What is agentic commerce? The 2026 primer for brands

Strip the buzzword and agentic commerce describes a transfer of labor. Online shopping has always required a human to do the work of the funnel: type the search, open the tabs, compare the specs, read the reviews, enter the card number. Agentic commerce hands those steps to software. The buyer expresses an intent in plain language, and an AI agent performs the research, forms a recommendation, and, on the surfaces where checkout is wired up, completes the transaction with the buyer's approval.

The definition matters because it separates two things that get conflated. The first is AI-assisted discovery, where an assistant recommends products and the human clicks out to buy. That has been growing since ChatGPT added search. The second is agentic execution, where the transaction itself happens through the agent. The second arrived in production on September 29, 2025, when OpenAI and Stripe launched Instant Checkout in ChatGPT for US users on Free, Plus and Pro plans, starting with US Etsy sellers. Wall Street registered the shift the same day: CNBC reported Etsy's stock jumped 16 percent on the announcement.

Both halves share one dependency that brands consistently underweight. Whether the agent merely recommends or fully transacts, it chooses from the options the AI answer surfaces. A brand missing from the answer is missing from the consideration set, and no amount of checkout plumbing fixes that. If the visibility side is new to you, the foundation is covered in what is AI visibility.

Follow one purchase end to end and the moving parts become concrete. A buyer tells ChatGPT they want a gift for a ceramicist friend, under 75 dollars, shipped within a week. The agent runs retrieval across product feeds and web sources, assembles a shortlist, and presents product cards with prices, images and availability. OpenAI states these results are organic and unsponsored, ranked on relevance. The buyer picks one. If the merchant supports Instant Checkout, a buy button appears in the conversation; the buyer confirms, payment executes through a Stripe-issued token, and the merchant fulfills the order exactly as if it came from their own site.

Google's version has the same skeleton with different plumbing. Agentic checkout began rolling out in November 2025 with Wayfair, Chewy, Quince and select Shopify merchants, including a striking capability: the buyer can ask Google to buy an item automatically when its price drops to a target, paying with Google Pay. In February 2026, checkout powered by the Universal Commerce Protocol reached AI Mode, making Etsy and Wayfair products buyable inside AI search itself, with Shopify, Target and Walmart named as next. How AI Mode composes those answers in the first place is covered in our Google AI Mode guide.

Notice what the merchant controls in this flow and what it does not. Feed accuracy, inventory freshness, pricing and checkout support are the merchant's to manage. The shortlist itself belongs to the model. That asymmetry is the defining commercial fact of agentic commerce, and it is why the discipline of tracking and improving your presence in AI answers has moved from marketing experiment to revenue infrastructure.

Every platform shift produces its standards war, and this one produced five protocol families in about a year. ACP, the Agentic Commerce Protocol from OpenAI and Stripe, defines how an agent executes a checkout inside ChatGPT, using a shared payment token so the agent never touches raw card credentials. UCP, the Universal Commerce Protocol, is Google's equivalent, co-developed with Shopify, Etsy, Wayfair, Target and Walmart and endorsed by more than 20 companies across payments and retail, including Visa, Mastercard, Stripe and Adyen.

Underneath the checkout layer sits the trust layer. AP2, the Agent Payments Protocol announced by Google in September 2025, uses cryptographically signed mandates to prove a human authorized what the agent bought. Visa's Trusted Agent Protocol, launched October 14, 2025, and Mastercard's Agent Pay solve the adjacent problem on the card rails: distinguishing a legitimate shopping agent from a bot attack. Merchants rarely implement any of this by hand; commerce platforms and payment providers carry the integration. The full breakdown, including which protocols a brand can safely ignore, is in ACP, UCP, AP2 and friends.

The forecast that anchors most board conversations comes from McKinsey: agentic commerce could orchestrate up to 1 trillion dollars of US retail revenue by 2030 and 3 to 5 trillion dollars globally. Any 2030 forecast is a scenario rather than a fact, but the present-tense data pointing the same direction is harder to dismiss. McKinsey's own consumer research finds 38 percent of European consumers already use gen AI tools to research products and decide what to buy.

The strongest present-tense evidence is Adobe's. Drawing on more than a trillion visits to US retail sites, Adobe Analytics measured AI-driven traffic up 393 percent year over year in Q1 2026. Quality rose with volume: by March 2026, AI-referred visitors converted better than traffic from traditional channels, a full reversal from a year earlier when they converted about 38 percent worse, and they generated roughly 37 percent more revenue per visit than non-AI traffic. Shoppers arriving from an AI answer land pre-persuaded, because the comparison work happened in the conversation. We unpack that dataset line by line in AI referral traffic in 2026.

