A shopper opens Gemini on her phone. She types, “Find me waterproof hiking boots under $120, size 8, in stock near me.” She never opens Google Search. She never visits a store website. An AI agent reads product feeds, compares prices and stock, and puts the best pair in a cart. The sale happens inside Google. The store that wins never gets a click.
That is agentic commerce, and it is already live in the United States. It changes what SEO for an online store has to do.
Agentic commerce is shopping where an AI agent finds, compares, and buys products for a shopper. The agent reads product feeds and structured data, checks price and stock, and completes the purchase inside Google Search, AI Mode, or the Gemini app. Google Universal Cart, launched in the US on October 15, 2026, is the cart where that purchase happens.
What is agentic commerce?
A shopper states what they want and the budget. The agent does the rest. It searches product data, compares options, checks stock, and either recommends a product or buys it. The shopper approves the purchase or sets limits in advance, and the agent stays inside those limits.
This is different from normal online shopping. A human browses pages, reads descriptions, and clicks through a checkout. An agent does not browse. It calls product feed APIs, reads Google Merchant Center data, and parses JSON-LD Product schema on the page. It compares price, specifications, ratings, and availability as numbers, not as marketing copy. When the structured data is incomplete, the agent falls back to reading prose, reviews, and comparison content. That fallback is slower and less reliable, so products with complete data win first.
Google built the biggest version of agentic commerce so far. Universal Cart launched in the US on October 15, 2026, in Google Search, including AI Mode, and in the Gemini app. Google previewed it at I/O on May 19, 2026. The cart holds products from many retailers in one place. It tracks prices and stock, sends alerts when a price drops or an item comes back, surfaces promo codes, and suggests the best card to use from Google Wallet. It runs on the Shopping Graph, which holds more than 60 billion product listings, according to Google.
Black Friday falls on November 27, 2026, and Cyber Monday on November 30. Adobe forecasts 12.9 billion dollars in US online spend on Black Friday and 15.1 billion on Cyber Monday. Adobe also forecasts AI traffic to US retail sites will rise 130 percent year over year from November 1 to December 31. The holiday season is the first real test of agentic commerce at scale, and the stores with clean product data will be the ones the agents recommend.
How agentic commerce changes who wins the sale
In traditional ecommerce SEO, the goal was a click. Rank on page one, win the click, and the product page did the selling. Agentic commerce removes the click from the middle of the sale. The agent evaluates products itself and returns a short recommendation, or it buys. A page that ranks first for a keyword but lacks machine-readable data can be skipped entirely. A product with complete data can win even when its page ranks lower.
Three things decide whether an agent picks your product in agentic commerce. First, can it find your product data at all. That means a Merchant Center feed, Product schema on the page, or an API it can query. Second, can it trust what it finds. Price, stock, shipping cost, and return terms must be current and consistent everywhere the agent looks. Third, does it have a reason to prefer your product over the same item at another store. Ratings, review count, delivery speed, and total price all feed that choice.
The agent also compares prices across retailers in the same cart. Universal Cart puts your product beside the same product from other stores. If your price is higher, the agent sees it at once. If your stock data is stale and the item is actually sold out, the agent learns that too. Product data quality is now a ranking factor in practice, even though no one at Google calls it that.
There is one more shift worth naming. A human search returns ten links. An agent returns one to three options. There is no page two in agentic commerce. If your product is not in the agent’s shortlist, the shopper never hears your name. That is why the seven steps below start with data and end with measurement, not the other way around.
Google Universal Cart: the checkout Google built
A shopper can add a product while browsing AI Mode and come back to it later in Search or Gemini without losing it. The cart watches prices and availability and alerts the shopper when a price drops or an out of stock item returns. It also surfaces promo codes for items already in the cart.
Checkout works two ways, and the retailer stays the seller of record on both. At participating merchants, the shopper pays with Google Pay in a few taps inside the cart. Otherwise, the items move into the merchant’s own cart and the shopper finishes the purchase on the merchant’s site. Either way, the merchant provides order confirmations and customer support.
Google named Nike, Sephora, Target, Ulta Beauty, Walmart, and Wayfair among the participating merchants, along with many Shopify retailers. Newegg has since joined. Eligible Shopify stores were switched on for checkout in Google by default in September, so many Shopify merchants are already part of it whether or not they noticed. Universal Cart is US only for now, though UCP checkout without the cart reached Canadian shoppers on November 4, 2026.
