Agentic Commerce Explained for Marketers: How AI Agents Discover, Compare and Buy

Agentic commerce went from conference slide to checkout button in about a year. Google’s AI Mode can now complete purchases and hotel bookings. Microsoft’s Copilot has a checkout. Google, Shopify, Walmart, Target, and others built a shared protocol so AI agents can buy from retailers.

Then the most famous launch stumbled. OpenAI’s Instant Checkout, introduced in September 2025, was deprioritized in March 2026.

If you’re a marketer or a founder, both halves of that story matter. This guide covers what agentic commerce is, what’s actually live, what stalled and why, how agents decide what to buy, which purchases they’ll take over first, and what to change in your marketing before the holiday season.

What Is Agentic Commerce?

Agentic commerce is shopping in which an AI agent carries out part or all of the purchase journey for a person: finding options, comparing them against that person’s needs, and in some cases completing the purchase within limits the person sets.

Google describes it as commerce “where AI completes tasks on people’s behalf” (Google).

The key word is delegation. Classic ecommerce gives shoppers tools: search bars, filters, comparison tables. Agentic commerce gives shoppers a helper that uses those tools for them.

You’ll also hear the phrase zero click commerce. It describes the end point of that shift: discovery, comparison, and payment all happen inside an AI surface, and the merchant fulfills the order behind the scenes. In the major programs live today, the retailer stays the seller or merchant of record and handles fulfillment and support.

How Agentic Commerce Works: Discover, Compare, Decide, Pay?

The scale of the comparison step is easy to underestimate. Google says its Shopping Graph holds more than 50 billion product listings, with 2 billion of them updated every hour (Google). That’s the pool your product sits in before a shopper ever sees your site.

It’s also worth knowing that “agentic” rarely means unsupervised today. Google’s agentic checkout, for example, asks for permission and only buys after the shopper confirms the purchase and shipping details (Google). The agent does the legwork; the person still says yes.

One request, four stages: an example

To make this concrete, here’s an illustrative request and what each stage demands of the brands involved. It’s a composite of documented behaviors, not a single platform’s exact process.

A shopper asks: “Find me a hard shell carry-on suitcase under $250 that fits most US airline size limits, with a good warranty, delivered before Friday.”

• Discover: The agent searches for carry on luggage and pulls candidates whose data mentions hard shell construction, dimensions, and price. A brand whose feed lists dimensions only inside an image never enters the list.
• Compare: It lines up dimensions, weight, warranty length, reviews, price, and delivery dates. A brand with a lifetime warranty buried in a PDF loses to one that states “10 year warranty” in plain text.
• Decide: It picks the option that meets every constraint with the strongest reviews and the most reliable delivery promise, and explains why in two sentences.
• Pay: If the retailer supports checkout inside the surface, the shopper confirms and pays with a saved wallet. If not, the agent sends the shopper to the retailer’s site with the choice already made.

Notice that the shopper never visited the losing brands. Their only chance to compete was the data the agent could read.

What’s Live: An Agentic Commerce Timeline

Here are the moments that matter for marketers, in order.

Sources for each row: AP2, Instant Checkout, Razorpay, Google November 2025, PayPal and Perplexity, Copilot Checkout, UCP launch, OpenAI March 2026, UCP Tech Council, Cooley on the Ninth Circuit, AI Mode hotels, Business Standard, Google September 2026, dots.

What Stalled, and What It Teaches?

In March 2026, Search Engine Land, citing The Information, reported that OpenAI was backing away from checkout inside ChatGPT because people researched products there but finished their purchases elsewhere, and only a small number of merchants were actively using native checkout (Search Engine Land).

OpenAI confirmed the change on March 24, saying the first version of Instant Checkout “did not offer the level of flexibility that we aspire to provide.” Merchants now use their own checkout while ChatGPT focuses on discovery. OpenAI expanded ACP so merchants can share product feeds and promotions, named Target, Sephora, Nordstrom, Lowe’s, Best Buy, The Home Depot, and Wayfair as integrated retailers, included Shopify merchants through Shopify Catalog, and launched an in ChatGPT Walmart experience with account linking and Walmart payments (OpenAI).

