What is agentic commerce?
The short answer
Agentic commerce is online retail where AI agents help people shop: they research products, compare options across merchants, answer questions, and narrow the choices, and on some platforms can complete checkout through standards like the Agentic Commerce Protocol (ACP) and Google's Agent Payments Protocol (AP2), once the shopper decides.
It differs from traditional e-commerce because the first visitor your store meets is often the shopper's agent, reading on their behalf before the human ever clicks, which changes how discovery, conversion, and payments work.
McKinsey estimates agentic commerce could orchestrate up to $3 trillion to $5 trillion in global retail revenue by 2030.
Verified sources
What is agentic commerce?
Agentic commerce is online retail where an AI agent helps a person shop. The shopper states a goal and the agent does the legwork: it searches across merchants, reads catalogs, compares prices and shipping, answers questions, and narrows the options. Ask ChatGPT for trail-running shoes under $150 that arrive by Friday, and the agent evaluates options across stores and hands back a short list for you to choose from. On some platforms, once you pick, the agent can also complete the checkout you approve. The shopper stays in charge of the decision; the agent does the work around it.
The popular image of agentic commerce is an AI buying things while you sleep. What is actually happening at scale is assistance: agents doing the research and narrowing the options while the person decides. For considered purchases especially, a sofa, a pair of glasses, a jacket you will wear for years, people still want to browse and choose; agents compress the research, not the desire to choose. The model has two layers maturing at different speeds, and the funnel makes the split clear. The top and mid funnel, where AI shopping agents handle discovery, evaluation, and Q&A, is live at scale today and is where the volume is. The bottom of the funnel, agentic checkout, where the agent completes the purchase the shopper approves, has been rolling out unevenly since late 2025, and the mechanics are still being renegotiated; OpenAI's early-2026 retreat from in-chat checkout toward richer discovery, covered in the timeline below, is the clearest public evidence of which layer is durable.
Agentic commerce is not conversational commerce, and it is not the chatbot on your own site. Those are friendlier interfaces a merchant offers a human. Here the agent works for the shopper, not the merchant, and it operates across every store at once. That one fact drives everything else in this guide.
How is agentic commerce different from traditional e-commerce?
In traditional e-commerce a human browses alone. Discovery happens in search, social, or on your homepage. Conversion happens on a product page built to persuade. Analytics see sessions and clicks. In agentic commerce, the shopper brings an agent along: discovery happens inside an AI answer, the agent reads your store so the human can decide, conversion depends on structured data the agent can parse, and your analytics see an agent crawl, if they see anything at all.
Traditional e-commerce
- DiscoverySearch, social, or your homepage
- ReadingA human browses alone
- ConversionA product page built to persuade
- AnalyticsSessions and clicks
Agentic commerce
- DiscoveryInside an AI answer
- ReadingThe agent reads your store so the human can decide
- ConversionStructured data the agent can parse
- AnalyticsAn agent crawl, if they see anything at all
The practical consequence: your website shifts from destination to data source. If an agent cannot read your catalog, your brand does not exist in the answer, no matter how persuasive the page looks to a human. You can rank first on Google and still be invisible to ChatGPT. We break the full comparison down in agentic commerce vs traditional ecommerce.
One thing has not changed: the traffic is worth having. Adobe's engagement data shows AI-referred visitors behave like better shoppers, not worse ones. The numbers are in the merchant section below.
How did agentic commerce happen so fast? 2023 to 2026
2023 to 2024: assistants that recommend
ChatGPT plugins from Instacart, Klarna, and Shopify in early 2023 were the first experiments in AI-mediated shopping. Amazon launched its Rufus assistant in February 2024. In November 2024, Perplexity shipped Buy with Pro, the first checkout completed inside an AI answer.
2025: the protocol land-grab
In April 2025 the payment rails moved: Visa announced Intelligent Commerce and Mastercard announced Agent Pay within a day of each other, both tokenized frameworks for agent-initiated payments. In September 2025, Google announced the Agent Payments Protocol (AP2), OpenAI and Stripe launched Instant Checkout in ChatGPT starting with Etsy and open-sourced the Agentic Commerce Protocol, and Shopify announced its ChatGPT integration the same day. By December, Salesforce was reporting a record $336.6 billion Cyber Week with AI agents influencing roughly one order in five.