Checkout runs on the Universal Commerce Protocol, or UCP. Google announced UCP on January 11, 2026, and co-developed it with Shopify, Etsy, Wayfair, Target, and Walmart. More than 20 other companies endorsed it, including Adyen, American Express, Mastercard, Stripe, and Visa. UCP defines how Google’s shopping surfaces talk to a retailer’s systems: searching the catalog, building a cart, linking accounts, checking out, and managing orders. The protocol is versioned by date, and version 2026-04-08 is the minimum for Google’s Cart API. You can read the specification at ucp.dev.

Agentic commerce: the 7-step system
Start with data, not with new tools. Most of the work below costs time rather than money, and the order matters. A feed fix in step one makes every later step work better.
Step 1: Fix your product feed first
Your Merchant Center feed is the first thing an agent reads. It is also the first place errors show up. Open the feed diagnostics and fix every critical error before you do anything else. The checks that matter most for agentic commerce are simple. Every product needs a valid GTIN where one exists. Titles should follow a brand, model, and key specification format, because agents match titles against natural language requests. Categories must use Google’s taxonomy, not your own labels. Images need to be at least 800 by 800 pixels.
Price and availability need to be fresh. Update them at least every four hours during a sale, and include sale prices with start and end dates. A sale window sent without a timezone is read as UTC, so set the timezone on every promotion. Shipping costs must be explicit for each zone, not hidden until checkout. If your feed says one price and your site says another, the agent trusts neither. Fix the mismatch at the source, not in the feed alone.
Step 2: Complete your Product schema
Product schema in JSON-LD is the vocabulary AI systems use to understand what you sell. Basic schema covers name, price, and availability. That is enough for a standard rich result. For agentic commerce, where an agent has to choose between ten similar products, it is too thin.
Add the properties agents actually use. GTIN lets the agent match your product against other databases and review sources. Brand as an object, not a plain string, helps brand-specific queries. AggregateRating with a review count gives the agent a quality signal it can compare. ShippingDetails and hasMerchantReturnPolicy answer the two questions shoppers ask most after price. Color, material, and size connect your product to queries like “faux leather jacket” or “size 10 running shoes.”
For products with variants, use ProductGroup schema instead of marking up each variant as a standalone product. That keeps price and stock consistent across the group. Validate the result with Google’s Rich Results Test. A schema error is invisible to a human visitor and fatal to an agent. Hashmeta’s agentic commerce guide reports that 65 percent of pages cited by Google AI Mode and 71 percent of pages cited by ChatGPT include structured data. That is the gap between being evaluated and being skipped. My structured data guide covers the schema types that still pay off, including Product markup.
Step 3: Let the AI crawlers in
Access is the prerequisite for everything else, and it is blocked more often than merchants realize. Open your robots.txt file and check for rules that block AI bots. The ones that matter are OAI-SearchBot and GPTBot for ChatGPT, Google-Extended for Google’s AI surfaces, and PerplexityBot. A disallow rule added during a security tightening exercise can silently cut off an entire AI platform from your catalog.
Firewalls and content delivery networks are the other common culprit. Many are configured to block non-standard bots by default. An AI crawler can be rejected before it ever reaches your robots.txt. If you use a web application firewall through Cloudflare or Sucuri, whitelist the AI crawler user-agents explicitly instead of leaving them under the default unknown bot ruleset.
Then check how your product pages render. Many ecommerce platforms load price, availability, and reviews with JavaScript in the browser. AI crawlers often read the page shell instead of the real content. Server-side rendering solves this, and it should be confirmed for every key product page. If your specifications live inside JavaScript accordions or image-based spec sheets, move them into plain HTML tables or definition lists. Agents extract attributes from tables far more reliably than from prose. The same rendering checks apply if you want to rank in Google AI Mode, where the citation also depends on readable content.
Step 4: Switch on Google Universal Cart in Merchant Center
This is the step most guides skip, because it is new. The UCP integration hub in Merchant Center lets you choose native checkout on Google, cart transfer to your own site, or identity linking, and validate each one in a sandbox before it goes live.
Start with cart transfer. It needs only the UCP Cart API, version 2026-04-08 or later, and a fast-track check. The shopper’s items move into your own cart, so you keep the checkout experience, the upsell, and the customer data. That makes it the safe first move for a small store.
Add native checkout later, once cart transfer works. Native checkout needs three REST endpoints, order webhooks, a Google Pay and Wallet Console account, and a Google review. Schedule that work for January, after the holiday peak. Do not try to build both paths in the same week as Black Friday.
For Shopify stores, check whether checkout in Google is already on. Eligible stores were switched on by default in September. If yours is on, your job is data quality, not setup. Confirm it in Merchant Center rather than assuming. My Shopify SEO guide covers the store-level fixes that pair with this step.
