# Article Name
What Is Agentic Commerce? (2026)

# Article Summary
Agentic commerce is online buying where an autonomous AI agent handles discovery, evaluation, and the purchase itself within guardrails you set. This guide explains how an agentic purchase works, the protocols competing to standardize it (ACP, UCP, AP2), how payments and trust work, why OpenAI's first checkout attempt stumbled, and how developers can monetize the discovery layer today with affiliate links.

# Original URL
https://www.getchatads.com/blog/what-is-agentic-commerce/

# Details

Online shopping has always asked you to do the clicking, the comparing, and the final buying yourself. That long-standing arrangement is changing fast now that AI agents can do the work. Agentic commerce is the name for a model where an AI agent runs that whole job on your behalf.

The term shows up everywhere in 2026, often stretched to cover things that are not really autonomous. Most shopping experiences people call agentic today still keep a human in the loop for the actual purchase. Knowing the difference matters before you build anything or spend a budget on it.

This guide breaks down what agentic commerce really means, how a purchase moves through it, which protocols are fighting to standardize it, and what all of it means for developers who want a cut. The short version is that the discovery layer already works, while the money layer is still being argued over.

Why agentic commerce matters in 2026:
- McKinsey and ICSC project $1 trillion in US B2C agentic retail by 2030
- Adobe measured an 805% year-over-year jump in AI traffic to US retail on Black Friday 2025
- Gartner expects 20% of digital commerce transactions to run through AI platforms by 2030
- Only about 14% of consumers currently trust an AI to place an order autonomously

## What Is Agentic Commerce?

Agentic commerce is online buying where an autonomous AI agent, not a person clicking through a site, handles discovery, evaluation, and the actual purchase. You set the goal and the limits, and the agent does the rest within them. The clearest test is who closes the transaction at the end.

It helps to line up three models side by side. Traditional e-commerce puts you in front of a website where you navigate and decide everything yourself. Conversational commerce adds a chat layer, so an assistant answers questions, but you still make the call and click buy. Agentic commerce removes that last manual step and lets the agent finish the job for you.

The reality in 2026 is messier than the marketing suggests. Forrester points out that most experiences sold as agentic are still really conversational, with a human approving every purchase.

Three Models of Online Buying:
- Traditional e-commerce: you decide, you close the sale on the site
- Conversational commerce: you decide with help, you close the sale after chatting
- Agentic commerce: the agent decides, the agent closes the sale within limits

## How Does an Agentic Purchase Actually Work?

An agentic purchase moves through a loop that starts with intent and ends with a delivered order. You give an instruction like restock my cleaning supplies, and the agent captures that intent as its goal. From there it searches machine-readable catalogs, builds a cart against your constraints, and gets ready to pay.

The human only touches two points in that entire loop. You authorize the agent up front, and you set the guardrails like spending caps and category limits. Everything between those bookends runs without you, which is a real shift in how buying works.

One detail underneath all of it keeps the system safe. The agent never holds your real card number, and it passes a scoped token instead that works only for that one purchase. That single design choice separates a useful agent from a standing security risk.

The five stages: intent capture, discovery, cart construction, scoped payment delegation, settlement.

## What Is the Agentic Commerce Protocol (ACP)?

The Agentic Commerce Protocol, or ACP, is the open standard OpenAI and Stripe released in September 2025 (https://www.agenticcommerce.dev/). It exists to connect three parties that previously had no shared language: the buyer, the buyer's agent, and the merchant. Stripe and OpenAI published it under an Apache 2.0 license so anyone can build on it.

ACP is built from a few clear pieces that handle each step of a sale. A product Feed lets the merchant expose its catalog, a Checkout API runs the order, and Delegate Payment passes a Shared Payment Token so money can move. Delegate Authentication handles the consent piece through OAuth so the agent acts with real permission.

The standard kept evolving in the months right after its launch. An April 2026 update folded in the Model Context Protocol, which turns ACP endpoints into tools an agent can call directly. That change matters because it lets any MCP-aware agent transact through ACP without custom glue code.

## ACP vs. UCP vs. AP2: How Do the Protocols Differ?

The three names that get confused most often are ACP, UCP, and AP2. ACP comes from OpenAI and Stripe, while UCP is the competing standard Google and Shopify launched in January 2026. Both try to do the same broad job, which is moving a shopper from discovery all the way through checkout.

AP2 sits in a different category than the other two, even though it gets lumped in with them. Google introduced it in September 2025 as a payment-authorization layer, not a full commerce protocol. It uses three signed mandates, for intent, cart, and payment, that act as tamper-evident proof of what the user approved.

