Agentic Commerce FAQ
Short answers to common questions about agentic commerce from consumers, brands, developers, and payments teams, with links to deeper pages.
In short: Concise answers to the questions people most often ask about agentic commerce, grouped by audience. Each answer links to the page that covers the topic in full. Facts are current as of October 2026.
General
What is agentic commerce in one sentence?
It is shopping in which an AI agent does some of the work of buying (finding, comparing, checking out, or paying) on a person’s behalf, within limits that person sets. See what is agentic commerce.
Is this actually happening, or is it still hype?
Both, depending on the stage. AI-assisted product research is mainstream: Amazon says Rufus helped more than 300 million customers in 2025. Agent-completed checkout is live at Google, Microsoft, Meta, Amazon, and others, but volumes are still small relative to e-commerce overall, and OpenAI scaled back its in-chat checkout in March 2026. See the timeline and market landscape.
How is it different from a chatbot on a store’s website?
A store chatbot works for one merchant and usually leaves the buying to you. A shopping agent typically works across many merchants on your behalf and may complete the purchase itself. See what is agentic commerce.
For consumers
Does the agent see my card number?
In well-designed systems, no. Agents increasingly use tokens: limited credentials scoped to a merchant, amount, or agent, issued by your card network, wallet, or payment provider. Some agents use single-use virtual cards. Browser agents that fill in checkout forms may handle card details directly, which carries more risk. See payments and trust.
Will an agent buy something without asking me?
Only if you allow it. Most in-assistant checkouts require you to confirm each purchase. Some features, such as price-triggered auto-buy, act later on a rule you approved in advance. See levels of autonomy.
What happens if the agent buys the wrong thing?
Usually you return it under the merchant’s normal policy, since the merchant generally remains the seller. Card dispute rights also apply, though how they apply to agent errors is still being defined. American Express announced protection for purchases made through registered agents in April 2026. See consumer protection and liability.
Are agent recommendations neutral?
Not necessarily. Agents may be influenced by data quality, partnerships, transaction fees, or advertising, and disclosure practices vary. Treat recommendations as you would any intermediary’s. See shopping agents.
For brands and merchants
Do I need to do anything to show up in AI shopping agents?
Yes. Agents rely on structured, accurate product data: complete attributes, identifiers such as GTINs, current price and stock, and clear policies. Submitting feeds to the major assistant programs and keeping your site’s structured data correct are the starting points. See for brands and product data.
Is “agent optimization” just SEO with a new name?
It overlaps, but the emphasis shifts from ranking pages for human clicks to supplying facts a machine can compare and act on. Missing attributes can exclude a product entirely from an agent’s shortlist. See agent optimization.
Do I lose the customer relationship if an agent checks out?
In most protocol-based checkouts the merchant stays merchant of record and receives order and customer details. The agent still sits between you and the shopper, which affects marketing and loyalty. See commerce platforms.
Can I block AI agents from buying on my site?
Technically and contractually, merchants can restrict agent access, and Amazon has done so with Meta’s Muse. Legally, an August 2026 US appeals ruling in Amazon’s case against Perplexity held that a user’s agent does not by itself violate federal anti-hacking law, so blocking relies mainly on terms of service and technical measures. Agent identity schemes let you allow verified agents while blocking others. See agent identity and fraud.
Which protocol should I support?
Most merchants reach agents through their commerce platform or payment provider rather than integrating protocols directly. Shopify, Stripe, and Adyen, among others, support multiple protocols. If you build directly, ACP and UCP cover checkout for different assistant ecosystems. See protocols overview.
For developers
What do I need to build a shopping agent?
Product data access (feeds, a catalog API, or search), a way to check out (a protocol integration, browser automation, or handoff), a payment approach (delegated tokens or network credentials), and clear user approval flows. See for developers and how it works.
What is the difference between MCP and ACP or UCP?
MCP is a general way for AI applications to connect to tools and data. ACP and UCP are commerce-specific protocols defining checkout and order objects; UCP can run over MCP among other transports. See protocols overview.
Should my agent use browser automation or APIs?
APIs and protocols are faster, more reliable, and more welcome to merchants. Browser automation covers merchants without integrations but breaks when pages change and increasingly faces blocking. Many agents use both. See how it works.
For payments providers
How does an issuer know the human really approved an agent purchase?
Through agent-bound tokens with user-set controls, signed mandates (as in AP2), step-up authentication at approval time, and agent registration with networks. Networks are adding risk signals; Mastercard began testing an agent-initiated transaction score in September 2026. See payments and trust.
Who is liable when an agent makes an unauthorized or mistaken purchase?
Rules are still forming. Existing card rules apply, but networks, issuers, and agent operators are negotiating how evidence such as mandates shifts responsibility. See consumer protection and liability.
Where do stablecoins and x402 fit?
x402 and similar rails mainly serve machine-to-machine payments, such as agents paying per request for APIs, data, or compute. Consumer retail agent payments as of October 2026 mostly run on card networks and wallets. See for payments providers.
Sources
- Meet Alexa for Shopping — Amazon
- OpenAI Ends Instant Checkout in ChatGPT — The Keyword
- American Express Debuts ACE Developer Kit — American Express
- Amazon Blocks Meta’s Muse AI Agent From Shopping — Dataconomy
- Ninth Circuit Vacates CFAA Injunction Against Perplexity’s Comet AI Agent — Jones Day
- Universal Commerce Protocol — ucp.dev
- Agentic Commerce Protocol — agenticcommerce.dev
- Mastercard Deploys New Risk Insights to Support Agentic Payments — PYMNTS
- Linux Foundation Announces Operational Launch of x402 Foundation — Linux Foundation