What is agentic commerce? A definitive guide for merchants

Short answer: Agentic commerce is online shopping where an AI agent, not a human clicking through a website, handles discovery, comparison, and sometimes checkout on a shopper's behalf. A person states what they want ("a waterproof hiking jacket under $150 that ships by Friday"), and the agent searches across merchants, evaluates the options, and recommends or buys. For merchants, it turns the AI assistant into a new sales channel and shifts the contest from "who ranks on the search results page" to "whose product data an agent can read, trust, and recommend."

This guide explains what agentic commerce is, how it works, which protocols and platforms now power it, where the hype has already run into reality, and what merchants should actually do to prepare. It reflects the state of the market as of June 2026, including the course corrections that happened early in the year.

Key takeaways

• Agentic commerce lets AI agents discover, compare, and buy products on a shopper's behalf, often without the shopper ever visiting a store.
• The category became real in 2025 when OpenAI and Stripe shipped the Agentic Commerce Protocol, followed by Google, Shopify, Microsoft, Visa, Mastercard, and PayPal.
• It is not a straight line up. OpenAI scaled back its in-chat Instant Checkout in March 2026 after weak adoption and conversion, and pivoted to retailer-operated apps inside ChatGPT.
• Agents do not browse the way humans do. They read structured product data, so catalog quality, accuracy, and completeness now decide whether your products get surfaced.
• It is still early. AI-driven shopping is growing very fast but remains a small share of total ecommerce traffic today.
• The clearest lesson from 2026 so far is that accurate, machine-readable product data is the deciding factor. Stale data breaks the experience.
  

What is agentic commerce? 

Agentic commerce is a model of online shopping in which an autonomous AI agent acts on a buyer's intent. Instead of the shopper running the search, opening tabs, comparing products, and filling in a checkout form, the agent does that work and presents a curated result, or completes the transaction directly.

The defining word is "agentic." An agent is software that can take goal-directed actions on a user's behalf, not just answer a question. In a commerce context, that means an assistant like ChatGPT, Gemini, or Microsoft Copilot can move from "here are some options" toward "I found it and I bought it for you" within the same conversation.

Three things make this different from a chatbot that recommends products: 

1. Intent comes first. The buyer's agent arrives with context, and sometimes payment authority, before a seller's website is ever involved.
2. The agent can transact. With the buyer's approval, the agent can assemble the cart and complete checkout, not just hand off a link.
3. The interface is a conversation. As Stripe and OpenAI put it when they launched the Agentic Commerce Protocol, "conversational surfaces are becoming a new kind of storefront".

How is agentic commerce different from traditional ecommerce? 

 Traditional ecommerce assumes the buyer comes to a seller-controlled environment, a website or marketplace, and the seller uses that environment to capture intent and convert it. Agentic commerce inverts that assumption. 

This is often called zero-click discovery. The shopper gets a recommendation and sometimes buys without ever loading a product page. That changes what merchants compete on. You cannot improve a button color the agent never sees. What the agent does see is your structured product data.

How does agentic commerce work?

A typical agentic purchase moves through four stages.

1. Discovery. The shopper expresses intent in natural language. The agent queries product catalogs from merchants that have made their data available in a structured, machine-readable format. Agents do not crawl and interpret a web page the way a person reads it. They rely on feeds, APIs, and structured markup.

2. Evaluation. The agent compares retrieved products against the shopper's stated criteria, things like price, material, dimensions, availability, shipping time, and reviews. The richer and more complete your attributes, the more accurately the agent can match your product to a query. Guidance from platforms and commerce vendors consistently points to the same lever: detailed, complete, accurate product data is what gets a product selected.

3. Checkout. With the buyer's approval, the agent completes the purchase. This is where the new payment protocols come in. Rather than the agent handling a raw card number, the networks issue scoped, tokenized credentials tied to a specific agent, a specific merchant, and a specific consent from the buyer.

4. Fulfillment. The order flows into the merchant's existing systems. A core promise of the main protocols is that merchants keep their existing commerce and payment infrastructure, their fulfillment, and their customer relationship, rather than handing them to the AI platform.

The two technical pieces that make this possible are a discovery layer (structured product data the agent can read) and a payment layer (a verifiable, scoped way for an agent to pay). The next sections cover the standards being built for both.

What is the Agentic Commerce Protocol (ACP)?

 The Agentic Commerce Protocol (ACP) is an open standard, co-developed by OpenAI and Stripe, that defines how AI agents interact with businesses to complete purchases on a buyer's behalf. OpenAI and Stripe announced it on September 29, 2025, alongside Instant Checkout in ChatGPT, and released the specification publicly so any merchant or developer can implement it.

ACP standardizes product discovery, checkout negotiation, and payment so that, in principle, any compliant agent can transact with any compliant merchant without a custom one-off integration. OpenAI and Stripe describe themselves as the founding maintainers, with a stated path toward broader community governance, and the protocol is open, so businesses that do not process payments with Stripe can still adopt it with their existing providers. The specification has continued to expand: the public spec history shows an April 2026 release that added cart, feed, orders, authentication, and Model Context Protocol support, broadening ACP from checkout-only toward discovery through fulfillment.

