How Criteo is turning LLMs into its next big advertising channel

The company, historically associated with ad retargeting, is attempting to reposition itself for an AI-driven commerce era, and its latest demos suggest the company now sees large language models — not just retailers or demand-side platforms — as the next major distribution channel for advertising. Far from treating LLMs as an existential risk to its performance business, Criteo is leaning into them, through early experiments with a (publicly unnamed) LLM to position its commerce dataset as the missing ingredient that can make product discovery inside tools like ChatGPT or Claude actually work.

The company has started piping structured signals — relevance, trendiness, retailer-level performance — into LLM environments via its Model Context Protocol server, effectively allowing any agent inside those models to hit Criteo’s API when recommending a product. The bet is that generic web-crawl data is simply not good enough for high-fidelity commerce recommendations. If LLMs want to play in retail media or product suggestions, they’ll need something closer to Criteo’s longitudinal, transaction-linked dataset.

“The capability in the engine drives the full funnel… our deep learning capability can make sense of scattered interactions to uncover product recommendation opportunities,” explained Criteo CEO Michael Komasinski, at a recent press event hosted in New York City.

That’s the strategic story the company is now telling: its dataset, built over almost two decades, is a differentiator in a world where AI becomes the front-end, and ad-tech vendors become the infrastructure supplying relevance behind the scenes.

Muscle memory

Criteo’s leadership is increasingly foregrounding its origins — first as a DVD recommendation engine, then as a performance ad tech business — to emphasize that the current shift toward LLM-driven recommendations is not foreign territory. The company stresses that deep learning work, much of it accelerated during the “signal loss” era and Privacy Sandbox transition, now powers a broader, more adaptive optimization layer.

That engine underpins Criteo’s argument that it can operate across a full-funnel workflow: understanding scattered browsing interactions, inferring product affinities, and generating the embeddings that allow neural models to reason about relevance at scale. This is the plumbing the company now wants to plug into LLM-based commerce experiences.

All of this rests on a dataset it claims is under-appreciated across the ecosystem: 720 million daily active users, more than a trillion dollars in observed online transactions, and billions of SKUs stitched together through retailer and brand integrations. To hear Criteo tell it, this is the kind of behavioral richness LLMs fundamentally lack.

AI as workflow

Internally, the company is pushing AI deeper into campaign set-up and optimization. The “audience agent” tools are designed to let a marketer describe a goal in plain language — “help me sell more camping backpacks” — and automatically assemble the segments, reach curves, and relevancy scoring that previously required sifting through taxonomies and manual logic.

Commerce Go — a fuller workflow tool that’s currently billed for a launch in Q1, 2026 — builds on this logic. It automates audience creation, channel mapping, and creative generation (text-to-image, image-to-video, URL-to-video) in something closer to 10 minutes than the multi-vendor, multi-day process many marketers still endure. Criteo claims these automated campaigns deliver materially higher return on ad spend, and says thousands have already run through the system.

Crucially, none of this relies on a user logging into a Criteo UI. Through MCP, the same campaign-creation flows now run inside ChatGPT or Claude. A campaign manager can sit inside an LLM, reference an advertiser ID and brief the system conversationally. The LLM passes instructions to Criteo’s APIs and returns a fully assembled campaign that can then be tweaked or launched inside Criteo’s platform.

Criteo CTO Todd Parsons added, “This is just an example of how we’re using AI to move our legacy products forward. In the past, you would have to look up our taxonomy, and you would have had to select or deselect a load of audience attributes.”

This is where Criteo sees the near-term efficiency play: collapsing media planning steps by using AI to perform the connective tissue work that was historically manual, slow and fragmented.

Retail media

Criteo’s retail media customers, meanwhile, are racing to layer AI chat experiences into their own sites — a shift driven partly by shopper behavior and partly by fear of ceding discovery to LLMs. Criteo is positioning itself as the bridge between retailer-controlled AI chat and ad monetization, ensuring sponsored product placements remain relevant and don’t undermine shopper trust.

One technical focus is giving retailers finer control over relevancy scoring — moving from binary eligibility to graded scoring where a brand may bid against nuanced relevance thresholds. It’s an attempt to resolve the long-standing tension in retail media: balancing monetization with genuine usefulness to the shopper.

The “agentic” use case cuts across both sides of the house. Whether the front-end is a retailer’s own chat interface or a general-purpose LLM, Criteo wants its commerce data to power the recommendation logic beneath it.

LLMs as the next ad channel?

The company is clear that the test-and-learn phase with a major LLM provider is still early. But the framing is telling: Criteo sees agentic shopping as an incremental channel, not a cannibalistic one. If the performance holds up, the company expects to allocate budgets the same way it already reallocates between open-web and social performance — wherever product sales can be measured and verified.

Commercial models are still fluid: Criteo talks about everything from data licensing and pay-per-query models to native ad formats inside LLMs. However, the thesis remains consistent: high-fidelity commerce signals will be required for quality product discovery, and this creates an opportunity for an ad tech company with a robust recommendation stack.

For a business that has spent years trying to reshape its identity after the third-party cookie era, the LLM shift offers something genuinely new to lean into — not just as a defensive posture, but as a route into a more diversified set of revenue lines.

In short, Criteo is trying to ensure that when AI becomes the shop window, it is the vendor supplying the shelves.