What Is Agentic Commerce? Definition, Examples, and How It Works

TL;DR

• Agentic commerce is AI agents finding, comparing and, with the shopper's permission, buying products on their behalf inside assistants such as ChatGPT, Google AI Mode, Gemini and Perplexity.
• Agents choose from structured product data (price, availability, shipping and returns), so a listing with incomplete or ambiguous data gets dropped before checkout.
• Commerce protocols (ACP, UCP, AP2, MCP and x402) handle discovery and checkout, and card networks issue agent-specific payment tokens so an agent can pay within limits the shopper sets.
• Promptwatch's ChatGPT Shopping tracking shows which of your products AI assistants surface in shopping answers, and its AI crawler logs show whether shopping agents can reach your product pages.

Agentic commerce is a model of online buying in which AI agents act on behalf of users to discover products, compare options and, with permission, complete purchases and handle follow-up tasks like tracking and returns. The shopper delegates the task to an assistant inside ChatGPT, Google AI Mode, Gemini, Perplexity or a dedicated shopping agent, and the agent does the browsing.

The word "agentic" describes AI that works toward a goal across several steps and takes actions on its own, within limits a person sets. For how that compares with a single-task assistant, see the difference between AI agent and agentic AI.

An agent skips hero images and headlines and queries merchant catalogs through structured data feeds, evaluating price, availability, shipping speed, return policy, reviews and relevance to the stated intent. Agents are built to be risk-averse, so merchants with complete, machine-readable catalogs get recommended, and a listing that leaves shipping cost or return policy implicit can lose the sale before a person ever sees it.

How does agentic commerce work?

Agentic commerce runs through a repeatable sequence, whether the agent is booking a flight or reordering printer ink.

1. Intent capture. The user states a goal and its constraints in a chat interface: "find a waterproof hiking jacket under $150 with free returns." The agent parses budget, product category and non-negotiables like the return policy.
2. Discovery and evaluation. The agent queries merchant catalogs through structured product feeds and commerce protocols, and compares options against the stated intent: price, stock status, shipping speed, reviews and return terms.
3. Transaction. Once the agent selects an option, usually after showing the user a shortlist to approve, the purchase happens through one of two routes. Some assistants hand checkout to the merchant: since March 2026, ChatGPT focuses on product search and discovery and sends the purchase to the merchant's connected app or site. Others complete checkout inside the answer, as Google does in AI Mode and Gemini through the Universal Commerce Protocol, with payment handled through Google Pay. For how the main protocols split this work, see AP2 vs ACP vs MCP vs UCP.
4. Post-purchase management. The agent, or a connected assistant, can track the order and start a return or reorder without the person going back to the original site.

What is agentic checkout?

Agentic checkout is the payment step of agentic commerce: the AI agent completes the purchase for the shopper, either inside the AI interface or through the merchant's connected app, after the shopper approves it or within spending limits set in advance. Google runs agentic checkout inside AI Mode and Gemini through UCP and Google Pay, while ChatGPT hands checkout to the merchant's app or site. Payment usually runs on a tokenized credential that a card network such as Visa or Mastercard binds to that specific agent and that the shopper can revoke.

Agentic commerce vs traditional ecommerce

The merchant infrastructure stays the same in both models: catalogs, payment processors, inventory and fulfillment. What moves is who performs each step of the purchase.

The last row changes budgets: a product page built to convert a human visitor can still fail an agent if shipping cost, stock level or return terms are missing from its structured data.

Real-world examples of agentic commerce

• A user asks ChatGPT for a waterproof hiking jacket under $150 with free returns. ChatGPT compares products from merchant feeds and shopping indexes, shows a ChatGPT shopping carousel with its reasoning, and hands the purchase to the merchant's connected app or site.

