The language arbitrage: your brand is invisible to half the world's AI

I'll say it without preamble: there's a question an international buyer could be asking ChatGPT right now — "best SEO agency in Spain" — and your brand doesn't appear. It's not that you appear in a bad position. It's that you don't exist. And worse, you're probably measuring everything in Spanish, seeing yourself doing fine, and sleeping soundly on top of a hole the size of half the market.

At Elevam Labs we've spent months measuring how the four big models — ChatGPT, Gemini, Claude and Perplexity — recommend brands. And the variable that moves the needle most is not Schema, not backlinks, not the llms.txt everyone is talking about. It's language. Language matters in GEO. We measured it, and the difference is enormous.

Here's what we found.

What we did

We took 15 real prompts — the kind a decision-maker actually types when looking for someone to hire — and ran them across the four models. First in Spanish. Then the same fifteen, translated into English. Same model, same intent, same day. The only thing that changed was the language of the question.

The methodology is published and anyone can replicate it. That's deliberate: if you're going to believe a number from me, I want you to be able to verify it yourself. Anyone publishing a figure while hiding the method is selling smoke.

The base ranking, the one that is publicly verified, came out in MarketingDirecto.com: Elevam scored 86/100, with the highest coverage in the market — 10 out of 15 prompts on Perplexity, 9 on ChatGPT. That's the floor we measure from. Everything that follows is the new layer: what happens when you change the language.

Finding 1: the two lists barely overlap

One would expect "mejor agencia SEO de España" and "best SEO agency in Spain" to return, roughly, the same names. Different order, sure. But the same names.

Nope.

Of the brands appearing in the top of Spanish-language answers, only 18% reappear in the English-language answers. The other 82% evaporates. And the reverse: English brings names that didn't show up anywhere in Spanish — directories, aggregators, international players with a landing page in Madrid.

They are two almost disjoint universes. The same question, in two languages, places you in two different markets with two different sets of winners. If you only play one, you're giving away the other.

Finding 2: in English, Spanish agencies barely show up

And here comes the part that stings. When we looked at who occupies the English-language market for "agency in Spain", Spanish agencies don't just lose positions. They almost disappear.

In the English answers, 71% of recommendations go to third parties that aren't even Spanish agencies: directories like Clutch, aggregators, and international consultancies with nominal presence in Spain. Local firms — the ones actually doing the work here — share the crumbs.

This is not chance or bad luck. It's structural. Models drag a known bias toward big brands that hurts niche players — the academic GEO literature itself documents this. Add that all your authority is built in Spanish, and you have a double penalty: invisible by language, invisible by size. The same brand, punished twice.

Finding 3: publish in English and they see you better… in Spanish

This is the counterintuitive one, and the one I struggled to believe until we saw the numbers.

The logical reaction from a Spanish CEO is: "I sell in Reus, my clients are Spanish, why would I write in English?". Precisely because of that.

The model does not store one entry for you in Spanish and another in English. It builds a single idea of who you are, and it feeds it with everything it finds, whatever the language. Every mention of you in English feeds that idea. And that idea, more solid, is the one the model uses also when someone searches for you in Spanish.

We checked it by comparing two groups of comparable Spanish companies: ones with a solid presence in English, others only in Spanish. Faced with the same Spanish questions, the first group appeared on average 2.3 times more than the second. Same country, same sector, same quality. The only systematic difference: some existed in English and others didn't.

Why this is an arbitrage and not "publish in English, done"

An arbitrage is exploiting an inefficiency the market has not yet corrected, while the window is still open. Here the inefficiency is textbook:

On one side, models weight English much more heavily than Spanish — they are trained mostly in English, that's where the density is. On the other, the supply of authoritative content in English about the Spanish niche is, practically, zero. No one is writing it.

High density on one side, zero supply on the other. That's an undervalued asset in a market with no buyers. The effort to translate and publish well in English is marginal compared to the return it captures, because you're competing against no one. It's the moment before everyone notices. And they will notice.

The window doesn't last. The minute three Spanish agencies start publishing in English seriously, the arbitrage closes and it's back to a normal fight. Right now it's wide open.

What to do with this (three things, not ten)

One: measure in both languages. If your AI visibility dashboard only runs Spanish prompts, you're blind in one eye. Duplicate the prompt set into English. It's an afternoon's work and it shows you the half of the map you weren't seeing.

Two: a canonical English landing page, now. Don't translate the entire site. One page — the one that defines who you are and what you do — written, not Google-translated, in English. It's the first brick of your existence in the other universe.

Three: make your external signals speak English. The founder's profile, the repo if you have product, the guest post, the original data you publish. Every English mention reinforces the entry the model uses to recommend you — also in Spanish.

Closing

Language is the variable Spanish GEO is not looking at. We measured it, and it moves the needle more than half the things the industry obsesses over. The good news is almost no one knows it yet, so the cost of moving today is ridiculous compared to what it captures. The bad news is that this is not a secret you can keep: the minute enough people read it, it stops being an arbitrage.

At Elevam Labs we'll keep measuring it and publishing the data as the set grows. The methodology is open. If you want to replicate it for your brand, the 15 prompts and the method are on the table — let's talk.

The divergence between languages is the first chapter of the "State of GEO in Spain 2026" study by Elevam Labs. The second — how much the recommendation changes by model — is coming next.

Asier López Ruiz is CEO of Elevam, a pioneering SEO and GEO firm in Spain. Creator of the HSA Protocol. 2× Forbes Best SEO Agencies Spain.