How AI recommends brands, and why it sometimes picks your competitor

When someone asks “which mattress brand should I buy in Mexico?”, the engine can answer with what it already knows, with what it searches for at that moment, or with a mix of both. Understanding which path carries more weight on each engine explains almost everything else.

Chat engines (ChatGPT, Claude, Gemini, and Meta AI, the assistant inside WhatsApp, Instagram and Facebook) lean heavily on the model’s memory, although some also search the web. ChatGPT search, for example, shares queries with third-party search providers, and OpenAI names Bing among them. Google’s AI Overviews are AI search: they build the answer from what Google ranks for that query and paraphrase it. Google also breaks each question down into several sub-searches.

Nobody outside each company knows the internal details, but the general path can be observed by reading many answers:

1. It interprets the question and its conditions: category, country, budget, use (“for back pain,” “for a small business”).
2. It gathers candidates: brands it remembers as associated with that category, or that appear on the pages it found.
3. It keeps the ones that meet the conditions and that different sources associate with them, not the ones that describe themselves as the best.
4. It writes an answer with a few options, often between three and five, in a particular order: showing up first is not the same as showing up fifth.

The practical consequence: AI doesn’t reward whoever promotes themselves the most, but whoever the most sources associate with the specific question.

• Clarity about what you are. Your site says in plain words what you sell, to whom, and in which countries. “Integrated sleep solutions” tells an engine nothing; “orthopedic mattresses shipped anywhere in Mexico” does.
• Presence in comparisons and lists. When the question is “which one should I choose?”, engines tend to lean on “best options for…” articles and on comparisons in your category. If you aren’t in them, you compete at a disadvantage.
• Consistency across sources. The same name, the same category and the same facts on your site, your profiles, directories and the press. Contradictory data makes the engine hesitate, and an engine that hesitates names someone else.
• Content in the market’s language that names the country. A Spanish-language page that says “in Argentina” answers a question asked in Spanish that says “in Argentina.” It sounds obvious, and it is often missing.
• A recognizable entity. A Wikidata item, a correct Google knowledge panel and an unambiguous name help the engine know that your brand is one specific thing.
• Answers to the sub-questions. Prices in local currency, coverage, terms, honest comparisons. These are the pieces AI search goes looking for.
• Third-party coverage and reviews. In Refractia’s data, a brand’s own site is only between 0.5% and 2.8% of what AI cites when it talks about the brand. The rest is third parties.

If AI names your competitor and not you, there is almost always a concrete reason, and it is usually one of these:

• They are in the comparisons and you aren’t. Check which pages AI cites for your category and whether your brand appears in them.
• They have more history on the web. A brand with years of articles, reviews and mentions is more present in models’ memory than a new brand, even if the new one is better.
• Their site says clearly what they do. Yours may be written to impress rather than to be understood.
• They have a presence in English. Many sub-searches run in English even when the question is in Spanish; if your brand only exists in Spanish, it doesn’t show up there.
• Your name is ambiguous. If your brand shares its name with a common word or with another company, the engine may not associate it with your category.
• You do well on a different topic. You can dominate the questions about price and disappear on the ones about trust or service. The question you asked lands right where your competitor is strong.
• The engine didn’t search this time. If the answer came from the model’s memory, your recent content never took part.

Ask the same question twice and you can get two different lists. It isn’t a bug: models generate each answer with an element of randomness, searches return results that change, models get updated, and the context of the conversation has an influence. A small difference in wording (“best” versus “most recommended”) also moves the answer.

That is why a screenshot proves nothing, for or against. To know whether AI recommends you, you have to ask many questions, repeat them and count: “it names you in 18 of 40 answers” is data; “it named me yesterday” is an anecdote. The guide to AI share of voice explains how to do that count without inflating it.

The same question asked in Argentina, Mexico and Spain can produce three lists with no brand in common. Google’s AI Overviews answer with what ranks in the country the search comes from, and the models’ memory is uneven: it knows a lot about some markets and little about others. A leading brand in Chile may not exist in the answers about Peru.

Language adds another layer. For a Spanish-speaking brand, that means two jobs: having Spanish content that names the country, for local questions, and having a presence in English if you want to show up when people ask in that language.

Three common situations, with made-up brands to illustrate the mechanism:

1. 01

   Measure before you change anything

   Which questions in your category name you, on which engines and in which countries, and who gets named when it isn’t you. Without that, you are optimizing blind.

2. 02

   Make sure AI can read your site

   Crawlers allowed in robots.txt and in your CDN, content in the server-rendered HTML, pages indexed. The GEO checklist goes through each point.

3. 03

   Say what you are, for whom and where

   On your homepage and on every product page, in the words your customers use, naming the country and the currency.

4. 04

   Answer the questions in your category

   A clear page for each important question: pricing, comparisons, how to choose, coverage. The answer first, the details after.

5. 05

   Earn a presence in the sources AI cites

   Comparisons, industry media, reviews and directories in your country. Earned coverage, not bought.

6. 06

   Measure again with the same questions

   To know whether something worked, repeat the measurement on the same engines, in the same country and with the same questions.

Refractia asks category, comparison and purchase-intent questions in your market’s language, naming the country in full, to five engines: ChatGPT, Claude, Gemini, Google’s AI Overviews (fetched with each country’s location) and Meta AI. Every rate is shown with its sample, an engine that didn’t answer doesn’t count as “didn’t name you,” and chat and AI search are reported separately, because they improve with different actions. Share of Voice shows your position against your competitors, and Influence Sources shows which sites the engines cite.

For Argentina, we also publish the Índice de Marcas IA, our own research into which brands AI recommends across consumer categories, with a published methodology. We don’t measure Perplexity, Copilot, Grok, or Google’s AI Mode as a separate surface.