Criteo introduces product recommendation service for AI shopping assistants

Criteo has introduced a new service aimed at supporting AI-powered shopping assistants with product recommendations based on real-world commerce data.

The service, called the Agentic Commerce Recommendation Service, is designed for use by LLM platforms and retailer-built AI chatbots that assist consumers with product discovery and purchasing. The service draws on shopping and transaction data rather than relying only on product descriptions available online.

As AI assistants play a larger role in how consumers search for and compare products, the recommendation systems need access to shopping behaviour and purchase signals to deliver relevant results. The new service builds on its previously outlined agentic commerce strategy.

Internal testing showed the service delivered up to a 60% improvement in recommendation relevancy compared with third-party approaches that rely only on product descriptions. The testing was conducted in January 2026 using transactional data.

The service is available through Criteo’s Model Context Protocol and connects AI assistants directly with merchant inventory. When a consumer submits a shopping request, the AI assistant queries Criteo’s system, which filters and ranks products using shopping and purchase signals such as availability, popularity and user intent. The service then returns a curated list of products rather than a full catalog.

The system supports both broad exploratory searches and specific product queries and can suggest complementary items when relevant.

Speaking about the new service, Michael Komasinski, CEO, Criteo, said, “The real competitive advantage in agentic commerce will come from access to high-quality commerce data at scale. This service brings that intelligence into AI-driven shopping experiences in a way that works for the entire ecosystem, delivering relevancy for consumers while respecting retailer data, brand integrity, and platform trust.” 

It has been testing the service with an LLM platform since 2025 and is expanding testing to additional platforms, retailers and brands.