Agentic Commerce in 2026: How AI Agents Buy Products

The process typically unfolds in four stages:

1. Intent capture. The shopper describes what they want in natural language. “I need a birthday gift for my sister, she likes skincare, budget $75.” The agent interprets this intent, extracting product category, price constraints, recipient preferences and urgency.

2. Discovery and matching. AI systems retrieve product information from web pages, feeds and merchant records. Clear descriptions and accurate attributes are what the system has to work with when it judges relevance to the questions shoppers ask.

3. Evaluation and recommendation. The agent compares options across several merchants, weighing the factors the shopper specified. It presents a short list with reasoning: “This moisturizer has 4.8 stars, ships free in 2 days and fits your budget at $68.” The shopper reviews and approves.

4. Transaction. Depending on the platform, the agent either directs the shopper to the merchant site to buy or completes the purchase with stored payment credentials, handles shipping selection and confirms the order inside the chat. Higher-value transactions require explicit shopper approval.

This four-stage flow runs on a growing set of open protocols. The Agentic Commerce Protocol (ACP), developed by Stripe and OpenAI, powers ChatGPT Shopping product discovery and the merchant handoff. Google's Universal Commerce Protocol (UCP), launched at NRF 2026, covers the full commerce journey from discovery to post-purchase. Anthropic's Model Context Protocol (MCP) provides the data connectivity layer that lets agents read real-time inventory, pricing and product details.

Shopper adoption is growing faster than mobile commerce did. In a Morgan Stanley AlphaWise survey of 2,000 U.S. consumers, 45% had used ChatGPT in the prior month (Morgan Stanley, Nov 2025). An IBM Institute for Business Value study found 45% of consumers already use AI for some part of the buying journey (IBM, Jan 2026). Adobe reported that 38% of consumers had used generative AI for online shopping (Adobe, 2025).

Platform infrastructure is live. ChatGPT Shopping recommends products to U.S. users and directs shoppers to merchant sites. Google AI Mode added agentic checkout. Microsoft Copilot supports checkout through ACP. Perplexity offers purchasing inside its answers. Each surface reads product information from pages, feeds and connected commerce platforms.

Open protocol standards reduce integration work. Before September 2025, reaching an AI shopping channel meant a custom integration for each platform. The launch of ACP (Sep 2025), the rapid enterprise adoption of MCP and the debut of UCP at NRF 2026 (Jan 2026) created a shared foundation.

The financial stakes are significant. AI-referred visitors completed purchases at a 38% higher rate than traditional search visitors (Adobe Analytics, Nov 2025), and AI-referred shoppers converted 42% better than other shoppers in the first quarter of 2026 (Adobe, Q1 2026).

November 2024. Anthropic launches the Model Context Protocol (MCP), the first open standard for connecting AI agents to external systems and data. MCP later becomes the foundation of every major commerce protocol that follows.

April 2025. Google launches Agent2Agent (A2A), the first interoperability protocol for AI agents to coordinate across systems. It is donated to the Linux Foundation by year end.

September 2025. Stripe and OpenAI release the Agentic Commerce Protocol (ACP) under the Apache 2.0 license and ChatGPT Shopping launches with its first merchants. Google announces AP2 the same month for secure agentic payments.

October to November 2025. Adobe Analytics records 805% year over year growth in AI-driven retail traffic during Black Friday 2025. Salesforce reports $67 billion in AI-influenced Cyber Week sales, and retailers with AI agent integrations see 7x the sales growth of those without (Salesforce, Dec 2025).

January 2026. Google unveils the Universal Commerce Protocol (UCP) at NRF 2026, co-developed with Shopify and a group of large retailers. UCP covers the full commerce journey from discovery through post-purchase, with more than 20 payment and retail partners at launch (Google, Jan 2026).

February to March 2026. Mastercard completes its first live agentic transactions in Singapore, South Korea and other Asia-Pacific markets. Visa launches its Trusted Agent Protocol commercially after sandbox piloting with more than 100 partners. The major card networks begin supporting agent-initiated payments at scale.

March 2026. OpenAI deprecates ChatGPT Instant Checkout. The model shifts from buying inside the chat to discovering in the chat and transacting on the merchant site, where brands keep the customer relationship, login data and loyalty engagement.

April 2026. American Express debuts its Agentic Commerce Experiences (ACE) developer kit and announces purchase protection for registered AI agent purchases, so all three major U.S. card networks now support agentic commerce. Microsoft adopts UCP in Merchant Center and wires Shopify catalogs into Copilot. Adobe Summit 2026 ships an MCP server for Adobe Commerce.

Commerce protocols and search visibility have different requirements. A protocol integration governs how an agent transacts with a merchant. Whether an agent recommends a product in the first place depends on the product information it can read from pages, feeds and connected commerce platforms, and on how well that information answers the questions shoppers ask. The AI visibility service tracks those appearances and improves the product information behind them.

The most consequential shift: in traditional ecommerce, brands compete for shopper attention through design, advertising and search ranking. In agentic commerce, brands compete for agent selection through data quality and catalog completeness. Making products discoverable and recommendable by AI agents is becoming as central as SEO was for the search engine era. The glossary entry on agentic commerce defines the related terms, and the industry labels generative engine optimization (GEO) and answer engine optimization (AEO) describe the same discipline from the content side.