The ACP standard is alive and growing, and partners now span AI platforms (including Microsoft Copilot, Anthropic, and Perplexity) and commerce platforms (including Shopify, Squarespace, Wix, WooCommerce, and BigCommerce). PayPal adopted ACP to enable payments inside ChatGPT, and Salesforce announced ACP support for Agentforce Commerce merchants in October 2025. 

What happened to Instant Checkout in ChatGPT?

 Instant Checkout was OpenAI's first consumer-facing agentic shopping feature, letting US shoppers buy from participating merchants without leaving the chat. It launched with Etsy sellers and was meant to extend to over a million Shopify merchants.

In March 2026, OpenAI confirmed it was ending Instant Checkout in its original form and moving commerce into retailer-built apps inside ChatGPT instead. A few reasons came together:

• Stale data. OpenAI partly populated product information by scraping retailer websites, so pricing, availability, and shipping estimates were often inaccurate at the moment of purchase, which eroded trust.
• Weak adoption. Fewer than thirty Shopify merchants went live before the change.
• Weak conversion. Walmart, which had tested Instant Checkout from late 2025, reported that in-chat purchases converted at roughly one-third the rate of click-throughs to Walmart.com, and it pulled out in favor of its own assistant, Sparky, embedded inside ChatGPT and Gemini.

What is the difference between ACP, UCP, AP2, and the card-network protocols?

 This is the most common point of confusion, because several standards launched within months of each other across 2025 and 2026, and they operate at different layers. They are not all competitors. Many are designed to work together.

 A useful way to hold this:

• ACP and UCP are merchant-facing checkout standards. UCP was co-developed by Google and Shopify and announced at NRF 2026 in January 2026, with backing from Etsy, Wayfair, Target, Walmart, and more than twenty endorsing partners. With both ACP and UCP, the merchant remains the merchant of record and keeps the customer relationship.
• AP2 is an authorization layer. Google's Agent Payments Protocol proves the buyer actually approved what the agent did, using signed intent, cart, and payment mandates, and it composes with MCP and Google's Agent2Agent protocol.
• Visa's and Mastercard's protocols are network-level implementations. Visa's Trusted Agent Protocol (built with Cloudflare) focuses on verifying that an agent is legitimate, and Mastercard's Agent Pay issues scoped tokenized credentials so a card can be charged without the agent ever holding the raw number.

The industry expectation, stated openly by people building these standards, is that agentic commerce will be multi-protocol for the foreseeable future. For most merchants, the takeaway is that you do not need to pick a winner. You need product data and a checkout that can plug into whichever surface your customers use, and a payment provider that supports the authorization layers underneath. 

What are examples of agentic commerce?

 The clearest live examples as of mid-2026:

• Retailer apps inside ChatGPT. After scaling back Instant Checkout, OpenAI moved to retailer-operated apps, with companies such as Walmart (via its Sparky assistant), Instacart, Target, and Booking.com building dedicated experiences inside the chat.
• Shopping in Google AI Mode and the Gemini app. Powered by UCP, Google enables checkout from eligible US retailers directly inside AI surfaces, using Google Pay and, soon, PayPal.
• Microsoft Copilot checkout. Copilot supports agent-driven purchasing through UCP and the network payment protocols.
• Perplexity. PayPal and Perplexity launched an in-chat "Instant Buy" experience covering thousands of merchants.
• Shopify Agentic Storefronts. Shopify introduced a way for merchants to list once and be syndicated across AI shopping surfaces including ChatGPT, Google AI Mode, Gemini, and Copilot, rather than maintaining a separate feed for each one.

The common thread is that the AI assistant is becoming a place where discovery and, increasingly, the purchase happen, while merchants keep control of their catalog, checkout, and customers. 

Why does agentic commerce matter now 

Three forces are converging:

• The discovery habit is shifting. A growing number of shoppers ask an AI assistant for recommendations instead of running a search and clicking through results. Shopify has reported that AI-driven traffic to its stores grew roughly eightfold year over year since January 2025, with orders from AI-powered searches up around fifteenfold over the same period. Those are one platform's figures, but Visa separately noted a surge of several thousand percent in AI-driven traffic to US retail sites, which points to a real change in behavior even though the base is small.

• The infrastructure is now standardized. Before 2025, agent-driven checkout happened, when it happened at all, through brittle workarounds like stored credentials and scraping, and networks had no reliable way to tell a human transaction from an agent one. The protocols described above exist to fix that, which is what makes the channel safe enough for large merchants and networks to adopt.

• Shoppers say they are open to it. Salesforce research found that 48% of shoppers who already use AI for shopping are open to letting an AI agent make a purchase for them, and Visa has reported that roughly 47% of US shoppers already use AI tools for at least one part of the shopping journey.