2026: course corrections and defaults
In early 2026, OpenAI scaled back the first version of Instant Checkout, saying it did not offer the flexibility it aspired to provide, and pivoted to richer product discovery plus merchant-built ChatGPT apps. Walmart's app handles account linking, loyalty, and Walmart payments natively. Shopify put millions of merchants into ChatGPT product discovery by default, and Target, Sephora, Nordstrom, Lowe's, Best Buy, Home Depot, and Wayfair integrated ACP for discovery.
The lesson of the timeline: the bottom of the funnel keeps changing, but the top and mid funnel grew through every pivot. Checkout mechanics are still settling; AI-mediated discovery and evaluation, where agents help people research and choose, is the durable layer, the one carrying the volume, and the one merchants can act on today.
Shopping is changing fast. People are discovering products in AI conversations, not just through search or ads.
Who are the key players in agentic commerce?
Four protocol stacks matter, and they attack the same problem from different directions. OpenAI and Stripe's ACP defines how agents read catalogs and complete checkout. Google's AP2 launched in September 2025 with more than 60 partner organizations, including American Express, Mastercard, PayPal, Coinbase, and Salesforce. Visa and Mastercard tokenize the payment credential itself. And the merchant platforms, Shopify above all, turn agent distribution into a default setting.
| Player | What it runs | Where it stands in mid-2026 |
|---|---|---|
| OpenAI / ChatGPT | ChatGPT shopping surfaces; the open-source Agentic Commerce Protocol (with Stripe) | Pivoted from its own Instant Checkout to product discovery plus merchant-built apps; merchants stay merchant of record |
| Agent Payments Protocol (AP2), AI Mode, Gemini shopping | AP2's signed Intent and Cart Mandates prove the user authorized each purchase; 60+ launch partners | |
| Perplexity | In-answer checkout (Buy with Pro), PayPal partnership | Earliest mover on agent checkout, live since November 2024 |
| Amazon | Rufus, Buy for Me, agentic auto-buy on price drops | Keeps agents inside its own walls; consolidated into Alexa for Shopping in 2026 |
| Visa / Mastercard | Visa Intelligent Commerce; Mastercard Agent Pay | Tokenized credentials bound to a specific agent, with user-set spending limits |
| Stripe / PayPal | Shared Payment Tokens; ACP payment processing | Tokens scoped to one merchant and cart total; PayPal joined ACP in late 2025; its wallet and merchant network rolled into ChatGPT through 2026 |
| Shopify | Default ChatGPT distribution for its merchants | Millions of stores surfaced with real-time pricing and inventory; orders land in Shopify admin with attribution |
Do not read the table as a race with one winner. The protocols are converging on the same primitives: machine-readable catalogs, scoped payment tokens, and signed proof of user intent. Readiness for one is mostly readiness for all.
How do AI agents actually pay?
Every payment protocol in agentic commerce answers one question: how do you prove a human authorized this exact purchase, at this exact price, when software clicks buy? Three mechanisms carry the industry's answer.
- Stripe's Shared Payment Tokens. The agent receives a token scoped to one merchant and one cart total. It never holds raw card numbers, and the token is useless anywhere else.
- AP2 mandates. Google's protocol uses signed Intent and Cart Mandates as verifiable credentials, so a merchant can prove the user authorized exactly this purchase.
- Network tokenization. Visa's Intelligent Commerce issues AI-ready cards and Mastercard's Agent Pay issues Agentic Tokens, each bound to a specific agent with user-set spending limits and conditions.
- ShopperAuthorizes this purchase, at this price. The decision stays human.
- AgentHolds a token scoped to one merchant and one cart total, useless anywhere else.
- MerchantCharges, and stays the merchant of record.
For merchants, the ACP model keeps you as the merchant of record: the customer relationship, fulfillment, and fraud responsibility stay largely as they are today. Liability and dispute rules for agent-initiated purchases are the open frontier; the industry has not settled them yet, which is exactly why the card networks moved so early.
Soon people will have AI agents browse, select, purchase and manage on their behalf. These agents will need to be trusted with payments, not only by users, but by banks and sellers as well.
Why does agentic commerce matter for merchants right now?