• A shopper in Google AI Mode buys a product without leaving the answer, through UCP checkout from a merchant whose Google Merchant Center listing is eligible for it.
• A shopper asks Amazon's Alexa for Shopping to buy a product, and the assistant completes the purchase on Amazon or another retailer's site on the shopper's behalf.
• A traveler whose flight is cancelled lets an assistant with booking access rebook the next nonstop option within a fare limit set in advance.
• A B2B procurement agent reorders supplies from an approved vendor list when stock falls below a threshold, within a monthly budget.
• A GEO team runs real buyer prompts through AI shopping assistants and audits whether its products are selected, mispriced or excluded because of incomplete catalog data.

The last example is the one most brands skip: running the prompts a real buyer would use to see whether your products get selected, and why a competitor's got picked instead.

Which companies and protocols power agentic commerce?

Three groups build the rails: protocol owners, AI platforms where agents shop, and payment networks that let an agent pay safely. Status as of October 2026:

Protocols: ACP, UCP, AP2, MCP and x402

ACP and UCP cover the shopping flow, from product feed to checkout. AP2 proves an agent was authorized to pay. MCP now works mostly as a transport: Shopify serves its agent catalog, cart and checkout tools through a UCP endpoint over MCP, and removed the old Storefront MCP catalog and cart tools by August 31, 2026.

x402 is the newest name in the stack. Coinbase's open protocol is named after the HTTP 402 "Payment Required" status code: a server answers an agent's request with a payment request, the agent pays in a stablecoin such as USDC, and access is granted in the same exchange. Stripe began using it for USDC payments by AI agents on Base in February 2026.

AI platforms where agents shop

ChatGPT focuses on product discovery and sends the purchase to the merchant. Google AI Mode and Gemini complete checkout inside the answer through UCP, and in May 2026 Google announced Universal Cart, a cart that works across retailers in Search and Gemini, with Nike, Sephora, Target, Ulta Beauty, Walmart and Wayfair among the first named retailers. Perplexity offers Instant Buy checkout through PayPal, and Amazon's Alexa for Shopping, launched in May 2026, can buy from Amazon and other retailers on a shopper's behalf.

Payment networks: Visa, Mastercard, Stripe and PayPal

Card networks answer the question every merchant asks first: is this agent allowed to spend this shopper's money? Visa Intelligent Commerce binds a tokenized version of the shopper's card to a specific AI agent, after the shopper authenticates and with consent they can revoke. Visa's Trusted Agent Protocol, co-developed with Cloudflare on Web Bot Auth, lets merchants recognize a legitimate shopping agent through cryptographic signatures on its requests. Mastercard Agent Pay uses agent-specific tokens too, starting with Citi and U.S. Bank cardholders. Stripe co-built ACP and processes agent payments by card and stablecoin, and PayPal handles checkout for Perplexity's Instant Buy.

No single AI platform is best for every brand. The right one to prioritize depends on where your buyers already ask shopping questions and how complete your product data is on each surface.

Where brands lose the sale: the evaluation step before checkout

Most coverage of agentic commerce focuses on checkout and payments. The decision happens earlier, at discovery and evaluation, and it runs on whatever structured data and content the agent can parse. Page design, copywriting and paid search position carry little weight at that step, while product pages carry the most. In Promptwatch's ChatGPT citation data for September 28 to October 4, 2026, product pages were the most-cited content type at 22.4% of citations, rising to 26.0% for commercial prompts, ahead of landing pages and listicles.

Those same pages are the hardest for machines to read. Adobe's April 2026 benchmark of US retail sites scored product pages at 66% machine readability on average, the lowest of any page type it measured, against 82% for returns pages and 80% for FAQ pages.

Each shopping prompt turns into several checks. "Waterproof hiking jacket under $150 with free returns" becomes sub-queries on price, stock in the shopper's size, delivery time and return terms. A product that answers three of them and leaves returns unclear drops off the shortlist. The merchant sees no bounce, no session and no abandoned cart, because the shopper never reached the site.

That's why a citation and a sale are separate bars. A brand can be cited in an AI Overview about its category and still miss the product cards, because a card depends on a complete product record that an agent can act on with confidence.

What are the risks of agentic commerce?

Four risks come up most often, for shoppers and merchants alike, plus an open question about whether adoption keeps its pace.