A note on the big projections you may have seen: Gartner forecasts that by 2030, 20% of monetary transactions will be programmable in ways that give AI agents economic agency, and that by 2028, 90% of B2B buying will be AI-agent intermediated. Those are broad predictions about programmable and agent-mediated transactions, not a specific claim that one in five retail checkouts will run through a chatbot. Treat them as direction, not a guarantee. 

Is agentic commerce worth investing in yet?

Honest answer: it is early, the volume today is small relative to the noise, and the first big consumer experiment was scaled back.

Independent analysis from Flagship Advisory Partners found that AI-driven traffic accounted for less than 1% of total retail and travel merchant traffic during the 2025 holiday season, and OpenAI's retreat from Instant Checkout in March 2026 showed that frictionless in-chat buying does not automatically convert. So the case for acting now is not "this is already your biggest channel." It plainly is not.

The case for acting now is different and more durable. The lever that decides whether an agent recommends you is the quality, accuracy, and structure of your product data, and that is something that compounds. It takes time to clean a catalog, fill in missing attributes, add accurate structured markup, and keep feeds current. The Walmart and OpenAI episode is the clearest evidence yet that this is the real bottleneck: the experience failed largely because the product data was stale and incomplete. Merchants who fix that while volumes are small are positioned to capture the channel as it matures. You cannot outspend competitors on ads inside an agent's recommendation. Data quality is the primary lever you control.

In short, treat it as a strategic investment in catalog readiness, not as a short-term traffic play. 

How can merchants prepare for agentic commerce 

The preparation work is unglamorous and concrete. It is mostly about your product data.

1. Audit and complete your product data. Agents skip products with thin or inconsistent data. Industry guidance consistently finds that merchants with high fill rates on core attributes, often cited around 95% on the most important fields, see meaningfully better agent visibility, while catalogs that fall well below that get routinely passed over. Make sure titles, descriptions, attributes (material, dimensions, color, compatibility, use case), identifiers like GTINs, pricing, and availability are present and correct across the catalog.

2. Add and maintain structured markup. Agents parse Schema.org Product markup from product pages and read structured feeds. Complete, valid structured data helps an agent understand your product accurately instead of guessing, and guesses rarely favor the merchant.

3. Keep feeds clean and current. For Google's path, getting started with UCP builds on your existing Merchant Center feeds, which means no disapproved products, no policy violations, frequent updates, and item updates enabled. Static feeds refreshed once a day are not enough when agents make real-time decisions and an out-of-stock surprise at checkout kills the sale. This was a central reason the early scraped-data approach failed.

4. Write product descriptions for both people and machines. Descriptions need to be accurate and richly attributed, covering the specifics an agent matches against a query, while still reading naturally for the human who sees the final recommendation. Outcome-focused, detailed copy outperforms thin marketing fluff in this channel.

5. Make your checkout agent-ready. Depending on your platform, this means adopting a checkout standard like ACP or UCP, or relying on a platform integration (Shopify's Agentic Storefronts, for instance) that handles the protocol plumbing for you.

6. Close the measurement gap. When purchases happen inside an AI assistant, traditional web analytics see less of the journey. Build a way to understand which products and which content are actually being surfaced and converted, so you can keep improving the catalog rather than guessing. Measurement is one of the genuine open problems in agentic commerce today, and the merchants who solve it for themselves will optimize faster than those who do not.

A related caution: not every surface is open. Amazon, for example, keeps shoppers inside its own assistant rather than opening its catalog to outside agents, so a multi-channel merchant should expect to maintain product data for several agent ecosystems with different feed structures and rules.

What this means for your product content 

Step back and the pattern is clear. In traditional ecommerce, the page was the product. In agentic commerce, the data is the product. An agent never sees your hero image animation or your clever above-the-fold layout. It sees your title, your attributes, your description, your price, your availability, and your structured markup, and it decides in milliseconds whether you match the buyer's intent.

The events of early 2026 made this concrete rather than theoretical. The most visible early attempt at agentic checkout stumbled because the product data behind it was scraped, stale, and incomplete. The fix the whole industry is converging on is merchant-provided, accurate, structured, continuously updated product data.

That reframes product content from a marketing afterthought into core commercial infrastructure. The catalog is no longer just something humans read after they arrive. It is the interface through which AI agents discover you, evaluate you, and decide whether to recommend and buy. Rich, accurate, machine-readable product content, kept current and measured against real outcomes, is the foundation of being found in this channel.

This is the shift that platforms like WriteText.ai are built around: generating and enriching ecommerce product content at catalog scale, with native integrations for WooCommerce, Magento and Adobe Commerce, and Shopify, so that the data agents depend on is complete, consistent, and optimized. As agentic commerce matures, the line between "content" and "commerce infrastructure" keeps blurring, and the merchants who treat their product data as a strategic asset will be the ones agents surface first.