AI Hotel Landscape

Week by week. Google is split out of what used to be counted as “other”: a link to google.com from an AI assistant is the assistant keeping the click, which is worth seeing on its own rather than filed under the vaguest available label. Metasearch (trivago, kayak) stays in other — those are third-party marketplaces, not the engine linking to itself. Google is a worldwide figure here; the other lines are scoped to this view.

One note on the other line, because it used to be the least trustworthy series here. “Other” is every link we cannot place as the hotel’s own site, an OTA, a chain page or Google — so it is partly a measure of our own coverage, not only of engine behaviour. It previously showed a band of 45–65% between 08 June and 27 July that we could not explain. That band was a bug on our side, and it is now fixed. Google AI Mode returns hotel websites as bare domains (ritzcarlton.com) rather than full URLs, and the query that derives the domain for classification expects a scheme — so 16,194 perfectly classifiable links, every one of them AI Mode, were counted as “other” purely for lacking an https://. Those URLs have been normalised, the ingest no longer stores them scheme-less, and every week has been rebuilt: other is now 3.5% across all engines and weeks, and the band is gone. Read direct and OTA as behavioural; still read a move in other as measurement first.

Weeks with no classified links are omitted rather than drawn as zero — a missing run and a genuinely zero week look identical on a chart, and only one of them is a finding. Link volume per week: 07-20 3 · 08-03 1 · 08-10 1 · 08-31 2.

Share of citations by source type, week by week, for Hong Kong. Search engine is its own bucket rather than part of metasearch, and covers Google (every regional domain), Bing and the rest. Google AI Mode routes its cited sources through google.com/searchviewer instead of linking the publisher, and those citations were landing under “meta-search” alongside trivago and kayak — which made the mix read as competitive pressure from metasearch when actual metasearch is under 1%. The dashed rule marks the week that started.

Which categories get a line is decided once over the whole window by total citations, not week by week — otherwise a category dropping out of the top N would look like it collapsed to zero. Everything below the cut is summed into “smaller categories” rather than dropped, so the lines always account for the full week. Citations per week: 04-27 108 · 05-04 336 · 05-11 7 · 05-18 36 · 05-25 191 · 06-01 441 · 06-08 362 · 06-15 149 · 06-22 157 · 06-29 242 · 07-06 265 · 07-13 124 · 07-20 138 · 07-27 127 · 08-03 109 · 08-10 104 · 08-17 117 · 08-24 55 · 08-31 102 · 09-07 55.