MercadoLibre
GenAI embedded in a marketplace: AI search and conversational assistant on the buyer side, a Seller Assistant that generates listings and short videos on the seller side, all feeding the ad business where AI optimizes bids. The brand owns the marketplace, hence the behavioral data and the ad inventory.
MercadoLibre embedded generative AI in its marketplace (AI search in Argentina, buyer assistant, Seller Assistant creating videos from a product photo), driving its advertising revenue to +67% in constant currency in Q4 2025 and ~20% of its GMV advised by the seller assistant.
Key points
- GenAI embedded in the marketplace: AI search, buyer assistant, Seller Assistant.
- Custom in-house building blocks feeding the ad business where AI optimizes bids.
- Advertising revenue +67% in constant currency in Q4 2025, ~20% of GMV advised by the assistant.
- Evidence A, confirmed status: figures from the Q4 2025 SEC 8-K and earnings call.
Objective
Make AI a cross-functional growth engine: improve product discovery and conversion on the buyer side, speed up and industrialize supply on the seller side, and above all monetize through an ad business where AI drives bids and campaigns. All without raising churn or degrading conversion.
The deployment
MercadoLibre, the number one marketplace and fintech in Latin America, integrated generative AI at several points of its ecosystem and documented it in its fourth-quarter 2025 results (published on February 24, 2026). On the buyer side: an AI-enriched search launched in Argentina, which starts from a broad term and expands it based on customer behavior. The word 'ball' returns tennis balls to a tennis player and soccer balls to a football fan, and may show brands or a price range depending on what the platform knows about the buyer. A conversational assistant refines a vague search like 'smartphone' by guiding the user on product attributes. On the seller side: the Seller Assistant speeds up sellers' progression toward better reputation tiers, improves listing quality, creates short videos from a single product photo, and handles requests that previously went through customer service. According to management, about 20% of GMV is advised by this assistant. These building blocks feed the ad business: advertising revenue climbed 67% in constant currency in Q4 2025 (70% in dollars), driven by search ads and by investments in bidding and campaign tools. In parallel, Mercado Pago's AI assistant, launched in October 2025, handled more than 9 million conversations over the quarter, nearly 90% of them resolved without human intervention.
Results Proof A
Figures published in the Q4 2025 financial results (SEC 8-K, exhibit 99.1, letter to shareholders), with the detail of the AI use cases and quantified advertising growth. Confirmed by the earnings call transcript (share of GMV advised by the assistant) and picked up by business press. Three concordant sources including a primary SEC document.
How it works
Documented architectureHow it runs, concretely
For ops teams-
1Understand intent at search AI (search model)
The AI engine expands a broad term into relevant results based on the buyer's profile, where a keyword search returned an undifferentiated list.
-
2Assist the buyer in conversation AI (conversational agent)
An agent guides the user on product attributes to turn a vague query into a precise selection.
-
3Industrialize seller supply AI + seller
The Seller Assistant generates short videos from a photo, improves listings and answers customer questions; the seller validates and publishes.
-
4Monetize through the ad business AI (bidding) + ads team
Search ads are optimized by bidding algorithms and automated campaign tools; advertisers pay on performance.
-
5Loop back on the data Data team
Each interaction (search, click, purchase, conversation) enriches the behavioral signal that refines the models on the next round.
First-party purchase and search behavior (history, transactions, viewed attributes). This is the signal that personalizes search, feeds ad bidding and trains the seller recommendations. Without this volume of proprietary data, personalization and bid optimization lose their basis.
How your customers perceive this type of use
Sourced studiesUn ecart net separe les annonceurs des consommateurs : 77% des annonceurs voient l'IA positivement contre 38% des consommateurs (Yahoo/Publicis, 2024). Les mesures implicites confirment le rejet declare : en EEG, les pubs generees par IA produisent une activation memorielle plus faible que les pubs traditionnelles et sont decrites comme agacantes, ennuyeuses et confuses (NIQ, 2024). La disclosure a un effet ambivalent : elle augmente fortement la confiance quand elle est remarquee (Yahoo/Publicis), mais 27% des jeunes consommateurs disent faire moins confiance a une entreprise dont la pub est creee par IA (IAB, 2024).
Acceptance conditions
- Une disclosure visible : quand la mention IA est remarquee, la confiance globale envers l'entreprise augmente de 96% (Yahoo/Publicis 2024)
- Une qualite visuelle suffisante : les visuels IA de basse qualite augmentent l'effort cognitif et distraient du message (NIQ 2024)
Red lines
- Le contenu IA non declare puis identifie : 72% des consommateurs disent que l'IA rend l'authenticite difficile a etablir (Yahoo/Publicis 2024) et les marques utilisant des pubs IA sont plus souvent jugees inauthentiques ou non ethiques par les consommateurs que par les dirigeants (IAB 2024)
- Les mannequins et personnes generes par IA : 46% des consommateurs n'en veulent pas dans la publicite, l'inquietude premiere etant les standards de beaute irrealistes (Attest 2025)
Sources: Yahoo / Publicis Media (terrain Ebco) 2024 · IAB (avec Attest) 2024 · NIQ (NielsenIQ) 2024 · Attest 2025
How to replicate
Inference, not sourcedData prerequisites
- Volume of first-party behavioral data (search, transactions)
- Structured product catalog and usable attributes
- Proprietary ad inventory to monetize
Org prerequisites
- Own the marketplace or the audience (not transposable to a pure advertiser)
- Durable internal AI engineering capacity
- An ad business or retail media program to run
Possible stack
- LLM and semantic search models (in-house or vendor)
- Video/image generation from a product photo
- Conversational agent connected to the catalog
- ML bidding engine for the ad business
The plan, step by step
- Step 1Map the available first-party behavioral data (searches, clicks, purchases, viewed attributes) and verify it is usable for personalization, including the legal basis in the EU.Deliverable: Data inventory and compliance framing.
- Step 2Start with search: move from a keyword engine to a search that understands intent and expands based on the profile, on a perimeter where failure is low-cost.Deliverable: AI search in test on one segment or category.
- Step 3Add a conversational assistant on the buyer side to refine vague queries, then a generative tool on the supply side (listings, visuals, short videos) to industrialize the catalog.Deliverable: Buyer assistant and seller tool in production on a subset.
- Step 4Connect these building blocks to monetization: optimize the ad business's bids and campaigns on the same behavioral signal, measuring the effect on ad revenue, conversion and churn.Deliverable: Instrumented discovery -> ads loop with conversion/churn guardrails.
First step: Spot a high-volume, low-relevance search, plug in an engine that understands intent, and measure the effect on conversion before extending to the other building blocks.
The same pattern, proven in other industries
Duolingo
Conversational genAI tutor inside a premium subscription
Grammarly
freemium AI writing assistant converted into premium subscriptions and organizational licenses
Sources
- S1 MercadoLibre Inc - Form 8-K, Exhibit 99.1, Letter to Shareholders Q4'25 Primary archive pending
- S2 MercadoLibre (MELI) Q4 2025 Earnings Transcript - The Motley Fool Established press archive pending
- S3 Mercado Libre Says AI Investments Support 45% Revenue Surge - PYMNTS Established press archive pending
An error, newer info, a source?
This page lives on its accuracy. If a figure has moved, if the deployment has changed, or if you have a higher-quality source, tell us. Every sourced correction is verified before publication.