Because the traffic is already here and it compounds. Adobe has tracked AI referrals to U.S. retail sites growing more than fourteenfold since October 2024, and those visitors arrive pre-qualified: an agent already matched them to your product before they clicked.
That last number is the punchline. Most brands are competing for AI-driven demand with a site the agents cannot parse, and the failure is silent: no error message, no analytics alert, no broken page. A competitor simply gets recommended instead.
What should merchants do now?
You do not need to bet on a checkout protocol this quarter. You do need to be readable, answerable, and measurable before the holidays. The end state is an AI-ready storefront: a version of your store that agents can parse, quote, and recommend. In order:
- Audit your AI visibility first. Find out what ChatGPT, Perplexity, and Google's AI actually say about your brand and your category queries. You cannot fix what you cannot see.
- Make your catalog machine-readable. Structured Product and Offer schema, clean feeds, accurate real-time pricing and inventory. Agents skip pages they cannot parse.
- Give agents an answerable version of your store. Publish content on your own domain that answers real comparison and best-X-for-Y questions, not just product specs.
- Get protocol-ready without betting on one winner. Shopify merchants get much of this by default; everyone else should evaluate ACP endpoints and watch AP2. The protocols share the same primitives.
- Measure AI referrals separately. Tag and monitor AI-source traffic. It converts differently, and it is your leading indicator.
- Do not panic-build checkout integrations. The lesson of OpenAI's 2026 pivot is that discovery readiness pays off now, while checkout standards are still settling.
Where Aigency fits
Full disclosure: this guide is published by Aigency, and the checklist above is the work we do. Aigency hosts a parallel, agent-readable version of your store on your own domain, so steps two through four ship without re-platforming or engineering work on your side, and the free scan is step one: it shows what ChatGPT already reads on your site, what it gets wrong, and where the answer goes instead. If you would rather build in-house, the checklist is the complete map. Either way, start with the audit.
Frequently asked questions
What is agentic commerce in simple terms?
It is shopping where an AI helps you do the work, while you make the call. You tell an assistant like ChatGPT what you need, and it searches across stores, compares options, answers questions, and narrows the choices; on some platforms it can also complete the purchase once you decide. For merchants, it means the next customer may be an AI agent reading your site instead of a person browsing it.
How is agentic commerce different from conversational commerce?
In conversational commerce, a human chats with a bot, often the merchant's own, as a friendlier interface to one store. In agentic commerce, the AI works for the shopper and executes tasks across many merchants at once: discovery, comparison, and checkout. The merchant's site shifts from destination to data source.
Can ChatGPT actually buy things for people?
Partially, and the mechanics are still evolving. OpenAI launched Instant Checkout in September 2025 with U.S. Etsy sellers live and Shopify merchants announced to follow (fewer than 15 ever went live in-chat), then scaled the first version back in early 2026 in favor of richer product discovery and merchant-built ChatGPT apps like Walmart's, which handle checkout, loyalty, and payments natively. Discovery-driven referrals to merchant sites kept growing through the change.
What is the Agentic Commerce Protocol (ACP)?
An open standard co-developed by OpenAI and Stripe, released in September 2025 under an Apache 2.0 license. It gives AI agents a shared language for reading merchant catalogs, pricing, and inventory, and for completing checkout with Shared Payment Tokens so agents never hold raw card numbers. Merchants remain the merchant of record, and retailers including Target, Sephora, Nordstrom, Best Buy, and Home Depot have integrated it for product discovery.
Is it safe to let an AI agent pay for things?
The major protocols replace raw card details with scoped tokens and signed authorizations. Stripe's Shared Payment Tokens are locked to one merchant and cart total, Visa and Mastercard issue tokenized credentials with user-set spending limits, and Google's AP2 uses signed mandates to prove the human approved exactly what was charged. The user confirms before purchase; liability rules for disputes are still being worked out.
How do I get my products to show up in ChatGPT or Perplexity?
Start by auditing what AI engines currently say about your brand, because most stores have blind spots they have never measured. Then make your catalog machine-readable with structured product data and accurate feeds, and publish content that answers real comparison questions on your own domain. Adobe's May 2026 data scores AI readability of retail product content between 47% in furniture and home and 63% in cosmetics, in many categories barely half, so fixing readability puts you ahead of most of your category.