One more number completes the picture. Adobe also graded retail sites on machine readability and found product detail pages scored just 66 percent on average, the worst of any page type. The demand side of agentic commerce is compounding while the supply side, the pages and data agents need to read, lags behind. Gaps like that are where early movers take share.

Retail strategy has always been a fight for shelf position, and every era redefines the shelf. In stores it was eye-level placement. In search it was page one. In agentic commerce the shelf is the answer itself: the three to five products or brands an AI names when a buyer states an intent. Everything downstream of that moment, the product card, the checkout button, the signed payment mandate, only executes a decision the answer already made.

This is why the checkout protocols, for all their engineering elegance, are the smaller half of the story for most brands. Supporting ACP and UCP is increasingly a checkbox your commerce platform handles. Being the brand the agent selects is a competition, and it is won upstream: in the product data agents parse, the third-party sources engines cite, and the content that answers buying questions the way engines like to quote. The mechanics of that competition are the subject of how ChatGPT decides which brands to recommend.

There is also a measurement consequence. Shelf position in a store is observable; your presence in AI answers is probabilistic and shifts without notice. The same buying question can name you on Monday and omit you on Wednesday. Treating a single spot check as the truth is the most common mistake teams make entering this channel. The honest approach is repeated runs on a schedule with stored answers you can audit, which is precisely the loop Reachroller automates: it tracks the buying questions in your category across engines, shows the raw answer behind every score, and generates the publish-ready content fix for each question you lose.

Weeks one and two: audit reality. Ask the engines your buyers use the ten questions that most often precede a purchase in your category. Record which brands get named and which sources get cited. Check whether your robots.txt admits AI crawlers such as OAI-SearchBot, and run your key product pages through a machine-readability check. Most teams discover they are invisible for the majority of their buying questions, which is unwelcome news and also a map.

Weeks three to six: connect the plumbing. If you sell on Shopify or a comparable platform, enable the ChatGPT and Google integrations as they reach your tier, and apply directly through OpenAI's merchant program if you qualify. Fix the product feed first: accurate availability, current pricing, complete attributes. OpenAI has said merchants are ranked on availability, price, quality and whether they are the maker or primary seller, so the feed is a ranking input rather than an operational afterthought. The step-by-step version is in how to get your products into ChatGPT Shopping.

Ongoing: run the visibility loop. Track your buying questions weekly, publish an answer-shaped page for each one you lose, get it indexed, and recheck whether the answer flipped. This is the part that compounds, and it is deliberately founder-sized in Reachroller: Starter is 29 dollars per month, every tracked question links to its raw answers, and each generated fix ships with slug, metadata, schema and indexing steps. The honest caveat stands: it is a young product, ChatGPT tracking is live today and the remaining engines are rolling out.

A primer that only sells the upside is a pitch, so take the frictions seriously. The first is habit. Most consumer purchases are reorders and routine restocks, categories where the buyer already knows the answer and an agent adds little. Agentic commerce wins research-heavy purchases first: the gift, the replacement appliance, the first stroller, the category the buyer enters once every few years. That is a large slice of retail margin, but it is a slice, and forecasts that assume every transaction becomes agentic should be discounted accordingly.

The second friction is trust, on both sides of the transaction. Consumers must be comfortable letting software spend money, and Adobe's survey suggests the early adopters already are: 39 percent of US consumers reported using AI for online shopping, and 85 percent of that group said it improved the experience. Merchants and banks must trust agent traffic, which is what the entire AP2 and card network protocol effort exists to manufacture. Progress here has not been a straight line; CNBC reported in March 2026 that OpenAI's first run at agentic shopping stumbled and the company was iterating with larger retail partners. Platform shifts routinely stumble in their first year and win anyway, but the stumbles are real and worth pricing in.

Here is why none of this changes the recommendation for brands: the preparation is identical either way. If in-chat checkout ramps slowly, AI-assisted discovery is already at scale, sending traffic that converts better than your other channels, and it rewards the same feed hygiene and answer visibility. If checkout ramps fast, the same work is what puts your product behind the buy button. There is no scenario on the table where being invisible in AI answers ages well.