UAE pilot · London luxury hotels · August 2026
The GCC AI Discovery Gap
When the same travel question is asked of AI answer engines in English versus Arabic, the hotels recommended barely change. The evidence used to justify those recommendations changes a great deal.
The Question
Ask an AI answer engine the same travel question in English and in Arabic. Do you get the same answer?
Gulf travellers increasingly research luxury hotels through AI answer engines rather than a page of search results, and they do it in both languages depending on device, browser and personal habit. Most brands assume that as long as the AI recommends them, the language of the question doesn’t matter.
This pilot tests that assumption directly, on one real, independently verifiable dataset: 1,426 AI executions, run against the identical set of 20 hotel-shopping prompts, once in English and once in Arabic.
Why This Matters Now
AI visibility tracking is becoming standard. Bilingual AI visibility tracking is not.
Brands now track how often ChatGPT, Gemini, Copilot and Google’s AI Overviews recommend them the way they once tracked search rank. Platforms built to do exactly that, including the one behind this data, exist because that question now has commercial weight.
What most of that tracking still treats as the finish line is a single number: did the AI recommend us. Almost none of it separates being recommended from being the reason given for the recommendation, and almost none of it asks whether that reason changes when the question is asked in a different language. In a region where the same customer might query in English on one device and Arabic on another, that gap is not academic.
The UAE Pilot
One category, one audience frame, two languages, five engines
Scoped deliberately narrow so the comparison would be clean: London luxury hotels, framed for a UAE/Gulf traveller audience, tested against the same 20 prompts written independently (not machine-translated) in English and Arabic.
The two language samples were captured in different, non-overlapping weeks. That timing gap is a possible confound and is carried through every finding below rather than smoothed over.
Headline Finding
Sources diverge far more than recommendations do, on every measure tested.
The 60% and 28% figures aren’t two estimates of the same thing, they answer different questions. 60% asks: weighting every citation by how often it happens, how similar do the two source mixes look overall? A few heavily-cited sites can carry that number. 28% asks something simpler: leaving frequency aside, how many of the actual distinct websites used show up in both languages? Answered that way, most sites turn out to be one-language-only, which is why this number is lower. See the method note below for exactly how each is worked out.
The Recommendation Shift
The hotel list barely moves. The rank order does.
Of the 14 hotels that place in either language’s top 10, six sit in both: the Mandarin Oriental Hyde Park, the Dorchester, the Savoy, Claridge’s, the Ritz London and Rosewood London. The other eight split evenly, four hotels that only make the English top 10, four that only make the Arabic one. Names were consolidated across scripts and spelling variants before this list was built, specifically because raw exports understate hotels like the Hyatt Regency Churchill, which appears under three different spellings in the source data.
The most dramatic mover in either direction: the Hyatt Regency London – The Churchill, the 9th most-recommended hotel in Arabic-language answers, and effectively invisible in English ones (rank 87). The Carlton Tower Jumeirah moves the other way, 8th in English, 21st in Arabic. Everything in between reshuffles by a handful of places rather than disappearing outright, which is exactly why the recommendation-similarity score still reads as high as it does.
Placing in both top 10s can still hide a large gap underneath. Claridge’s is one of the six hotels that makes both lists, but Rankscale’s own per-brand detection data, tracked independently of our top-10 reconstruction, has it far stronger in Arabic: a 10.6% detection rate and an average rank of 4.0 in English, against a 14.4% detection rate and an average rank of 2.4 in Arabic, a visibility score of 12.8 versus 8.2. Same hotel, same top-10 placement in both languages, a materially different weight behind it. “We’re on the shortlist in both languages” and “we carry equal weight in both languages” are not the same claim, and only one of them is usually being reported.
Source Shift, English
English answers lean on the brand’s own voice and Western consumer platforms
Ranked by raw citation-count gap across all 20 prompts, one week of runs. These are the domains English-language AI answers cite disproportionately more than Arabic ones do for the identical questions.
reddit.com, brand-owned sites like thesavoylondon.com, and UK consumer/travel press (thenationalnews.com, yalla-london.com, expedia.co.uk, kayak.ae) make up most of this list. This is the ecosystem a typical English-language digital PR or SEO program is already built to influence.
Source Shift, Arabic
Arabic answers lean on regional booking platforms, travel agents and social
The same ranking, run the other direction. These are the domains Arabic-language AI answers cite disproportionately more than English ones do, for the identical questions.
Almost no brand-owned domains appear on this list. almatar.com, almosafer.com, ae.almosafer.com, safravip.com and arab-tours.com are regional OTAs and travel agents; instagram.com is social; bing.com is a search pathway inside Copilot. The hotels themselves are barely citing their own websites in Arabic-language answers, third parties are speaking for them instead.
A Necessary Distinction
Being recommended is not the same as being the source of the recommendation
Most AI-visibility tracking stops at stage two below. The gap that matters for a brand’s own narrative control opens up at stage four.
A brand can be recommended in Arabic-language answers while having almost no say in what is said about it there, because the AI is citing a regional OTA, a travel agent, or a social post, not the brand’s own site or press. Recommendation parity can coexist with a near-total loss of narrative control, and the two require different fixes.
Beyond Hotels
Where else this pattern probably shows up
This pilot only tested one category. Nothing below is measured, it is a hypothesis worth the next test, listed here so it doesn’t get lost.
Checklist
What to do with this, starting Monday
- 1Audit Arabic-language AI citations separately from English ones. Don’t assume the source list overlaps just because the recommendation does. In this pilot it didn’t, by a wide margin.
- 2Localize owned content into Arabic, not just the recommendation-facing summary. An English-only site and press library won’t surface as a citation in Arabic answers even when the brand still gets recommended.
- 3Build presence on the platforms that actually gate Arabic answers. In this category that meant regional OTAs and travel agents (Almosafer, Safar VIP) and Instagram, not Western travel press.
- 4Consolidate your brand name across scripts and spellings before trusting any AI-visibility number. Unmerged duplicates undercount a brand and can silently flip its apparent rank against competitors.
- 5Re-run this comparison periodically. Source ecosystems move faster than recommendation lists. A quarterly check is more useful here than a one-off audit.
Being recommended is not the same as being known.
In the Gulf’s bilingual AI search environment, a brand can win the recommendation in both languages and still lose the narrative in one of them. The fix isn’t a better English press strategy. It’s a separate Arabic one.
UAE pilot · August 2026 · Produced in partnership with Rankscale · Analysis: Marta Warren