The AI audience stopped speaking English. The retrieval didn't.
More than half of ChatGPT's users now work in another language — yet its background search still runs in English, and those English sources favor global chains over your hotel
by Markus Busch
The shift. OpenAI just put a number on it. In its June 30 Signals release, the company reported that people who predominantly use a language other than English now make up more than half of active ChatGPT consumers. The leading non-English languages are Spanish, Portuguese, and Arabic, and the fastest growth is in Africa, Asia, and lower-income countries. English speakers built the early platform; they are no longer the majority inside it.
The catch in the machine. The retrieval layer didn't get the memo. Peec AI, analyzing more than 10 million prompts and 20 million of ChatGPT's background "fan-out" queries, found that 43% of those fan-out steps run in English even when the original question wasn't — and 78% of non-English prompts triggered at least one English-language search along the way. Turkish questions detoured through English 94% of the time; Spanish, the lowest, still 66%. A traveler can ask in Arabic and have the model quietly consult English pages before it answers.
Who that hands the answer to. This is where it bites a hotel. In Peec's examples, those English fan-out queries pulled English-language sources, and those sources favored global brands over local competitors. So even when the guest asks in their own language, the model's English detour can route the recommendation toward the international chain with the deep English footprint, and past the independent property down the street. The machine's second language is quietly an advantage for whoever is biggest in English.
What it means for your marketing. The reflex — translate the website — is half the answer, and the wrong half if it's the only move. Two jobs run in parallel now. Keep strong, accurate English content, because the retrieval leans on it even when your guest never types a word of English. And build genuine presence in your feeder markets' languages, because that is the language the answer arrives in, and the non-English sources the model can draw on are growing fast. Then do the unglamorous part: check which of your actual source markets have shifted to planning in their own language. The global "half" is not your number; your feeder mix is.
The stake. The audience inside AI stopped being English-first. The machine that serves them hasn't caught up. Until it does, your discoverability answers to two languages at once — the traveler's and the model's — and the hotels that notice will get found in both.
Limits: OpenAI's figures cover consumer plans and general use, not travel specifically; Peec's come from its own platform sample, and whether the English detour is deliberate or an artifact of training is still unknown. The direction is clear; the exact shares will move.
Enjoying this analysis? Hospitality.today delivers daily insights on hotel distribution, AI trends, and travel commerce — straight to your inbox. Subscribe for free at Hospitality.today →