Your brand disappears
in Portuguese and Spanish
You rank in English. Then a buyer in São Paulo asks ChatGPT which tool to use, in Portuguese, and your name is not in the answer. AI visibility does not carry across languages, and the translated version of your site is competing against you. We fix that, natively, and we measure it.
Three things that are true and expensive
Visibility resets per language
A brand's presence in generative engines starts from zero in every language it has not explicitly optimized for. Your English authority does not transfer.
Translation makes it worse
Translated pages carry anglophone sources, US market data and a structure built for another reader. Models detect weak translation, and it competes at a disadvantage against native content citing local sources.
The click is disappearing
When an AI summary appears in search, only 8% of queries end in a click on a traditional result, against 15% without one. Just 1% of users click a source cited inside the summary. Pew Research Center, 2025.
What actually moves the needle, measured
We do not work on intuition. The techniques below were measured in a controlled experiment over roughly 10,000 queries, across 9 datasets and 7 domains, in the study that coined the term Generative Engine Optimization.
“Including citations, quotations from relevant sources and statistics significantly boosts source visibility, with an increase of over 40% across various queries.”
Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande. GEO: Generative Engine Optimization, KDD 2024.
Direct quotation · +41%
The strongest single technique: +41% in PAWC and +28% in subjective impression. An attributed quote gives the model a discrete, verifiable unit to reproduce.
Statistics · +31%
+31% in PAWC and +23% in impression. A specific percentage with a named origin is quotable. “Many companies” is not.
Cited sources · +31.4%
Weak alone, but adds +31.4% on average when combined with the two above. Fluency plus statistics was the best combination in the study, over 5.5% above any single technique.
The equalizer effect: rank 5 gains +115%
Pages sitting at position 5 that started citing sources saw +115.1% in relative visibility. Pages already at position 1 saw minimal change. In other words, GEO returns disproportionately for brands that do not dominate traditional search. If you have scaled back SEO investment, this is the more efficient entry point.
An honest caveat: the 40% figure circulating in the market is the study's maximum, not the average. The real range runs from roughly 22% to 41%, and gains are domain dependent.
Why we do not charge for llms.txt
Nearly every GEO proposal lists llms.txt as a deliverable. The available
evidence does not support it:
- Google documents that the file has no effect, positive or negative, on Search rankings or AI Overviews. John Mueller compared it to the old keywords meta tag, ignored for over a decade.
- None of the major labs (OpenAI, Google, Anthropic, Meta, Mistral) has publicly committed to using it in production.
- Across 500 million AI bot visits monitored over 90 days, 408 touched the file. Limy, 2026.
- Ahrefs analysed 137,000 sites: 97% of llms.txt files received no traffic at all. Ahrefs. SE Ranking checked 300,000 domains and found no correlation with AI citations.
We still publish the file on your site. It takes twenty minutes and may serve agents later. But it is a routing artifact, not a visibility lever, and it does not go on the invoice. When a competitor lists it in a quote, ask for the data.
Four phases, each with its own verifier
Baseline citation matrix
20 to 40 buying-intent questions from your category, in Portuguese and Spanish, run with repetition across ChatGPT, Perplexity, Gemini and AI Overviews. Output: your citation share against competitors, per engine. We also check whether AI crawlers can read your site. A blocked site is the most common finding and the cheapest to fix.
Native rewriting, not translation
Priority pages get attributed quotes, sourced numbers and explicit external references, written natively with local sources and local market data. This is the part an agency in Chicago cannot replicate.
Entity and structure
A canonical page per entity, JSON-LD consistent with visible text, a clear definition in the first paragraph. The model has to be able to answer “what is this company” from a sentence you wrote.
Weekly measurement
The same questions, every month, with repetition. Plus server logs separating GPTBot, ClaudeBot, PerplexityBot and OAI-SearchBot from Googlebot, which is proof of reading rather than an estimate.
Frequently asked
What is GEO (Generative Engine Optimization)?
The set of techniques that get a brand cited inside the answers of generative AI systems such as ChatGPT, Perplexity, Gemini and AI Overviews. It differs from SEO because there is no “position 1” in a synthesized answer: what gets measured is your brand's share of the generated text.
Why do I need this in Portuguese and Spanish specifically?
Because a brand's AI visibility resets to near zero in every language it has not explicitly optimized for. Ranking in English does not carry over. If your buyers in São Paulo or Mexico City ask an AI which tool to use, your English-language authority does not enter that answer.
We already translated our site. Is that enough?
No, and it can work against you. Translated content keeps anglophone sources, US market data and a structure written for a different reader. It competes at a disadvantage against native content that cites local sources. Machine translation in particular is detectable and actively hurts international AI visibility.
Isn't Brazilian Portuguese the same as European Portuguese?
No. Vocabulary, idiom and, most importantly, reference sources differ enough to require separate content programs. The same applies to Spanish, which splits into Spain, Latin America and US Hispanic markets with different search behaviour.
Do I need an llms.txt file?
No. Google documents that the file has no effect, positive or negative, on Search rankings or AI Overviews. We publish one anyway because it takes twenty minutes and may serve agents later, but we do not charge for it and do not present it as a deliverable.
How long until we see results?
The honest range is 90 to 180 days. Model training and refresh cycles are outside any vendor's control. That is why we work on a three-month minimum: selling this month to month guarantees cancellation before the effect appears.
Can you guarantee we appear in ChatGPT?
No, and anyone who guarantees it is selling something they cannot deliver. Generative engines are not deterministic: the same question produces different answers. We measure each question across repetitions and report citation share, not position.
Start with the diagnostic
We run the citation matrix for your category in Portuguese and Spanish and show how often you and your competitors appear. No commitment. If the number is good, we say so too.