Trust is the other brake. Forrester found that between February and October 2025, US adoption of answer engines for product discovery barely moved, from 18% to 19%. By October 2025, only 8% of US online adults had used Instant Checkout, and 54% weren’t comfortable giving personal information to generative AI tools (Forrester via Forbes).

Infrastructure is the third. In the same Search Engine Land piece, Semrush’s Leigh McKenzie pointed out that normalizing product catalogs in real time across tens of millions of SKUs is a decade scale problem, one Google had already solved with Merchant Center, and that shoppers still default to checkout flows they already trust (Search Engine Land).

The lesson for marketers is simple. Discovery moved to AI faster than payment did. That is where most of your near term opportunity sits, and it’s why product data and brand trust deserve a budget before any checkout integration does.

How AI Agents Decide What to Buy?

No platform publishes a complete formula for agent recommendations. What the platforms document, plus what shows up consistently in testing, points to eight inputs.

1. Fit with the request. Agents match the constraints in a request, such as size, compatibility, budget, and use case, against product attributes. Google says conversational attributes submitted by lululemon were incorporated 50% of the time in relevant AI Mode product recommendations during testing (Google).
2. Fresh, accurate data. Price and stock have to match between your feed and your page. Agents compare in near real time, and a mismatch is an easy reason to choose someone else.
3. Reviews and ratings. Ratings, review counts, and what reviews actually say feed comparison tables and summaries.
4. Delivery and returns. Total cost and risk matter. Shipping fees, delivery dates, and return windows are part of the comparison.
5. Offers and loyalty. Google’s Direct Offers pilot lets advertisers present exclusive discounts in AI Mode (Google), and Google Ads is now testing a Direct Offer asset for AI Mode promotions (Search Engine Land). Merchant Center loyalty data can show member pricing to signed in members.
6. Brand trust. Some brands appear again and again in AI recommendations even though the exact list changes run to run; SparkToro’s research found certain brands in 60% to 90% of responses for a given intent (Search Engine Land).
7. Ability to transact. If the agent can complete checkout or booking with you inside the surface, there’s less friction in choosing you. Eligibility for programs such as UCP checkout matters here.
8. The shopper’s own context. Past purchases, linked accounts, and loyalty memberships increasingly shape what an agent suggests. Walmart’s ChatGPT app, with account linking and loyalty, is an early example.

Which Purchases Agents Take Over First?

Not every purchase will be delegated at the same speed. Kalicube calls the line up to which customers hand decisions to agents the delegation boundary (Kalicube). Here is how we expect it to move across categories, based on what has already launched.

Industries likely to feel agentic commerce first: groceries and household replenishment, travel and hospitality, consumer electronics and home goods, and event ticketing. Fashion and beauty follow close behind on discovery, while B2B services, healthcare, and financial products will mostly see AI influence the shortlist rather than the transaction for now.

If you want to see where your own brand sits, our assistive agent optimization guide includes a five question delegation check and a 15 point readiness scorecard.

Three Myths About Agentic Commerce

• Myth 1: Agents will make brands irrelevant. The opposite is happening. When an agent has to pick one option and explain why, the brands it understands and trusts win more often. Brand building doesn’t go away; it gets judged by machines as well as people.
• Myth 2: It’s only for the biggest retailers. Google says UCP already enables direct checkout for hundreds of thousands of brands and retailers (Google), and OpenAI says Shopify merchants’ product data is integrated into ChatGPT through Shopify Catalog with no extra work required from individual merchants (OpenAI). Small brands with clean data can compete with large ones on the comparison step.
• Myth 3: You need to build your own agent first. You don’t. Agents need good data and clear policies far more than they need a brand’s own chatbot. If you want a branded assistant, Google’s Business Agent lets eligible US retailers activate one inside Merchant Center, but it’s an add-on, not the foundation.

