When an AI agent searches, compares and buys for a person, it talks to shops, services and other agents. Protocols and conventions for that are taking shape right now. These are the terms you will run into, each with what it means for your marketing.
Agent to agent
As a concept: a person sends their AI agent, and that agent talks to the agent of the shop, the service or the brand. The person no longer visits the pages. In 2025 Google introduced an open protocol of the same name that lets agents from different providers hand tasks to each other.
For your marketing: the selection happens inside the agent, not on your page. What an agent cannot read, it will not recommend. What your brand needs now is on the agent to agent page.
MCP, the Model Context Protocol
An open standard from Anthropic, published in 2024. It defines how an AI model talks to tools and data sources: a server exposes functions and data, and the agent calls them in one uniform way.
For your marketing: through MCP an agent can query a product catalogue, stock levels or a booking calendar, for example. If you expose your data this way, agents can make enquiries without every provider building its own integration.
ACP, the Agentic Commerce Protocol
An open standard from OpenAI and Stripe, published in 2025. It describes how an agent starts and completes a purchase with a merchant, from the product list to payment, without the person opening the shop.
For your marketing: the purchase happens in the conversation with the agent. Your product feed, prices and availability have to be right where the agent reads them, or the purchase does not go through.
UCP, the Universal Commerce Protocol
Google's open standard, introduced in early 2026, through which agents find and compare products and buy from merchants. Merchants provide their range in one shared format, and agents from different providers read it.
For your marketing: the same product data that feeds Google Shopping today becomes the basis for agents. One set of data that is identical in feed, shop and product page pays off twice.
Agent card
An agent's business card in the agent to agent protocol: a small file on the server describing what the agent can do, which tasks it accepts and at which address it can be reached. Other agents read it before they talk to it.
For your marketing: once your brand has an agent of its own, the agent card is the entry through which others find it. Until then: readable first, then available for enquiries.
llms.txt
A convention from 2024: a text file in the root of a website that tells language models in a few lines what the site is about and where the important content lives. Comparable to robots.txt, but as a signpost rather than a list of restrictions.
For your marketing: an llms.txt does not replace product data, but it makes it easier for models and agents to get started on your site. hurra.com provides one.
The terms are means, not ends. What counts: product data, structured data and product pages an agent understands, plus measurement that recognises agent visits. More on the agent to agent page.