Step 5: Make your price match everywhere
An agent compares your price in three places: your feed, your Product schema, and your checkout. If the three disagree, the agent penalizes you or drops you. PageTraffic’s agentic commerce guide states this plainly: price differences among schema, feeds, and checkout endpoints result in instant trust penalties and disqualification from recommendations.
The fixes are unglamorous. Keep one source of truth for price and stock in agentic commerce, and let the feed, the schema, and the site read from it. Show the total price the agent will compare, including shipping, not a base price that grows at checkout. Use PriceSpecification schema for discounts and tiered pricing so the agent evaluates the real net price. Put sale start and end dates on every promotion, with the timezone set. When a product goes out of stock, update the feed and the schema at the same time, not hours apart.
Step 6: Build the trust signals agents read
Agents do not only read your site. They read third-party sources: review platforms, editorial roundups, Reddit threads, and YouTube videos that mention your products. Each platform weights these sources differently, but a gap in any of them weakens your visibility in agentic commerce.
Reviews matter most. Ask buyers for reviews that mention the use case, not just a star rating. “Fits a wide foot, true to size” helps an agent match a query. “Great product” does not. Keep your review count and average rating in your schema, and keep them honest. A rating with no review count behind it is ignored.
Editorial mentions help too. A product that appears in a roundup with real detail earns a stronger signal than one that appears only in your own store. That is a slow build, and it starts with products that are genuinely easy to recommend: accurate specs, clear photos, honest stock, and fast delivery.
Step 7: Measure agent traffic, not just clicks
Clicks stop being the main metric in agentic commerce, because the sale happens inside Google. The metrics that matter now are different. Track AI citation rate, which is how often your brand appears in AI shopping responses. Track feed approval rate, the share of products approved across Merchant Center and the other portals. Track schema completeness, the coverage of GTIN, rating, shipping, and return policy across your catalog. Track AI referral revenue, the sales you can trace to AI citations against your baseline.
You can measure most of this with tools you already have. Check server logs for AI crawler user-agents to see which agents visit and which product pages they read. Watch Merchant Center diagnostics for feed approval and disapproval trends. Watch Google Search Console for shopping queries that bring impressions without clicks, because an agent may have read the listing and bought without a visit. Salesforce reports that 73 percent of consumers now use AI agents or AI-powered assistants at some point in their purchase journey. That share will show up in your logs before it shows up in your revenue reports.
Google shopping optimization for agents
Better Shopping ads used to be the whole job. For agentic commerce, the job is making your feed easy for an agent to match against a request. Titles carry the most weight. Put the brand first, then the model, then the single specification that separates your product from the rest.
Take a title like “Nike Pegasus 41 running shoes, men’s size 10.” It matches a request better than “Men’s running shoes, lightweight, breathable, best seller.” The first title gives the agent a brand, a model, and a size. The second gives it adjectives.
Use Google’s product taxonomy at three or more levels deep, and use custom attributes for the details your category needs, like material, compatibility, or certification. Agents can query custom fields directly. The same detail buried in a paragraph is invisible to them.
Comparison content helps in agentic commerce as well. Publish a short guide that compares your top products in a table, with columns for the products and rows for the attributes. Agents extract a full comparison from a table in one pass. The same information written as prose forces the agent to infer the comparison, and inference is where errors creep in. DigitalApplied’s guide reports a 40 percent improvement in attribute extraction accuracy from HTML tables versus the same information in paragraph form.
Agentic commerce platform choices: Shopify, ChatGPT, Perplexity, Copilot
Google is not the only platform building agentic commerce. Each one has its own protocol, and each one reaches different shoppers. An agentic commerce platform is simply the place where the agent meets the shopper, and the setup work is the same everywhere.
Shopify is the easiest starting point for most stores. Its catalog and Shop Pay are open to developers building AI shopping agents, and eligible stores already have Google checkout switched on. If you run Shopify, your first job is the feed and the schema, not a new platform.
ChatGPT uses its own merchant portal and the Agentic Commerce Protocol, or ACP, which OpenAI built with Stripe. It powers product recommendations and in-chat checkout. If your customers already ask ChatGPT for buying advice, submit your catalog to the merchant portal.
Perplexity runs a merchant program for live product and inventory data. Microsoft Copilot uses Microsoft Merchant Center and UCP transactions. Both matter less today than Google and ChatGPT, but the setup is the same work: a clean feed, complete schema, and open crawler access. Do the work once and it serves every platform.