What Changes for Marketers?: An Eight Point Checklist

1. Your product data becomes your storefront. Feeds now decide whether you’re even considered. Fill every attribute you can, and add Merchant Center conversational attributes such as product Q&A and related products (Google Merchant Center Help).
2. Promises have to be machine readable. Shipping times, return windows, warranties, and cancellation terms should live as plain text on your pages and in your Merchant Center settings.
3. Trust has to be visible beyond your site. Reviews, press coverage, comparison lists, and creator content are how agents decide you’re a safe choice.
4. Offers move into the agent. Plan promotions that can be expressed as structured offers, such as Direct Offers in AI Mode, promotions shared through ACP, and member pricing through loyalty data.
5. Measurement gets harder, so rebuild it. Expect fewer site visits per sale. Segment AI referrers in analytics, and if you sell in Australia, Canada, India, New Zealand, or the US, use Merchant Center’s AI performance insights to track share of voice in AI Mode and AI Overviews (Google Merchant Center Help).
6. Protect the customer relationship. In UCP checkout and Copilot Checkout you remain the merchant of record, but the experience is mediated. Make post purchase emails, loyalty sign ups, and account creation count.
7. Set an agent policy. Decide which agents can access your site and accounts, and make sure your bot management matches. The Ninth Circuit’s August 2026 ruling narrows anti hacking claims against user directed agents but leaves terms of service claims open (Cooley).
8. Keep a human path. Business Agent on Google, live chat, and fast support catch the edge cases agents can’t handle.

What It Means for Paid Media?

Agentic commerce doesn’t remove advertising. It moves it closer to the moment of decision.

• Google’s Direct Offers pilot puts exclusive discounts in front of shoppers in AI Mode when Google’s systems judge the offer relevant (Google), and Google Ads has begun testing a Direct Offer asset for AI Mode promotions (Search Engine Land).
• Business Agent is now in beta inside YouTube ads in the US, letting viewers ask product questions without leaving the video (Google).
• Adobe’s 2026 holiday forecast expects affiliate and partner revenue share to rise 12% and social media’s share to rise 17%, compared with 4% growth for paid search (Search Engine Watch). Discovery is spreading across more surfaces, and agents are one of them.

The practical shift: budget for offers and creatives that an agent can surface and explain, not just for clicks. A discount with clear terms and an end date is easier for an agent to recommend than a vague “big savings” banner.

Agentic Commerce in the US, India, and the UK

United States

The US is where nearly every program launches first: Google’s agentic checkout, UCP powered checkout in AI Mode and Gemini, Copilot Checkout, Perplexity’s Instant Buy, and hotel booking in AI Mode. Google’s UCP checkout help page currently lists product eligibility in the US, Canada, and Australia, available to select merchants (Google Merchant Center Help). If you sell in the US, agentic commerce is a 2026 holiday issue, not a 2027 one.

Adobe expects AI traffic to US retail sites to rise 130% year over year this holiday season, although it remains a modest share of total traffic (Search Engine Watch). McKinsey estimates agentic commerce could orchestrate up to $1 trillion in US B2C retail revenue by 2030, and $3 trillion to $5 trillion globally (McKinsey). Treat that as a long range forecast, not a near term plan.

India

India has a unique advantage: UPI. The Razorpay, NPCI, and OpenAI pilot uses UPI Circle and UPI Reserve Pay so users can preauthorize an agent to spend within limits they set, with demos covering bigbasket groceries and mobile recharges (Razorpay). But in September 2026, NPCI put the launch of its broader Unified Agentic Protocol on hold while it awaits clearance from the Reserve Bank of India, with a focus on user protection and liability when agents go wrong (Business Standard).

For Indian brands, that means discovery comes first. Merchant Center’s AI performance insights already cover English queries in India, so use them. Get your product data, INR pricing, delivery promises, and reviews in order now, so you can move quickly when payment rails open.

United Kingdom

UK shoppers can use AI Mode and the major assistants, but most agentic checkout programs haven’t reached UK merchants yet, and Merchant Center’s AI performance insights don’t currently list the UK. That makes the UK a discovery and trust market for now: clean feeds with GBP pricing, visible delivery and returns terms, strong reviews, and accurate brand information.

How to Prepare: A 60 Day Plan

Weeks 1 and 2: Find out where you stand

• Ask Google AI Mode, ChatGPT, Perplexity, Gemini, and Copilot your top 20 buyer questions, three times each. Log whether you appear, how you’re described, and which competitors win.
• Pull Merchant Center diagnostics and the AI performance insights report if you’re eligible.
• List every place your prices, stock, and policies live, and check that they match.

