For two years, “AI will shop for us” lived in keynote speculation. Then Google published its Rethink Retail 2026 hub and quietly shipped the vocabulary of a different era: Business Agents that answer shoppers’ questions inside ads, a Universal Commerce Protocol for machine-to-machine transactions, product feed attributes designed for conversation, and a metric called AI share of voice. None of those are features for humans clicking ads. They are infrastructure for agents doing the shopping.
Agentic commerce, in one sentence: shopping where an AI agent does the finding, the comparing, and increasingly the buying on a customer’s behalf, with the human setting the goal (“find me winter boots under $150 that ship this week”) and approving the result. This guide explains what just became official, what it means for Canadian retailers on real budgets, and what to do now, while preparation is still cheap. We write this as an agency that runs eCommerce accounts daily and has been tracking the AI-search shift since it was measurable in our own analytics.
The Moment We Are Actually In
The honest picture, before any hype: humans still make the overwhelming share of purchases by hand, AI referral traffic is still a minority of most stores’ visits, and nobody’s holiday season will be decided by autonomous agents this year. And yet the direction is unambiguous, and the numbers behind it are the same ones reshaping search generally. ChatGPT reached roughly 900 million weekly active users by early 2026, its outbound referral traffic grew 206% during 2025 per Semrush, and the same research finds AI-driven visitors converting at about 4.4 times the rate of standard organic traffic, because a person who asked an assistant for a recommendation arrives pre-decided. OpenAI shipped shopping product-feed ads in ChatGPT in May 2026, the platform layer we covered in our ChatGPT Ads guide. And now Google, the company with the most to lose from getting this wrong, has published the retail rails. When both sides of the duopoly build for agents, the debate about whether is over; the only live question is when, and the sequencing answer for retailers is the same as it was for mobile: the ones who prepared during “early” were the ones who won “suddenly.”
What Google Actually Shipped, Piece by Piece
Four announcements from the Rethink Retail hub form the agentic skeleton, and each rewards a different piece of homework:
Business Agents. AI agents attached to your brand, set up through Merchant Center and now appearing in YouTube Ads, that answer shoppers’ questions and bridge, in Google’s words, inspiration and purchase confidence. Read that carefully: the agent speaks for your business, and what it can say is bounded by the product data, policies, and content you have given it. A thin feed produces a thin agent.
Universal Commerce Protocol (UCP). A standard, with an integration hub in Merchant Center for configuring and testing, for how commerce features and transactions work across surfaces. Protocols are how the web scales things: HTTP scaled pages, RSS scaled feeds, and a commerce protocol is what lets an agent complete a transaction without a human steering a checkout. Stores connected to the standard become shoppable by machines; stores that are not become invisible to them.
Conversational Attributes. New product feed fields designed for discovery in the conversational era, describing products the way people ask for them (“waterproof,” “fits a small kitchen,” “good for sensitive skin”) rather than only the way filters sort them. This is the tell about how agents choose: they read your data like a language, not a database.
AI Performance Insights. A report showing your brand’s share of voice in AI conversations, with steps to capture conversational demand. Google building a metric is Google declaring a battleground: AI share of voice is about to become a number your competitors track, the same way rankings were.
Around the skeleton sit the supporting pieces: Checkout Links compressing purchase to a single step, Ask Advisor letting merchants converse with their own feeds, and Google’s published AI optimization guidance for keeping sites visible as Search evolves. The full announcement rundown is in our Rethink Retail 2026 breakdown.
How an Agentic Purchase Actually Works
Walk one purchase through the new pipeline and the optimization surfaces reveal themselves. A customer tells an assistant: “I need running shoes for flat feet, under $180, delivered by Friday.” The agent interprets the need, including constraints no keyword ever carried. It retrieves candidates, reading product feeds, availability, pricing, and the conversational attributes that say “supportive” and “stability” in machine-checkable form. It compares against the constraints and against trust signals: reviews, return policies, merchant reliability, what the wider web corroborates about each brand. It asks your Business Agent the clarifying questions a floor salesperson would field: sizing, returns, arrival dates. Then it presents a short list, or with permission and a protocol like UCP plus stored payment, it completes the purchase. Notice what decided the sale at every step: data completeness, honest structured content, corroborated trust, and transactional readiness. Notice what appeared nowhere: your homepage carousel, your pop-up, and most of what retail marketing spends its meetings on.
