Agentic Commerce: How to Market When AI Agents Do the Shopping

March 16, 2026 Marketing by Cliff Lin

Adobe measured a 693.4% year-over-year jump in traffic from generative AI tools to U.S. retail sites during the 2025 holiday season. Then, in early March, OpenAI pulled back Instant Checkout, the in-chat buy button it had launched five months earlier. Together, those two facts give a fair picture of agentic commerce right now. AI agents already shape what people buy, but nobody has settled where the purchase itself should happen.

McKinsey estimated in October 2025 that agentic commerce could orchestrate up to $1 trillion in U.S. retail revenue by 2030. My lesson from the past six months is narrower. Most of the marketing work agents reward is data work, and you can start it now.

What Agentic Commerce Looked Like in March 2026

Six launches between late September and mid-January set the shape of the market:

September 29, 2025: OpenAI launched Instant Checkout in ChatGPT with U.S. Etsy sellers. It also open-sourced the Agentic Commerce Protocol, which it built with Stripe.
October 14: Walmart announced that shoppers would be able to buy from it inside ChatGPT.
November 13: Google added agentic checkout to Search. You set a price alert, and Google can buy the item with Google Pay once you confirm. Wayfair, Chewy and Quince were among the first merchants.
December 10: Shopify announced Agentic Storefronts, which syndicate merchant catalogs to ChatGPT, Perplexity and Microsoft Copilot.
January 2026: Microsoft launched Copilot Checkout in the U.S. with PayPal, Shopify, Stripe and Etsy.
January 11: Google announced the Universal Commerce Protocol, co-developed with Shopify, Etsy, Wayfair, Target and Walmart, along with checkout inside AI Mode and Gemini.

Amazon took its own route. In November 2025, it sued Perplexity to stop the Comet browser agent from shopping on Amazon. At the same time, it kept building its own assistant. Amazon says more than 300 million customers used Rufus in 2025. It credits Rufus with nearly $12 billion in incremental annualized sales.

Then checkout wobbled. In early March, The Information reported that OpenAI was scaling back Instant Checkout. OpenAI said the feature is moving to apps. Shoppers will pay inside a retailer’s ChatGPT app or finish on the retailer’s site. So discovery inside the chat keeps growing, while the companies are still working out the buy button.

Product Feeds for AI Shopping Agents

Agents skip your homepage design and read data instead: titles, attributes, prices, stock, reviews and policies. Google’s Shopping Graph holds more than 50 billion product listings, and Google says 2 billion of them update every hour. OpenAI’s launch post lists what ChatGPT weighs when several merchants sell the same item. The factors include availability, price, quality and whether the merchant is the primary seller. Each of those signals comes from your data.

What to Fix in the Feed

Titles and descriptions that carry the attributes people type into a chat, such as size, material, use case and compatibility.
GTINs on every product that has one, so an agent can match your listing to the same item elsewhere.
Price and availability that update as fast as your inventory changes.
Shipping cost, delivery speed and return terms in structured fields, instead of buried in a policy page.
Ratings and review counts wherever the platform accepts them.

Google also said in January that it is adding dozens of new Merchant Center attributes for conversational shopping. They cover answers to common product questions, compatible accessories and substitutes. Similarly, Shopify asks merchants using Agentic Storefronts to load policies, FAQs and brand voice into its Knowledge Base app. Both point to the same job: write down the answers your best salesperson would give.

Structured Data on Your Own Site

Agents that browse still read your pages. Schema.org Product and Offer markup gives them the same facts in a format they can parse. Include price, availability, shipping details and return policy. Keep the markup in sync with the feed. If you have already worked on making content discoverable in AI tools, treat this as the product-level version of that work.

What Agentic Commerce Changes for Brands

When an agent builds a shortlist, the shopper may never see your ad, your landing page or your packaging. As a result, agentic commerce puts more weight on two places where brand still shows up.

The first is the request itself. A shopper who names your brand in the prompt has already chosen you. The agent only has to find the best offer. That demand gets built outside the chat, through distinctive products, word of mouth, reviews and press. I’ve argued before for building a semantic brand footprint: consistent, specific mentions of what you are known for across the web. The models behind shopping agents learn from those same mentions.

The second is the evidence an agent weighs when it compares options: ratings, review volume, return terms, delivery speed and price. McKinsey’s report suggests brands design the agent experience as carefully as the customer experience. In plain terms, make your product the easiest one for software to recommend with confidence.

Search and Ads Inside AI Shopping

Paid placements are arriving in AI shopping, slowly and with guardrails:

Google’s Direct Offers pilot shows exclusive discounts in AI Mode when Google’s AI judges them relevant. Petco, e.l.f. Cosmetics, Samsonite and Rugs USA are early testers.
On January 16, OpenAI said it would test ads at the bottom of ChatGPT answers. The test covers logged-in U.S. adults on the Free and Go tiers. OpenAI also said the ads would carry labels, sit apart from the answer and not influence it.
Google’s Business Agent lets shoppers chat with a retailer’s own branded agent on Search. Lowe’s, Poshmark and Reebok were among the first to use it.

For search teams, this looks like zero-click search taken one step further. The agent answers, compares and sometimes buys without a site visit. So set up a separate analytics segment for referrals from AI assistants. Also watch branded search volume as a signal that people still ask for you by name.

Measurement needs a change too. A purchase completed inside a chat may never touch your web analytics. Shopify says every Agentic Storefront order flows into the merchant’s admin. So reconcile at the order level instead of the session level.

Checkout Options and Commerce Protocols

Checkout is the least settled layer of agentic commerce. Two open standards compete for the plumbing. One is the Agentic Commerce Protocol from OpenAI and Stripe. The other is Google’s Universal Commerce Protocol, which Google says works with its Agent Payments Protocol (AP2) and with MCP.

My advice is to let your commerce platform and payment provider carry the protocol work. Shopify, Stripe and PayPal already connect to several agents. Then put your own effort into what carries over to any agent: clean product data, clear policies and fast fulfillment. Also watch who holds the customer. Google and Shopify both say the retailer stays the seller of record. That keeps the customer relationship with you.

An Agentic Commerce Checklist for Next Quarter

Audit your feed for missing GTINs, vague titles and stale prices, starting with your top 100 sellers.
Add structured shipping, return and warranty data to the feed and to the Product markup on your site.
Turn on the agent channels your platform supports. Then search for your own products in ChatGPT, Gemini and Copilot to see how they appear.
Create an analytics segment for AI assistant referrals and reconcile it against orders each month.
If you join the Direct Offers pilot or the ChatGPT ad test, set up a holdout so you can measure lift.
Keep funding brand demand outside the chat, because that is what gets your name into the prompt.