Weeks 3 and 4: Fix the data

• Resolve feed errors and price or availability mismatches.
• Add missing identifiers and attributes, then conversational attributes for your top products.
• Publish shipping and returns as plain text and in Merchant Center settings.

Weeks 5 and 6: Build trust signals

• Push review volume and recency on the platforms your buyers use.
• Publish or refresh comparison and proof content that agents can quote.
• Correct inaccurate third party listings and old pages that agents keep repeating.

Weeks 7 and 8: Test and integrate

• Run agent task tests: try to buy or book your own products through each assistant and record every failure.
• If you’re a US retailer, check UCP checkout requirements and register interest; confirm your Shopify or payment partner integrations for ChatGPT and Copilot.
• Write down your agent access policy and align your bot settings with it.

How YuvGro Helps?

We help brands get chosen in agent led shopping across the US, India, the UK, and our other markets. That includes product data and feed audits, agent visibility tracking, trust and review programs, and the strategy work to decide where agentic commerce fits in your growth plan. Explore our assistive agent optimization services, AI optimization services and digital strategy services.

The Bottom Line

Agentic commerce is real, uneven, and moving fast. Agents are already deciding which brands make the shortlist, and in a few categories they’re completing the purchase too. The brands that win won’t necessarily be the biggest. They’ll be the ones agents can understand, trust, and buy from without friction.

Fix your product data before the holidays. Plan your checkout integrations for early 2027.

Frequently Asked Questions

What is agentic commerce?

Agentic commerce is shopping in which an AI agent carries out part or all of the purchase journey for a person: discovering options, comparing them against the person’s needs, and sometimes completing the purchase within limits the person sets. In most programs live today, the shopper still confirms the purchase and the retailer remains the seller of record.

How do AI agents decide what to buy?

AI agents match a shopper’s constraints against product attributes, then weigh price, availability, reviews, delivery and return terms, offers, brand trust, and whether they can complete checkout. Fresh, complete product data matters most, because agents compare options in near real time and skip products whose details are missing or inconsistent.

Is agentic commerce actually happening yet?

Yes, in specific lanes. Google AI Mode supports agentic checkout and UCP powered checkout for eligible US retailers and hotel booking for US users, and Microsoft Copilot has a checkout. But OpenAI deprioritized Instant Checkout in March 2026, and Forrester found only 8% of US online adults had used it by October 2025. Most agent influence today is on discovery and shortlisting.

Which industries will AI agents buy from first?

Repeat, low risk purchases such as groceries and household items, travel and lodging, spec driven goods such as electronics and home products, and time released purchases such as event tickets are leading. Taste driven categories like fashion follow on discovery, while high consideration and regulated purchases will mostly see AI shape the shortlist rather than complete the sale for now.

What is the difference between UCP, ACP, and AP2?

The Universal Commerce Protocol (UCP), launched by Google with retailers including Shopify, Walmart, and Target, standardizes how agents discover, buy, and handle support with merchants. The Agentic Commerce Protocol (ACP), from OpenAI and Stripe, now centers on sharing product feeds and promotions with ChatGPT. The Agent Payments Protocol (AP2), from Google, focuses on how agents make payments securely on a person’s behalf.

Do brands keep the customer relationship in agentic commerce?

Partly. In Google’s UCP powered checkout and Microsoft’s Copilot Checkout, the retailer remains the merchant of record and handles fulfillment and support. However, the shopping experience happens inside the AI surface, so brands need to work harder on post purchase communication, loyalty programs, and account linking to keep the relationship.

Is agentic commerce available in India?

It’s in pilot. Razorpay, NPCI, and OpenAI began a private beta of UPI based agentic payments in ChatGPT in October 2025, using UPI Circle and UPI Reserve Pay so users can set spending limits. NPCI’s broader agentic payments protocol was put on hold in September 2026 pending regulatory clearance, so Indian brands should focus on discovery and product data first.

How should brands prepare for agentic commerce?

Start with product data: complete, accurate feeds with conversational attributes, plus visible shipping and return terms. Build trust signals agents can read, such as reviews and third party coverage. Then measure your visibility in AI answers, test whether agents can buy from you, set a deliberate agent access policy, and evaluate checkout programs where you’re eligible.