Fraud and agent impersonation

Payment and fraud systems were designed to verify a human at checkout. Merchants now have to tell a legitimate shopping agent from a scraping bot or a fraudster running one. Signed agent requests (Web Bot Auth, Visa's Trusted Agent Protocol) and agent-specific card tokens limit what an impostor or a stolen credential can do.

Wrong purchases and liability

An agent can buy the wrong size, the wrong quantity or a non-refundable fare, and the split of responsibility between AI platform, merchant and card network is still being worked out. The precedent businesses watch is from February 2024, when a Canadian tribunal ordered Air Canada to honor a refund policy its customer-service chatbot had invented. Spending limits, approval steps for high-value orders and a return policy stated in structured data all reduce the exposure.

Prompt injection and manipulated listings

Agents read untrusted content: product pages, reviews and marketplace listings. Hidden instructions in that content can try to steer the agent toward one product or away from a competitor. Merchants should monitor what AI assistants say about their products so a manipulated comparison shows up quickly.

Privacy and consumer trust

An agent that buys needs payment credentials, a delivery address and often purchase history. Many shoppers hesitate to hand that over. Network models address it with authentication before an agent gets access and consent the shopper can revoke at any time.

Why agentic commerce could stall

Most consumers still browse and check out manually. The protocols change fast: OpenAI changed ChatGPT's checkout model within six months of launching it. Few merchants can staff an integration for every surface. The discovery side is already live, though: assistants pick products for shoppers today, even when the final click happens on the merchant's site.

Is agentic commerce the future of ecommerce?

Agentic commerce is a new transactional layer on top of existing ecommerce infrastructure. The product catalogs, payment processors, inventory systems and fulfillment operations that power a webstore today are the same systems an agent queries and buys through. What changes is who, or what, does the browsing and clicking.

Real adoption sits behind the term. Grand View Research estimates the agentic commerce market at $7.7 billion in 2026, growing at a 35.7% CAGR to $65.5 billion by 2033. McKinsey's October 2025 analysis estimates AI agents could orchestrate $900 billion to $1 trillion of US B2C retail revenue by 2030, and $3 trillion to $5 trillion globally. The two figures measure different things: Grand View sizes the agentic commerce market itself, while McKinsey estimates the retail spend that could flow through agents.

Shoppers already arrive through AI. Adobe measured AI traffic to US retail sites up 393% year over year in the first quarter of 2026, and in March 2026 that traffic converted 42% better than non-AI traffic, after converting 38% worse a year earlier.

What changes for each group:

• Retailers: product data, inventory and policies have to be readable by agents, and checkout has to accept agent-initiated payments.
• Brands: being picked by an AI assistant starts to matter as much as ranking in search, and it's measured in shopping answers.
• Payment providers: they have to verify that an agent is authorized to pay, which is the job of agent tokens and agent authentication.
• Shoppers: buying gets faster, in exchange for giving an agent payment authority within set limits.

Agent-mediated purchases are still a small share of online retail, and most merchant catalogs aren't fully agent-ready. Treating agentic commerce as an emerging channel to prepare for is the more defensible position for teams planning 2026 and 2027 budgets.

How to prepare your product catalog for agentic commerce

For the full sequence, with an owner and a finish line for each step, see how to prepare for agentic commerce. These are the five moves that matter most for the catalog itself:

