For most of the history of online stores, the person reading the product page and the person pressing the buy button were the same. That assumption is starting to loosen. A growing number of shoppers now ask an AI assistant to compare options, check delivery times, find a size in stock and, in some cases, complete the purchase for them.

For a small or mid-sized retailer, this raises practical questions rather than philosophical ones. If a customer’s agent visits your store before the customer does, what does it need to find? How do you know the agent is acting with the customer’s permission? Who is responsible if the order is wrong? And how do you keep a relationship with someone who may never see your homepage?

This article looks at what has been announced, what it means for an ordinary store and what you can prepare now without betting the business on any single platform.

What changed in agent-led shopping

Two announcements from the past year show how quickly the infrastructure is forming. On September 16, 2025, Google Cloud announced the Agent Payments Protocol (AP2), an open protocol for agent-led payments developed with more than 60 organizations, including payment networks and technology companies. The post noted that today’s payment systems generally assume a human is directly clicking “buy” on a trusted surface, and that autonomous agents break that assumption.

On January 11, 2026, Google announced the Universal Commerce Protocol, described as a new open standard for agentic commerce across the entire shopping journey, from discovery and buying to post-purchase support. The announcement said it establishes a common language for agents and systems across consumer surfaces, businesses and payment providers, and that it is compatible with existing protocols including AP2. It also said the protocol would power a checkout feature on eligible Google product listings in AI Mode in Search and the Gemini app for eligible U.S. retailers, with retailers remaining the seller of record.

None of this means every store must integrate a protocol tomorrow. It does mean the shape of the problem is clear: agents need structured, trustworthy information, and merchants need a way to trust the agents.

What an agent needs to find on your site

An AI assistant comparing products works with whatever it can read reliably. Pages that bury key facts in images, hide prices until a pop-up is dismissed or describe shipping only in a vague paragraph make the assistant’s job harder and its answer less accurate.

Start with the basics a careful human shopper would also want. Each product should have a clear name, a complete description, current price, availability, variants such as size and color, and identifiers where they apply. Policies should be written plainly and kept on stable pages.

Shipping and returns deserve special attention. Google’s November 12, 2025 post More ways to share your shipping and returns policies with Google said shipping speed, cost and return policies are critical factors for online shoppers. It described two ways for merchants to provide that information: configuring policies in Search Console or adding organization-level shipping policy structured data, with product-level policies still taking priority for specific items. If a store’s own policy page is unclear or contradicts its markup, an assistant may relay the wrong answer to a customer.

A good test is to ask an assistant yourself. Pick a product, ask about price, delivery to a specific city and the return window, and compare the answer with what your store actually offers. Mismatches point to content that needs clarifying. Repeat the test with a product that has several variants, since sizes, colors and bundles are where answers most often go astray. Keep a simple log of each question, the answer and the correction you made, so you can see whether the fixes take hold over the following weeks.

Hands in green sleeves pack lilac tissue into an orange box on a lilac bench beside tape.

Trust, permission and accountability

The harder questions are about authority. When an agent places an order, how does the merchant know the customer really wanted it?

The AP2 announcement framed the problem around three questions: authorization, proving the user gave an agent specific authority for a particular purchase; authenticity, letting a merchant be sure the request reflects the user’s true intent; and accountability, determining responsibility if a fraudulent or incorrect transaction occurs. Its proposed answer uses what it calls mandates, cryptographically signed records of a user’s instructions. In a real-time purchase, an Intent Mandate captures the request and a Cart Mandate records the user’s approval of the exact items and price. For delegated tasks, the user signs a more detailed Intent Mandate upfront with conditions such as price limits and timing.

For a store owner, the takeaway does not depend on adopting AP2 specifically. It is that agent purchases will increasingly arrive with evidence of what the customer approved. Your order system should be able to keep that evidence alongside the order, and your customer service team should know how to read it when a dispute comes in.

Ask your payment provider how it handles agent-initiated transactions today, what fraud screening applies and what records it keeps. Ask the same of any platform that offers to list your products in an AI shopping experience. If the answers are unclear, that is useful information too, because it tells you which questions to raise again before you rely on the channel for meaningful sales.

