AI shopping answers in Mexico: what the Spanish-language sample shows

Mexico has its own shops, brands and prices, so this post describes what the Dopbot Index recorded there and stops short of explaining it. The sample is small and was collected once, so read it as a description of 297 answers.

What is in the Mexico sample?

The Mexico run covers 30 categories, from mattresses and washing machines to dog food. Every prompt is written in Mexican Spanish, and the budget and premium prompts state prices in pesos. In the October 2026 edition each prompt was asked once on each engine: OpenAI's API with web search (gpt-6-luna) and, for Perplexity, its Agent API with the Sonar model (perplexity/sonar). That gave 297 answers, 150 from OpenAI and 147 from Perplexity, because 3 of the 150 Perplexity requests failed.

The web search location is set to Mexico on both engines. As of October 2026, OpenAI documents an approximate-location setting for this purpose, and the methodology describes ours. All 30 Mexico categories are the October 2026 edition, are not updated, and have about 10 answers each. You can browse them in the Mexico index.

Two example prompts show the style. For mattresses, the budget prompt reads "mejor colchón matrimonial por menos de $6,000 pesos" and the premium prompt reads "¿Cuál es el mejor colchón de lujo, de $20,000 pesos en adelante?". Both are on the mattresses page.

How do brands per answer compare across the three markets?

Brand counts below leave out the alternative-to-a-brand question, which names a brand in the prompt and lists more brands than the other four. Both engines are pooled. The engine overlap divides the brands both engines recommended in a category by the brands either recommended, then averages over categories.

October 2026 edition, excluding the alternative question. Mexico and Canada have 30 categories each, the United States 100.

"No brand" means no recommended brand: retailers and marketplaces do not count as brands, and a brand an answer only mentions does not count either.

Brands per answer and engine overlap barely move across markets. The no-brand share does. About 25 of Mexico's 237 answers recommended no brand. With the alternative question included, the share is 8.4% of 297 answers.

We cannot say why. The Mexico categories and prompts are not translations of the US ones, the answers are in Spanish, and the sample is one run. The published figures do not split Mexico's no-brand answers by question type, so we do not know which prompts they came from. Treat the gap as something to check in your own category, not as a feature of Spanish or of Mexico.

Which domains show up in Mexico's source lists?

The next table is pooled: it counts the domains in the source lists of Mexico's answers with both engines together. OpenAI's list is the pages its answer links, and Perplexity's is the pages its search retrieved, and Perplexity's lists are much longer, so they supply most of the domains. The share is of the 296 Mexico answers that had any source. The "Kind" column is our reading of the domain name; only consumidores.com.mx and gob.mx were checked against the sites themselves.

October 2026 edition, Mexico, 297 answers, OpenAI and Perplexity pooled.

19 of the 30 most frequent domains in Mexico's lists end in .mx, and 10 of the 30 are on our hand-made retailer list. Several more look like stores but are not on it, so the retail share is understated. International editorial sites appear further down, for example nytimes.com in 17 answers (5.7%). The government portal gob.mx is in 16 answers (5.4%).

Two Mexican retailers have counts we can split by engine, because their pooled count equals the sum of the two engines' totals. That means every appearance was in a Mexico answer.

Both domains appear much more often in Perplexity's retrieved lists than in OpenAI's links. That fits the recording difference: Perplexity's lists hold a mean of 15.2 distinct domains per answer and OpenAI's hold 2.0, over all markets. A domain in a list does not show that it shaped a recommendation.

What can a brand selling in Mexico take from this?

The sample cannot tell you what makes a brand appear, so the steps are about measuring and about what the source lists suggest.

1. Check the retailers first. In these lists, Mexican retail domains are the most frequent entries. If you sell through Amazon México, Liverpool or Walmart México, compare each product page there with your own site, as in the retail listings post.
2. Run your questions in Mexican Spanish with peso budgets. Our budget and premium prompts use that wording, and a question in another language or currency is a different question.
3. Record each engine on its own. Engine overlap in Mexico was 46.3%, so your brand can be named by one engine and not the other.
4. Look for your domain and its variants in the lists. Your .mx domain, a store subdomain and a parent company domain. The own-website post shows how strict matching misses some of these.
5. Collect enough answers. About 10 answers per category supports examples, not rankings.

What this sample cannot tell you

• It is the API, not the app. The apps can differ in personalization, memory, location and interface.
• It is one run. In the October 2026 edition each prompt was asked once per engine. We have not compared Mexico over time and the categories are frozen.
• It describes answers, not shoppers. We did not observe what anyone searched.
• It is not a comparison with English. The categories, prompts and prices differ, so no difference here can be put down to language.

Frequently asked questions

Will the Mexico numbers be updated?

No. All 30 Mexico categories stay published as the October 2026 edition, and each category page says so.

Do the answers in Spanish recommend different brands than the English ones?

This sample cannot answer it. The Mexico categories and prompts are different from the US and Canada ones, and our one same-name comparison is between the US and Canada, which the Canada post covers.