The protocols will keep multiplying. UCP, ACP, MCP, and the Agent Payments Protocol all overlap, and Google says UCP is compatible with the Agent Payments Protocol, Agent2Agent, and the Model Context Protocol. Do not pick a protocol to bet on. Make your product data clean and let the protocols come to you.
What agentic commerce costs and how long it takes
Most of the work is time, not money. A small store can run the whole system in two to three weeks without buying a new tool. The table below shows a realistic timeline.
| Week | Focus | What gets done |
|---|---|---|
| Week 1 | Feed and schema | Audit the feed, fix GTIN gaps, and complete Product schema on the top 20 products |
| Week 2 | Access and cart | Open AI crawler access, fix robots.txt and firewall rules, and switch on cart transfer in Merchant Center |
| Week 3 | Price and measurement | Fix price mismatches, publish one comparison guide, and set up measurement in server logs and Merchant Center |
| After that | Maintenance | Feed freshness, review requests, and a monthly schema check |
The costs that do exist are small. A developer may need a few hours for server-side rendering or UCP endpoints. A feed tool may be worth paying for once the catalog passes a few hundred products. Native checkout is the only step with a real project cost, which is why it waits until January. Skip any vendor that promises agentic commerce visibility for a monthly fee before your feed is clean. No tool can sell an agent a product it cannot read. If you want a broader baseline first, my ecommerce SEO checklist covers the store-wide fixes this system builds on.
5 mistakes that get your products skipped
All five mistakes below have the same result in agentic commerce: the agent skips your product and recommends a competitor.
- A price in the feed that does not match the price on the site. Agents compare all three sources and drop the product when they disagree.
- Stale stock. An out of stock item that still shows as available wastes the agent’s time and your credibility.
- Blocked crawlers. A robots.txt rule or a firewall default can hide your whole catalog from an AI platform.
- Specifications trapped in JavaScript or images. If the agent cannot read the spec, it cannot match the product to the request.
- No GTIN. Without a global identifier, your product cannot be matched against other databases, reviews, or price comparisons, so the agent evaluates it alone and usually passes.
Frequently Asked Questions about agentic commerce
What is agentic commerce protocol?
There is no single protocol with that name. The term usually means Google’s Universal Commerce Protocol, or UCP, the open standard that lets AI agents search a catalog, build a cart, check out, and manage orders. OpenAI and Stripe built a separate protocol called ACP for ChatGPT checkout. When someone asks about the agentic commerce protocol, they usually mean UCP, which Google announced on January 11, 2026.
What is agentic shopping?
Agentic shopping is the shopper’s side of agentic commerce. A person tells an AI assistant what they want, and the assistant finds options, compares them, and either recommends one or buys it. Google Universal Cart is the best-known example. The shopper’s items sit in one cart across Search, AI Mode, and Gemini, and checkout happens with Google Pay or on the merchant’s site.
How does agentic commerce work?
It works in four moves. The shopper states what they want and the budget. The agent reads product feeds, Merchant Center data, and Product schema to find matches. It compares price, stock, ratings, and delivery, then recommends a product or buys it inside the limits the shopper set. The merchant stays the seller of record and handles the order.
Can you give me an example of agentic commerce?
A shopper asks Gemini for waterproof hiking boots under $120 in size 8. The agent checks feeds from several stores, finds a pair in stock at the best total price, and puts it in Google Universal Cart. The shopper pays with Google Pay without opening the store’s website. The store still owns the sale and ships the boots.
What are the differences between agentic AI and agentic commerce?
Agentic AI is the general idea: AI systems that take action on their own, from booking travel to writing code. Agentic commerce is that idea applied to shopping. The agent’s actions are finding products, comparing them, and buying. Every agentic commerce system uses agentic AI, but most agentic AI has nothing to do with shopping.
What are the risks of agentic commerce?
Three risks matter most. The agent can recommend the wrong product when the data is thin or stale. A price mismatch between the feed and the site can mean the shopper pays a different price than the agent quoted. And the merchant can lose the direct relationship with the shopper when the sale happens inside Google. Clean data, price checks, and cart transfer all reduce those risks.
The bottom line
Agentic commerce is not a future trend. Google Universal Cart is live, Black Friday is five days away, and AI agents are already choosing which products to recommend. The stores that win are the ones with clean feeds, complete schema, open crawler access, and prices that match everywhere. Do the seven steps in order, start with cart transfer, and measure what the agents do.
If you want help getting your store ready for agentic commerce, see the services I offer and get in touch. The work starts with a feed audit, and the best week to do it is this one.