Agentic commerce: understanding ACP and UCP protocols

Charles Cortés, chief operating officer of Converteo Spain, supports brands in optimizing their acquisition performance and leveraging data. Damián Bourgeois, director of Dataiads, is a specialist in product feeds and their activation across all channels.

Key takeaways

• The shopping journey is moving from search engines to answer engines. With ChatGPT and Gemini, users discover, compare, and buy without leaving the AI interface. The challenge is no longer to be present, but to be understood by the machine.
• Two protocols shape this new market: OpenAI’s ACP and Google’s UCP. Two different philosophies — one close to a commission-based marketplace, the other free and backed by the advertising model — but the same underlying shift: integrated purchasing within AI.
• 2026 will be a pivotal year. Few companies have deployed these protocols, but many are preparing for them. The risk for laggards: disappearing from generated answers while competitors establish themselves.

For years, the online shopping journey followed a stable logic: a query on a search engine, a click to a site, browsing a catalog, then payment. This world where the goal was to “present a catalog” is giving way to a world where the consumer wants to “converse with a catalog”. Agentic commerce is not just a matter of technology, it is a strategic repositioning.

From search engine to answer engine

User habits have already shifted. A growing share of internet users no longer systematically go through classic search: they directly ask ChatGPT, Gemini, or other AIs, which return contextualized and personalized results. OpenAI and Google have understood this and are deploying new features at a steady pace.

Agentic commerce encompasses several realities. On one hand, brands that integrate their own agents on their site to fluidify the experience. On the other hand, the agents of major generative AIs that offer to complete the purchase without leaving their platform. In between, consumers can also mobilize their own agents to search and buy.

ACP: OpenAI’s marketplace vision

The ACP (Agentic Commerce Protocol) protocol allows ChatGPT users to search for a product, get results that take their context into account, consult descriptions, and buy without leaving the interface. OpenAI started with a strong partnership with Shopify and entrusted payment to Stripe, before opening up to other partners.

The economic model evokes a marketplace logic: a commission of around 4% for OpenAI, to which is added that of Stripe, making a total of around 8%. A competitive level compared to the 8% to 15% charged by Amazon depending on volumes and sectors.

UCP: Google’s integrated response

The UCP (Universal Commerce Protocol) protocol works on a similar principle, but integrates into Gemini and the AI mode of search. Google’s vision is broader: the collaboration covers the entire value chain — deliveries, physical points of sale, financial flows. Another major difference: Google does not charge a commission, at least for now. A clear counter-attack against OpenAI, relying on its historical advertising model, with more than twenty strategic partners already announced.

Rufus: what the data already proves

Among the most advanced cases is Rufus, Amazon’s agent, which offers a hybrid navigation between traditional experience and conversational assistance. In the United States, Rufus can now go as far as payment; in Europe, not yet — but it is probably only a matter of time.

Amazon’s data confirms the intuition: users who go through the agent show a higher conversion rate and a higher average basket. A strong signal for brands that are still hesitating.

SEO vs GEO: the race for mentions

In this new world, it is no longer enough to fight for the click in classic search: you must also be mentioned by answer engines. However, these engines have an inverse logic: they do not try to make the user leave, but to retain them by synthesizing information. To do this, they must deeply understand the content to produce a personalized answer: this is the whole point of GEO (Generative Engine Optimization).

The essential point: robots are blind. They do not see your design, they read your structure and your content. If a product page does not clearly specify the price, sizes, or the fact that it is a purchasable product, the AI will not understand it. You have to work simultaneously for humans and machines, and this requires optimizing the product feed and the product page.