The cleanest way to hold these apart is by the layer each one owns. ACP and UCP are the shopping and checkout layer, while AP2 is the proof-the-user-authorized-this layer underneath them. A merchant can run AP2 beneath either commerce protocol rather than choosing among all three.

## How Do Payments Work When an Agent Checks Out?

Agentic checkout works because of tokenization, the same idea that protects your card in a phone wallet. Instead of handing an agent your real card, the rails issue a scoped token that stands in for it. A leaked token is close to worthless, since it only covers one merchant and one cart total before expiring.

Several networks shipped their own version of this within months of each other. Stripe issues Shared Payment Tokens, Mastercard built Agentic Tokens under Agent Pay scoped to rules like groceries only, and Visa launched Intelligent Commerce. The first live Mastercard agentic transaction ran on September 29, 2025, which showed the rails working in the real world.

Payments also have to fit merchants that do not want a heavy rebuild. PayPal's Agent Ready lets existing merchants accept agent payments with no extra integration work. The store stays the merchant of record through all of it, so refunds and support run exactly where they always did.

What a safe agent payment needs: a scoped mandate naming the merchant, amount, and consent; a single-use token standing in for the real card; spending limits the agent cannot exceed; the merchant staying merchant of record.

## How Does a Merchant Know an Agent Is Trustworthy?

Trust is the hardest problem in agentic commerce, and it breaks the signals fraud systems depend on. Agents produce none of the human signals that fraud systems lean on, like device fingerprints, typing cadence, or mouse movement. They can also fire from data-center addresses at machine speed, which looks a lot like an attack.

Two camps answer the trust question from two opposite directions. Visa's Trusted Agent Protocol asks whether the agent itself is a legitimate registered party, using signed HTTP requests that expire in about eight minutes. Mastercard's Verifiable Intent asks a different question, checking whether the cart actually matches what the human meant to buy.

Both approaches exist because an agent can fail in two separate ways. It might be a fake agent impersonating a real one, or a real agent that drifted and bought the wrong thing after a prompt injection. Catching both takes identity checks and intent checks working together rather than either one alone.

## What Does It Take to Make a Store Agent-Ready?

Becoming agent-ready is an infrastructure project, not a marketing campaign you can launch in a week. The foundation is a structured product feed, usually a gzipped JSONL, CSV, or XML file with clean titles, attributes, and prices. Agents read that feed instead of your pretty website, so messy data simply makes your products invisible.

Beyond the feed, a store needs a Checkout API that implements the spec end to end. It also needs schema.org markup so agents can parse catalog data that never renders in plain HTML. Most larger merchants will end up supporting both ACP and UCP to reach ChatGPT traffic and Google traffic at once.

One step gets skipped constantly and costs merchants real insight later. You have to tag agent-initiated traffic separately from human traffic from the very start. Nobody can optimize a channel they cannot even see in their own analytics, so that tagging is not optional.

## How Big Is the Agentic Commerce Market, Really?

Market projections for agentic commerce are huge and wildly inconsistent with each other. McKinsey and ICSC put US B2C agentic retail at $1 trillion by 2030, while Morgan Stanley pegs the US figure at $190 to $385 billion. The gap is that wide because nobody agrees on what should count as agentic in the first place.

The optimistic numbers deserve a skeptical read against the adoption reality. Gartner expects 20% of digital commerce transactions to run through AI platforms by 2030, yet also predicts 40% of agentic projects get canceled by 2027. Only about 14% of consumers currently trust an AI to place an order on its own.

The demand signals are genuine even if the forecasts swing around a lot. Adobe measured an 805% year-over-year jump in AI traffic to US retail sites on Black Friday 2025. Growth that steep means the behavior is forming now, well ahead of the trust that has to catch up to it.

## Why Did OpenAI's First Checkout Attempt Stumble?

OpenAI's Instant Checkout is the clearest cautionary tale in agentic commerce so far. It launched inside ChatGPT in late 2025, starting with Etsy and then Shopify merchants, and charged a flat 4% fee. That cut sat on top of the Shopify and Stripe fees merchants already paid, which stung from day one.

The rollout ran into trouble much faster than almost anyone expected. Fewer than a dozen Shopify merchants fully integrated, in-chat conversion came in weak, and OpenAI had not built sales-tax collection. Native checkout got pulled in March 2026, only months after the launch headlines celebrated it.