Keep the customer relationship when the agent does the clicking

A buyer who purchases through an assistant may never browse your category pages or see your brand story. That changes where loyalty is built.

The order confirmation, the shipping updates, the packaging and the first support conversation become the main places a customer meets your business. Make sure confirmation emails are clear about who sold the item and how to reach you. Include a plain explanation of how returns work. If you offer an account or loyalty program, explain its benefits after purchase rather than requiring it beforehand.

The UCP announcement also described Business Agent, a way for shoppers to chat with brands directly in Search, answering product questions in a brand’s voice. Whether or not you use that particular tool, it signals that the words describing your products and policies may be quoted, summarized or spoken by software. Product copy that is specific, consistent and honest serves both human readers and assistants. Our copywriting team writes descriptions and policy pages with that dual audience in mind.

Design for a person who arrives halfway through

Not every agent-led purchase will be fully automatic. Many shoppers will let an assistant narrow the choices, then visit the store to finish. Those visitors land deep in the site, often on a product page, with a specific question already half answered.

That makes the product page carry more weight. The price, options, delivery estimate and return policy should be visible without hunting. The checkout should accept a visitor who has never been to the homepage and should not demand account creation before payment. If the assistant described a particular variant or offer, the page should make it easy to confirm that exact item.

Our UI/UX design work often starts with these deep-entry journeys, tracing what a visitor sees when they skip the front door entirely.

A man in an orange jacket smiles as he picks up a lilac parcel from his front step.

When the agent gets it wrong

Assistants make mistakes. An agent may order the wrong size, misread a delivery promise or buy an item the customer later says they never wanted. Planning for those moments now prevents confusion when the first complaint arrives.

Start by deciding how your store will treat an order that a customer says their assistant placed in error. Will the normal return policy apply, or will you offer a simpler cancellation window before the item ships? Write the answer down so every staff member gives the same response.

Next, look at what your order records show. If a dispute reaches your payment provider, you will want to see when the order was placed, which channel it came through, what items and price were confirmed and what messages the customer received. Ask whether your current platform keeps that information and how long it is stored.

Finally, train support staff to ask neutral questions. A customer who feels blamed for their assistant’s mistake is unlikely to return. A calm, quick fix often turns an awkward moment into a reason to trust the store.

Questions to ask before joining an AI shopping program

Platforms will keep inviting retailers to list products in new AI experiences. Before agreeing, ask a few plain questions and get the answers in writing.

Who is the seller of record, and who handles customer service and returns? What product data and policies will the platform display, and how often does it refresh them from your store? How are payments authorized and what fraud protections apply? What fees or commissions are charged, and are they different from your existing channels? Can you see which orders came through the program, and can you leave without losing your product listings elsewhere?

Clear answers will help you compare programs on their merits rather than on their novelty. Vague answers are a signal to wait.

Practical steps for the next six months

You do not need to rebuild your store to prepare. A modest list, worked through one item at a time over weeks, covers most of the ground.

First, audit product data. Check that every active product has complete, accurate details and that prices and stock levels on the page match your inventory system.

Second, clarify policies. Rewrite shipping and returns pages in plain language, keep them on stable URLs and make sure any structured data or platform settings say the same thing.

Third, review payments. Talk with your payment provider about agent-initiated transactions, fraud screening and dispute records.

Fourth, test assistants. Ask several AI tools about your products every few weeks and note where the answers are wrong or out of date. Fix the source page first, then check again after the next refresh.

Fifth, strengthen post-purchase communication. Make sure every order produces clear confirmation and tracking messages that identify your business.

Finally, watch the standards without rushing to adopt every one. The protocols described above are new and will change. A store with clean data and clear policies will be ready to connect when an integration makes sense.

Where this leaves a small retailer

Agent-led shopping is not a reason to panic or to chase every announcement. It is a reason to make your store easy to understand, for humans and for the software acting on their behalf.

Some of that work is content and design. Some of it is technical: exposing accurate data, connecting inventory and order systems and keeping integrations secure. If your store’s foundation makes those changes difficult, Motta’s web development team can help you build a setup that serves customers however they choose to shop.