Mercado Libre Offers Food Delivery. AI Still Sends You to PedidosYa and Rappi.

Mercado Libre is the most visible brand in Argentine e-commerce. Ask an AI for the best sites to shop online and it comes first in all 510 answers we recorded. Ask the same AIs where to order fast food in San Telmo, La Plata, Mendoza or Rosario, and it appears in none of 410 answers, even though you can order from restaurants inside its app.

This is a case study from Lumen AI monitoring. Mercado Libre is not a Lumen customer: we tracked it in two Lumen accounts as a public benchmark, using only public AI answers. The pattern it shows applies to any company that grew by adding business lines.

How did we test the food delivery gap?

We used two kinds of prompts, the same mix that Lumen suggests proposes when it analyses a brand. The San Telmo prompt below was added from one of those suggestions.

• Ranking, San Telmo (Buenos Aires), May–June 2026: "Pedir comida rápida en San Telmo por app". 220 answers from OpenAI and Gemini.
• Ranking, La Plata, May–October 2026: "Qué plataforma existe para pedir comida rápida en La Plata". 130 answers from OpenAI, Gemini and Google AI Overviews.
• Ranking, Mendoza and Rosario, July 2026: "En qué app puedo pedir comida en [ciudad]". 30 answers from the three engines per city, 60 in total.
• Ranking, Córdoba, June 2026: "Alternativas a PedidosYa para delivery de comida en Córdoba". 60 answers from the three engines. This prompt is separate because it names a competitor.
• Precision, May–October 2026: "¿Mercado Libre ofrece envío de comida?". 140 answers from the three engines.

One San Telmo answer was flagged as a mention because it listed Mercado Pago as a payment method. We do not count it: the brand was not recommended as a delivery option. The numbers above are answers, not independent questions: several come from repeated runs of the same prompt.

Does Mercado Libre really offer food delivery?

Yes. In June 2022 Perfil reported that Mercado Libre would launch food delivery inside the Mercado Pago app, with Argentina first and restaurants handling their own deliveries. Today Mercado Pago’s delivery page tells users they can order delivery from the app, and Mercado Libre has a landing page inviting users to order food through its own app.

What did each AI engine say when asked directly?

1. 01
   OpenAI: confidently wrong: In 50 of 50 runs it said Mercado Libre does not usually work as a restaurant delivery app like Rappi, Uber Eats or PedidosYa, and pointed to grocery and food categories on the marketplace, citing Mercado Libre’s Mexican site.
2. 02
   Gemini: changes its mind: Across 50 runs it mostly hedged that the service depends on the country and sent the user to a "Delivery" section in the Mercado Pago app. Two runs, 21 minutes apart on the same day, gave opposite answers: yes, it offers food delivery, and no, it is not a consumer food ordering app.
3. 03
   Google AI Overviews: right, but only if you name the brand: In 40 of 40 runs it described ordering from restaurants inside the Mercado Libre and Mercado Pago apps, with a free-shipping threshold and a first-order discount. 30 of the 40 answers listed CABA, GBA, Córdoba, La Plata and Rosario as covered (details as stated by the AI, not verified by us). For the generic questions, the same engine cited city-level pages from Rappi and PedidosYa and left Mercado Libre out.

What happened in Córdoba?

Córdoba is the one place where the brand showed up without being named. We asked for "alternatives to PedidosYa", and Gemini listed Mercado Libre in second place in all 20 runs, behind Rappi. It called the service "Mercado Envíos (Mercado Libre)", which is the company’s logistics brand, not the restaurants product. OpenAI and Google AI Overviews did not name Mercado Libre in any of their 40 answers to that prompt.

A prompt that already points at a competitor’s users is an easier question than "where should I order food?". Only one of three engines made the connection, and it used the wrong name for the product.

Why does a giant get missed in its own new category?

AI engines tie brands to categories. For Mercado Libre that is marketplace, shipping and payments. Food delivery is a newer business line attached to a brand whose identity is already full. Four signals in our data are consistent with that explanation, though they are not proof of how any engine weighs sources:

• The service has more than one name: "Delivery" in the Mercado Pago app and "Restaurantes" in the Mercado Libre app. Gemini called it "Mercado Envíos" in Córdoba.
• The incumbents own city-level pages. For the La Plata question, Google AI Overviews cited Rappi’s "Delivery de comida La Plata" listing and PedidosYa’s La Plata page.
• Mercado Libre’s own sources are thin. For the direct question, Google AI Overviews cited one generic Mercado Pago page and two Instagram reels, and the text we could extract from Mercado Libre’s restaurants landing page is a headline and one sentence.
• The engines disagree with each other and with themselves. OpenAI says no every time, Google AI Overviews says yes every time, and Gemini changes between runs.

Why would a mention-rate dashboard have missed this?

Mercado Libre’s shopping prompts return 100% mention rate and rank #1, so a headline AI visibility number looks excellent. The precision prompt also reports a mention in 140 of 140 answers, because the brand name is in the question, while OpenAI gives the wrong answer in 50 of them. Mention rate counts the name, not whether the answer is right.

The gap only appears with two kinds of prompt tracking: a ranking prompt for the category and place where a customer decides, and a precision prompt that checks what each engine believes about a specific service. Together they show where the brand is absent and why. This is the core of GEO: measuring the product, not just the brand.

How can a brand close a category gap in AI answers?

1. 01
   Give the service one name: Use the same name for the product across apps, site, press and partner pages, so engines can attach it to the brand and stop confusing it with logistics.
2. 02
   Publish city and category pages: Answer "pedir comida en [ciudad]" with the facts: where it works, how it works, fees and restaurants. The incumbents win these queries with exactly that.
3. 03
   Earn third-party citations: Press, reviews and local media that describe the service in the category’s own words give engines sources other than the brand’s own pages.
4. 04
   Re-measure by engine: Correcting OpenAI’s wrong belief, steadying Gemini and connecting Google AI Overviews to the category are different jobs. Track each engine separately and re-run weekly.

Which other companies have the same blind spot?

Any company that grew by adding business lines: a bank that added insurance, a retailer that added a marketplace, a telecom that added payments. Brand-level metrics can look healthy while a whole product line is invisible. If an AI does not know what you sell, it cannot recommend it.