OpenAI did not abandon shopping. It just changed which part of it the company owns. It pivoted to a discovery model where ChatGPT surfaces products and retailers like Walmart and Target close the sale on their own systems. The lesson is blunt: owning discovery is far easier than owning checkout.

## What Does Agentic Commerce Mean for Affiliate Revenue?

Agentic commerce quietly breaks the machinery that affiliate revenue has run on for years. The $13 billion US affiliate industry is built on cookies and last-click attribution, and agents carry no cookies at all. They fire no JavaScript and load no thank-you page, so the tracking that pays publishers never triggers.

The twist is where agents actually get their product knowledge in the first place. Affiliate content is feeding these systems, since one analysis found roughly 70% of ChatGPT's cited sources for a single brand came from affiliate articles. Those publishers shape the recommendation yet often earn nothing on the purchase that follows.

The revenue model is still being worked out, but the early signals lean one way. Sam Altman has said on record that he prefers a roughly 2% affiliate fee over running display ads. Agent-mediated attribution could even beat cookies in the long run, since an API-logged purchase is deterministic rather than guessed.

## How Can Developers Monetize Agents Without Owning Checkout?

Most developers building a chatbot or agent will never run their own payment rails, and they do not need to. The affiliate model lets you earn without owning feeds, protocols, or tokens, because it works precisely when there is no cookie to rely on. You read the conversation, spot real product intent, and return a tracked link only when one fits.

A few monetization paths exist for AI apps that skip checkout entirely. Ad SDKs built for AI surfaces, like Koah, report around $10 eCPM and 7.5% click-through, while subscription tiers convert only about 5% of users. Affiliate insertion tends to land higher on revenue per message, with benchmarks running from $0.08 to $0.30 and up in strong categories.

This is the gap ChatAds was built to fill for developers who want revenue without the rebuild. It reads the assistant's reply, finds genuine buying intent, and returns an affiliate link through one API call. You keep 100% of the commission and never touch checkout, feeds, or tokens to do it.

Agentic commerce in 2026 is real but earlier than the biggest headlines imply. The discovery layer already works well, while autonomous checkout keeps tripping over payments, trust, and merchant plumbing. The protocols are still competing and the market forecasts still disagree. Consumer trust has a long way to go before it catches up to the technology.

For developers, the takeaway is that you can join this shift now without rebuilding the entire stack. Detect intent, disclose honestly, and return a useful link only when it genuinely helps the person reading. The teams that win early monetize discovery now instead of waiting for autonomous checkout to settle.

## FAQ

Q: What is agentic commerce?
A: Agentic commerce is online buying where an autonomous AI agent handles discovery, evaluation, and the actual purchase on your behalf. You set the goal and the guardrails, like spending caps and category limits, and the agent completes the transaction within them. The defining test is who closes the sale: in agentic commerce, the agent does.

Q: What is the Agentic Commerce Protocol (ACP)?
A: ACP is the open standard OpenAI and Stripe released in September 2025 to connect buyers, AI agents, and merchants. It defines a product Feed, a Checkout API, Delegate Payment via Shared Payment Tokens, and OAuth-based Delegate Authentication, all under an Apache 2.0 license.

Q: What is the difference between ACP, UCP, and AP2?
A: ACP (OpenAI and Stripe) and UCP (Google and Shopify) are competing commerce protocols that cover discovery through checkout. AP2 is Google's payment-authorization layer that uses signed mandates to prove what the user approved. A merchant can run AP2 underneath either commerce protocol rather than choosing among all three.

Q: How do AI agents pay for purchases?
A: Through tokenization: the payment rails issue a scoped, single-use token that stands in for your real card, so the agent never holds your card number. Stripe's Shared Payment Tokens, Mastercard's Agent Pay, and Visa's Intelligent Commerce all work this way, with spending limits the agent cannot exceed.

Q: How big is the agentic commerce market?
A: Estimates diverge widely because nobody agrees on what counts as agentic. McKinsey and ICSC project $1 trillion in US B2C agentic retail by 2030, while Morgan Stanley pegs it at $190 to $385 billion. Adoption is earlier than the forecasts suggest, with only about 14% of consumers trusting an AI to place an order autonomously.

Q: How can developers make money from agentic commerce?
A: The most accessible path is affiliate monetization of the discovery layer, which works today while autonomous checkout is still settling. An API like ChatAds reads your assistant's replies, detects genuine buying intent, and returns tracked affiliate links without you owning feeds, protocols, or payment rails. You keep the commission and never touch checkout.