Last week I ran an experiment in Spanish, asking three AI models which are the best PR and influencer marketing agencies in Mexico. This week I ran the same three prompts in English, phrased the way a US brand manager actually types them: “which ones should I consider,” “recommend top PR agencies for a US brand entering the Mexican market.”
The answers came from an almost entirely different set of sources.
In Spanish, Perplexity leaned on Clutch, Sortlist, marketing4ecommerce.mx, and a handful of B2B directories. In English, those same directories showed up, but a new name dominated every single response: everything-pr.com, a listicle site I’d never had a reason to check before this experiment. It appeared in all three answers, sometimes cited four or five times in one response. TheManifest.com, a sister site to Clutch that I’d also never tracked, showed up almost as often.
None of this is random. Perplexity is the only one of the three models that actually searches the live web when it answers, and search results in English simply favor different sites than search results in Spanish. everything-pr.com ranks well for “best PR firms Mexico” in English. It probably ranks nowhere for “mejores agencias de relaciones públicas México.” Two separate visibility games, running on two separate scoreboards, and most agencies are only aware one of them exists.
The two models that don’t search live, OpenAI’s GPT-4o-mini and Google’s Gemini, cited nothing at all in either language. They answered from training data, the same way they’d answer a trivia question, which means whatever they say about your agency is frozen at some point in the past and nobody is checking if it’s still true.
There’s a detail in the English results that changed how I think about this. One of the citations was LLYC’s own press release announcing an award it won from a Mexican trade magazine. Not a directory, not a review site: a company’s own announcement of recognition it received, indexed and picked up as a source. That’s a much cheaper play than fighting for a spot on Clutch, and almost nobody I know treats award wins or press mentions as inputs into an AI visibility strategy. They treat them as a line in a year-end recap deck.
The practical implication, if you’re marketing an agency, a hotel, or really any B2B service with a Mexican and an international buyer at the same time: auditing your visibility in one language tells you almost nothing about the other. A brand can be reasonably well positioned when a Mexican CMO asks ChatGPT in Spanish and be invisible the moment a Miami-based VP of marketing asks the same question in English, because the model is drawing from two different pools of sources built by two different search ecosystems.
I ran this specific test with my own agency in mind, and the honest result is that Lithiumpr didn’t appear once, in either language, across nine calls. That’s not a complaint about the tool. It’s the reason I ran it. If I hadn’t checked, I’d still be assuming that ranking on Google in Spanish was enough.
If you sell into more than one language market, the question worth asking isn’t just “what does ChatGPT say about us.” It’s “what does it say about us in the language our actual buyer uses to search.” Those can be two very different answers, and most companies have only ever checked one of them.
