The four assistants were Claude, Google Gemini, Microsoft Copilot and a fourth general-purpose assistant. Each got the identical prompt with no follow-up steering.
The lists barely overlapped. One assistant returned four properties, another returned 21. Claude named Esmé Miami Beach and The Moore, and no other assistant mentioned either. The fourth assistant named Faena, The Vagabond and Arlo Wynwood, and no other assistant mentioned any of them. Copilot and Gemini both led with The Standard Spa and The Savoy, neither of which appeared on the other two lists at all.
Twenty-four of 30 properties came from a single assistant. For an operator, that is the number that counts: appearing in one assistant's answer told us almost nothing about appearing in another's.
The narrower query fragmented further, with zero properties named by all four
"Independent hotels near freedom tower" returned 11 distinct properties and no unanimous pick. Eurostars Langford came back from three of the four. Nothing came back from all four.
The four assistants also disagreed about which properties qualified. Claude and Gemini both recommended The Elser and YOTEL Miami. Copilot named the same two properties specifically to rule them out, telling us they do not fit an independent-hotel requirement. The fourth assistant recommended YOTEL as an independent brand, and in the same answer recommended Kimpton EPIC while noting Kimpton is part of InterContinental Hotels Group.
One assistant contradicted itself across consecutive turns. It listed Langford Hotel under Brickell and Downtown in its answer to the boutique query, then left it out of its answer to the independent-hotels query one message later, recommending RS Boutique Suites and YOTEL instead.
Different retrieval stacks, not different opinions, explain most of the gap
The four answers were not four readings of the same evidence. They were four different data pipelines.
Claude ran a live places lookup and returned Google ratings and review counts with each property. Gemini answered from general knowledge, then attached a Google Hotels module with live nightly rates. The fourth assistant cited 43 sources and worked from booking-site aggregate scores, quoting the Langford at 8.8 out of 10 where the other three showed 4.1 out of 5. Copilot returned a map card and a ranked shortlist.
Rating figures diverged because the sources diverged. Distance figures diverged for the same reason: on the Langford, Copilot said about half a mile from the Freedom Tower and Gemini said roughly a ten-minute walk, both consistent with the property's actual position at 121 SE 1st Street. The fourth assistant put it at 0.1 to 0.2 miles, a two to four minute walk, and used that proximity as a reason to recommend it. The real distance is closer to half a mile.
Nothing an operator does to their own website changes which pipeline an assistant queries. That is a different problem from the one travellers using AI to plan trips present, and it has a different fix.
What the fragmentation does not rule out
Our test is one city, one day, two queries, four assistants. It carries the same limits as the concentration findings it complicates, and it cannot show a trend.
Two readings survive it. The first is that concentration is real at the top and we measured the tail: Mayfair House and The Betsy did recur, and a larger sample might show a stable handful recurring across many markets while everything below them churns. The second is that fragmentation is the durable state, because the assistants are querying structurally different sources and will keep doing so.
The test cannot separate those. What it does settle is that on the day we ran it, in one large market, a property's presence in one assistant's answer was close to uninformative about its presence in another's.
Some operators are already treating each assistant as a separate channel and checking them separately rather than looking for one ranking to influence. Others read the same evidence as a reason to wait, on the grounds that pipelines this unstable are not yet worth building process around. The eleven properties that came back from our landmark query included one, Breezy Stays Luxury Residences, that surfaced only inside Gemini's Google Hotels module and in no assistant's written answer at all.
Guneet Lamba does content and SEO at PriceLabs, where she writes about dynamic pricing, revenue management, and how operators actually run their portfolios. Her work appears across the PriceLabs blog and Rental Scale-Up.
