AI search sends traffic differently across 10 global markets

What happened

Per Aleyda Solis, AI search does not behave like a single global channel. Solis analysed top click-producing AI search domains across 10 markets and found that the dominant narrative ("AI assistants surface the biggest brands, marketplaces win ecommerce, OTAs win travel") collapses once you look country by country. The mix of cited domains, the role of local players, and the categories where AI assistants actually drive clicks all shift across borders.

Solis reports that the recognised-brand playbook is only partly true. In some markets, AI assistants concentrate citations on a handful of global incumbents. In others, local publishers, regulators, and category specialists take a meaningful share of the click traffic that ChatGPT, Perplexity, Gemini, and Copilot pass on to publishers.

The takeaway is uncomfortable for global brand teams: a single English-language AI visibility strategy, built on US data, will misread the citation landscape in seven or eight of the 10 markets Solis looked at.

Why it matters for your brand

For B2B marketers running global programmes, this finding upends a quiet assumption baked into most AI visibility work over the last 18 months. The assumption: if you win citations in ChatGPT's English answers, you have won the category, and the rest of the world will follow. Solis's data says the opposite. The domains AI assistants trust to answer a question about retail banking in Germany are not the domains they trust in Mexico or Indonesia. The domains they trust for industrial procurement in France are not the ones they trust in Japan.

For financial services brands, this is the most operationally significant point. A global bank or asset manager that has invested in being the cited authority on, say, ESG reporting or private credit in US-language ChatGPT outputs cannot assume that authority transfers. AI assistants in non-English markets often lean harder on local regulators, local financial press, and local trade bodies, because those are the sources their training data and retrieval layers treat as credible for that jurisdiction. Citation share in one market tells you almost nothing about citation share in another.

For multilaterals and the UN system, the implication cuts a different way. These institutions already produce content in multiple languages, but they tend to optimise for the English version and let translations follow. Solis's market-level variance suggests that the local-language versions of UNDRR, World Bank, or WHO content may be doing more visibility work than the English originals in any given non-English market, and they are almost certainly under-instrumented. If you cannot see which language version of your report an AI assistant is citing in Brazil or Indonesia, you cannot defend that citation when a model update changes the retrieval pattern.

For major industrial groups, the practical question is whether your country sites are being cited at all, or whether the assistant is routing every query back to the global English domain. The answer determines whether you need a country-level content investment or a consolidated hub. Solis's data implies that the right answer is different for, say, Holcim in Switzerland versus Holcim in the Philippines, and the wrong call costs you visibility in the markets where AI search adoption is actually growing fastest.