How Does Agentic Marketing Support the New Era of Agentic Commerce?

Artificial intelligence (AI) agents can now find, compare and buy products for people. In September 2026, Shopify let agents complete checkout on all of its eligible merchant stores. In the United States, the share of AI users who bought through AI rose from 10 percent to 17 percent in one year.

New protocols, shared rules that let an agent read a store and pay, make this possible. The Agentic Commerce Protocol (ACP) comes from OpenAI and Stripe. The Universal Commerce Protocol (UCP) comes from Google and Shopify. The Agent Payments Protocol (AP2) lets an agent pay on a person's behalf.

The protocols handle the transaction. They do not decide which agents you serve, what you offer, at what price, or under which rules. Those are still marketing decisions, and most marketing tools were built for a human buyer.

Each of those decisions gets harder when the buyer is software. Here is what shifts.

What changes when the buyer is an agent

• The persuasion surface goes away. Agents read structured data, reviews and price, not creative. In a 2026 Harvard Business Review study, only star ratings raised selection in a consistent way.
• The signal goes away. The shopping journey happens inside the assistant. You see only the order.
• The buyer keeps changing. Each AI model chooses differently, and every update can change the result.
• The offer becomes a negotiation. Agents compare offers in seconds. A weekly promotion calendar can't keep up.
• Governance gets harder. Which agents do you serve, what can they get, and who owns the customer?
• Measurement breaks. With no impressions or clicks, path-based attribution stops working.

Six capabilities to build

Each shift points to something your marketing system must do well:

1. A brand agents can read. One source of product truth, published to every protocol and assistant.
2. New signals. Agent catalog and cart requests, assistant recommendations, and order outcomes.
3. Continuous testing. Simulated shoppers run against each AI model, again after every update.
4. Real-time offers. A decision engine answers in seconds, inside the margin and brand rules your team sets.
5. Rules machines can enforce. Your team writes the policy. The system applies and logs it on every request.
6. Measurement that follows the agent. Share of agent recommendation, agent conversion rate and offer acceptance rate.

Consent runs through all six. Each agent request carries a person's permission for specific data, a specific purpose and a specific time. The system has to honor those limits.

One loop, not six tools

Thirty years ago, payment gateways let stores take money online, but didn't bring customers. Brands built a second system for that: analytics, ad technology, data platforms and customer relationship management (CRM).

Agentic commerce needs its own second system. The six capabilities work best on one data model and one loop: observe, decide, act and measure. Bought as separate tools, they become new silos. Built as one system, they move at machine speed while your team sets the rules.

For more depth on this topic, we invite you to read our full white paper, Why Agentic Commerce Needs a Marketing System.