Short answer. Yes, AI agents can already buy from websites, and whether they can buy from yours depends on five things you control: prices an agent can see, forms it can fill, standard checkout steps, availability data it can check, and policies it can read. An agent that gets stuck does not call your sales team; it buys from a competitor instead.
Key takeaways
- AI agents from OpenAI and Anthropic already operate real websites on customers' behalf, and the Agentic Commerce Protocol (OpenAI and Stripe, September 2025) makes in-chat purchasing standard infrastructure.
- Whether an AI agent can buy from a given website depends on visible prices, standard labelled controls, structured offers, predictable checkout and readable policies.
- Hidden pricing and fragile custom checkout interfaces quietly send agent-carried demand to competitors.
- Agent traffic should be measured on task completion, not engagement, because normal metrics such as pageviews and scroll depth make agent sessions look worthless.
- Most early agent-readiness wins are clarity fixes rather than engineering projects, so a business can start with a self-test this week.
What exactly is an AI agent, and is this real yet?
An AI answer tells a customer which hotel to book; an AI agent books it. Agents plan a task, open a browser, click, type, compare options and, increasingly, pay, all on the customer's instruction, and they already exist in shipping products.
An AI agent is software that carries out a multi-step task on a person's behalf, operating websites the way a human would. OpenAI's ChatGPT agent (July 2025) carries out multi-step tasks in a virtual browser, and Anthropic gave Claude "computer use" in October 2024, meaning it can look at a screen, move a cursor, click and type.
What this changes for you: the agent skips your funnel. No landing-page persuasion, no pop-up it reads, no brand video it watches. It checks your price against the customer's budget and your policy against their conditions, such as "under $200, free cancellation", then either completes the task or moves on. Clarity and functionality, not charm, win agent-carried demand. This is also where answer engine optimization stops and agent readiness begins: the answer ends at a recommendation and a link, while the agent carries on through comparison, form-filling and payment.
Can agents complete a purchase on my site today?
Probably, if a careful first-time human visitor could. Agents act on what is actually rendered, such as button labels, form fields, visible prices and error messages, so they succeed on sites with standard controls and fail on sites that hide the information they need.
They stumble on exotic custom widgets, surprise pop-ups mid-checkout, and prices hidden behind "contact us". A quick self-test your team can run this week: have someone who has never used your site complete a purchase or booking without asking anyone for help, narrating every point of confusion. Every stumble they hit, an agent hits harder, because an agent cannot improvise around ambiguity the way a person can. Crawler access matters too, so review which AI bots to allow and which to block before assuming agents can reach your pages.
What is the Agentic Commerce Protocol, and do I need it?
The Agentic Commerce Protocol is a shared standard that lets an AI agent, a shopper and a business complete a purchase together. You do not strictly need it today, because agents can already operate ordinary websites, but you should decide this quarter whether and when to adopt it.
The Agentic Commerce Protocol (ACP) is an open standard, co-developed by OpenAI and Stripe, that defines how an agent, a buyer and a merchant coordinate an order, its payment and its fulfilment. It was released in September 2025 alongside Instant Checkout in ChatGPT, with Etsy sellers first and Shopify merchants announced as next. You keep your own systems and your customer relationship; the protocol handles the order, payment and fulfilment handshake.
The barrier is low: Stripe says merchants already on its payment rails can enable agentic payments with minimal integration work. That puts in-chat checkout within reach of mid-sized brands, not just platform giants. The flow runs as follows: the customer instructs the agent, the agent reads your product data, an order is created through the protocol, Stripe-powered payment completes it, and you fulfil and keep the customer.
What should I fix first to make my site agent-ready?
Fix the items that stop an agent from seeing a price, completing a form or trusting a policy, in this order: published prices, standard controls, structured offers, a surprise-free checkout, quotable policies, and a protocol decision.
Agent readiness is the degree to which an agent can find, verify and complete a transaction on your site without human help. It shares foundations with AEO but adds working forms, checkout and policies. Six items, in order of impact:
- Publish your prices. "Contact us for pricing" is an instruction an agent cannot follow; it will substitute a competitor whose price it can verify.
- Use standard, labelled controls. Real buttons, properly labelled form fields, native date pickers. Agents, like screen readers, depend on them.
- Mark up your offers. Structured data for products, prices, availability and opening hours keeps your facts readable outside the page design; see the technical AEO checklist and the guide to structured data and entity identity.
- Remove checkout surprises. Pop-ups mid-flow, forced account creation and vague error messages are each a point where an agent abandons you.
- Write policies a machine can quote. Returns, cancellation and delivery terms in plain, extractable sentences, because agents check them against the customer's conditions.
- Put ACP on the roadmap with a named owner and a decision date.
How will I see AI agent traffic in my analytics?
Agent traffic looks strange, small and more valuable than it appears. Agents execute rather than browse, so their sessions are short, with few pageviews and no scroll depth, and purchases may complete inside a chat with no referral path your dashboard recognises.
Judged by your normal engagement metrics, your best new customers will look like your worst traffic. So give this segment its own yardstick: identify agent sessions (declared user-agents, protocol-originated orders, machine-fast navigation) and track one number, which is whether the agent finished the task. A rising abandonment rate among agent sessions is your early warning that your flows only work for humans. For broader measurement, see how share of answer tracks visibility in AI responses.
How does agent readiness connect to AI search visibility?
Agent readiness and AI search visibility are two halves of the same job: being the option an AI recommends and the option an AI can actually buy from. A brand that is cited but cannot transact still loses the sale.
Lifewood Data Technology provides AEO/GEO and AI data services, so agent readiness sits close to its work. On the training side, its AI data service line covers annotation, RLHF/SFT/evaluation and field collection, the kind of interaction data that teaches models to operate interfaces. On the visibility side, its AEO services help brands become citable, and the GEO service page covers the generative-engine side. Stripe and Shopify are building the transaction rails; Lifewood's focus is the content-and-data layer. To compare providers, see the best AEO and GEO agencies, a list Lifewood publishes and appears in.