• Make product data machine-readable and complete. Price, availability, shipping speed and return policy all need to be present in structured formats like JSON-LD, stated explicitly in each listing instead of buried in a shipping FAQ. Keep these details in plain HTML text, since information that only appears inside an image, an animation or an interactive widget is easy for an AI crawler to miss. The guide on how to optimize product listings to be selected by AI shopping agents covers the fields one by one. Promptwatch's Site Health crawl flags thin content and crawl-blocking risks on a site, the same gaps that can disqualify a listing from an agent's shortlist.
• Check that shopping agents can reach your product pages. A perfectly structured feed is worthless if a 403 or a crawl-rate limit blocks the agent first. The free agent readiness checker tests your robots.txt against nine AI crawlers and checks agent discovery standards in about 20 seconds, with no signup. Promptwatch's AI crawler logs show crawler identity, page requested and status code per hit, so a blocked shopping bot shows up as a specific, fixable event.
• Track whether your products actually surface in shopping answers. Promptwatch's ChatGPT monitoring for product pages shows brand-level share of voice in shopping responses and which of your product pages get recommended or skipped.
• Refresh product content on a cycle. Promptwatch's citation data shows relevance holds for roughly 8 to 14 weeks before a refresh helps, as models re-crawl and re-index regularly, and a stale price or shipping policy can quietly cost a listing its spot. Content Agent drafts and update suggestions are grounded in a brand's own citation and content-gap data, so refreshes target what has gone stale.
• Test real buyer prompts against your catalog the way a shopping agent would. Write prompts for the specific buyers you want to win: the same white sneaker can be the answer for a commuter, a runner or a shopper who wants sustainable materials, and each prompt rewards different product details. Prompt Tracking runs a brand's buyer-intent prompts across tracked AI platforms on a repeating schedule, so gaps in product data surface before a lost sale does.

How Promptwatch helps brands measure agentic commerce readiness

Promptwatch treats citations and agent selection as one visibility problem: what content and data a brand exposes to AI systems, and whether it's enough for those systems to choose that brand.

Content Coverage and Content Gap Analysis apply the same fan-out logic that shapes agentic search to a transaction. Promptwatch breaks each tracked shopping prompt into the sub-queries an agent checks (price, then availability, then return policy) and re-runs them against a brand's own indexed content to flag which ones go unanswered.

On the shopping side, Promptwatch extracts product results from ChatGPT search and Amazon Alexa responses and matches them to your catalog. Google AI Overviews and AI Mode shopping results are visible inside tracked responses but aren't matched in the Shopping section yet. For a side-by-side of the tools that track product cards, see the roundup of the best tools to track AI visibility for ecommerce.

Frequently asked questions

Is agentic commerce the same thing as agentic AI?

No. Agentic AI is the broader term for AI systems that plan and adapt across multiple steps toward a goal. Agentic commerce is that behavior applied to shopping: an agent researching, comparing and completing a purchase.

Is agentic commerce just hype?

Real infrastructure sits behind the term: published protocols, live checkout in Google AI Mode and Gemini, agent payment programs from Visa and Mastercard, and a market Grand View Research sizes at $7.7 billion in 2026. The standards are still moving and most consumers still check out manually. Agentic commerce adds a channel on top of the same catalogs, payment rails and fulfillment systems that run ecommerce today.

How does an AI agent decide which product to recommend?

It evaluates structured product data against the user's stated intent: price, availability, shipping speed, return policy and reviews, pulled from merchant feeds and shopping indexes. Incomplete or ambiguous data gets a product skipped.

Is it safe to let an AI agent buy for you?

It's safest when the agent pays with a tokenized credential tied to that agent, when you set a spending limit, and when you approve purchases above that limit. Visa's and Mastercard's agent programs both use tokens of this kind, and Visa's model lets the shopper revoke an agent's access at any time. Check the return policy before approving, since an agent can still pick the wrong size or quantity.

Do merchants need ACP or UCP to be found by AI shopping agents?

Discovery mostly runs on product feeds, shopping indexes and crawlable product pages. A study published by Search Engine Land on March 5, 2026 found that 83% of ChatGPT carousel products strongly matched a product in Google Shopping's top 40 organic results. A protocol integration matters for checkout inside the AI answer: UCP for Google AI Mode and Gemini, and ACP or Shopify Catalog for a direct product feed to ChatGPT.

Which AI platform is best for agentic commerce?

It depends on where your buyers shop. ChatGPT focuses on product discovery, Google AI Mode and Gemini offer in-answer checkout through UCP, Perplexity offers Instant Buy through PayPal, and Amazon's Alexa for Shopping can buy on a shopper's behalf. Prioritize the surface your buyers already use and check that your product data is complete there.