Who Should Care How Much
- eCommerce brands: most, and first. Feeds, agents, and checkout protocols target you directly, and your product data is about to be interviewed by machines on every purchase.
- Retailers with local stores: the same shift arrives through “find me X nearby” agent queries, where profile data, inventory accuracy, and reviews decide who gets recommended, the agentic extension of the local SEO game.
- Service businesses: agents book as well as buy; the consultation-heavy version of this shift runs through the citation work in our GEO program, and it is further along than most owners think.
- Marketplaces-dependent sellers: watch closest of all, because agents that compare across the whole web weaken the walled gardens’ grip, an opening for direct brands that are machine-readable.
The Preparation Checklist: Cheap Now, Expensive Later
- Treat the product feed as your primary sales copy. Complete every attribute, write titles and descriptions the way customers talk, adopt conversational attributes as they roll out, and keep availability truthful to the hour. In agentic commerce, the feed is the pitch.
- Stand up your Brand Profile and Business Agent early. Bounded, tested, and fed with your real policies, so the agent speaking for you says true things.
- Make your trust machine-verifiable. Review volume and recency, consistent entity data everywhere, clear return and shipping policies in crawlable text: the corroboration layer agents check before recommending, the same layer that wins AI citations today.
- Evaluate UCP and frictionless payment. You do not need to be first through the protocol door; you need to know where the door is and what standing in front of it requires.
- Keep the human funnel excellent. Agents are additive, not a replacement this year: the store that converts humans well while becoming machine-readable wins both eras at once. Fast pages, honest pricing, and clean checkout serve both audiences, which is convenient, because they were already your job.
The Honest Caveats
What a trustworthy read of this moment includes: timelines are uncertain and Google’s own materials say evaluate, not migrate; protocols sometimes stall, get renamed, or lose to a rival standard, which is why the checklist above invests in things (feed quality, trust signals, tracking) that pay off regardless; agent adoption will vary wildly by category, with commodity repurchases automating years before considered purchases do; and measurement is genuinely immature, so treat AI share of voice as a directional instrument, not gospel. The reason to act anyway is asymmetry: every preparation step doubles as an improvement to your current-era store, so the downside of preparing early is roughly zero, while the downside of preparing late is watching agents recommend competitors whose feeds could answer questions yours could not.
Frequently Asked Questions
Shopping where an AI agent finds, compares, and increasingly completes purchases on a customer's behalf, with the human setting the goal and approving the outcome. Google's Business Agents, Universal Commerce Protocol, and conversational feed attributes are the first mainstream retail infrastructure built for it.
The infrastructure is real and shipping: Google's UCP hub and Business Agents, ChatGPT's product feed ads, and checkout-in-a-link formats all exist today. Autonomous purchasing at scale is not here yet, and pretending otherwise would be hype; building the readiness while it is cheap is just planning.
By reading structured product data against the customer's stated constraints, then checking trust: reviews, policies, entity consistency, and what independent sources corroborate. Complete, honest, machine-readable data beats persuasion, which is a genuine change in what retail marketing optimizes.
Google's standard for powering commerce features and transactions across surfaces, with an integration hub in Merchant Center for configuring and testing. Think of it as the plumbing that lets an agent transact with your store without a human driving the checkout.
No; it changes the website's audiences from one to two. Humans still browse, compare, and buy, and now machines read the same store on customers' behalf. Sites built with clean structure, honest data, and fast performance serve both; sites built entirely on visual persuasion serve one shrinking audience.
The first three checklist items: feed completeness, Business Agent setup, and machine-verifiable trust, all doable inside a quarter on a small team, all valuable even if agents took five more years to matter. The audit version of that sentence is free, and it is below.
