AI recommends many of the same London luxury hotels in English and Arabic. But the evidence behind those recommendations is surprisingly different.
A few weeks ago, I argued that luxury brands cannot afford to treat Arabic AI discovery as an afterthought, so, I teamed up with Rankscale AI to test it. We ran 20 London hotel queries designed around a UAE and Gulf travel audience, in English and Arabic, across five AI environments, generating more than 1,400 executions. The headline finding sounds reassuring: the overall hotel shortlists were 79% similar between languages. But, underneath those recommendations, only 28% of the non-platform sources cited overlapped.
Same brands. Very different evidence.
The shortlist barely moves, but the evidence does
Mandarin Oriental Hyde Park ranked first in both languages, while familiar names including The Dorchester, The Savoy, Claridge’s, The Ritz London and Rosewood London appeared in both top tens.
But, individual performance still moved significantly.
Hyatt Regency London – The Churchill went from effectively invisible in English, at rank 87, to ninth in Arabic. Claridge’s offers an even more interesting example. It ranked fourth in both languages, yet Rankscale detected it much more frequently in Arabic: a 14.4% detection rate versus 10.6% in English. The ranking looked identical. The strength underneath it was not. For communications teams, that matters.
A good ranking can mask a thin evidence base. If Arabic-language mentions, coverage and corroboration are weaker than their English equivalents, the position itself does not tell you how resilient that visibility is.
But, the bigger difference appeared in the citations.
Same recommendation – different evidence
Across non-platform domains cited by the AI systems, only 28% appeared in both the English and Arabic runs. The English citation landscape included sources such as Reddit, hotel websites, TimeOut, Tripadvisor, Visit London, Condé Nast Traveller, Michelin Guide and Agoda.
Arabic answers drew much more heavily on a different mix. Regional travel platforms became particularly prominent. Almatar generated 157 citation occurrences. Almosafer and its UAE domain added another 140. Instagram appeared 94 times. There were also less predictable sources. SafraVIP appeared repeatedly. So did a French cultural website with an article about Ritz pricing.
That does not mean these sources are inherently more authoritative than established travel media. It means they were accessible, retrievable and useful enough for AI systems to keep surfacing them.
And that changes the communications brief.
Your Arabic media strategy cannot be your English one translated
If Arabic AI answers are being constructed from a substantially different source ecosystem, then an English-first media list translated into Arabic is optimising for the wrong thing.
PR teams need to understand which publications, platforms, creators, communities and travel intermediaries are informing Arabic-language answers. Some will be familiar media targets. Others may sit outside traditional PR planning entirely. That matters because being recommended and being the evidence behind a recommendation are not the same outcome. If the credible Arabic evidence around a brand is thin, AI will not leave the space blank. It may construct its answer from whatever accessible evidence it can retrieve.
The facts a brand wants repeated need to exist in accessible Arabic content: interviews, articles, owned pages, travel platforms, creator content and other credible third-party sources.
Not simply translated in an internal presentation.
One regional visibility score is no longer enough
This also creates a measurement problem. Most AI visibility dashboards still gravitate towards one deceptively simple question: does the brand appear? PR teams need to ask more.
Does it appear for the questions that matter? Does performance change by language? Which sources support the recommendation? Do those sources carry the brand story we intended? And are they sources we can realistically influence?
A single regional visibility percentage can flatten two very different communication realities into a single neat number.
Multilingual audits also need basic entity discipline. In our data, hotel names appeared across different spellings and scripts and had to be consolidated before rankings could be trusted. For comms professionals, this is a reminder that your name must be spelt consistently everywhere – across languages, platforms, and markets – because even small variations can fragment visibility and distort how AI systems understand and surface your brand.
Measurement without that consolidation can create false precision.
There may be a third ecosystem coming
This study tested ChatGPT, Gemini, Copilot, Google AI Overviews and AI Mode. It did not test Arabic-first systems such as HUMAIN Chat, Fanar 2.0 or Jais 2. If Gulf audiences increasingly use models built specifically around Arabic language and regional context, their source behaviour may create a third evidence ecosystem that most Western AI-visibility platforms are not yet measuring. That is worth watching, but brands do not need to wait for the measurement tools.
The groundwork is the same: credible Arabic coverage, accessible facts, consistent naming and strong regional corroboration. For PR teams working across the Gulf, language can no longer be the final localisation step after the strategy is finished. It belongs much earlier in the brief.
Because the shortlist may translate. But the reputation doesn’t.
The full research report is available here.