Your store's next sale may be made by an AI agent

For twenty years, an online store was built around one journey: the person arrives, searches, opens a category, compares, reads the product page, adds to basket and pays. The whole industry optimised that path, with better menus, filters, bigger photos and shorter checkouts.

Now imagine the customer does none of that. They simply write that they need waterproof hiking shoes, size 43, under 120 euros, delivered before Friday. The AI searches, compares, checks price, size and delivery time, offers two options, and they reply that they will take the first.

The store made a sale. That person may never have seen the home page.

This has stopped being a prediction. Through 2026, Google and Shopify have been extending the ability to complete purchases inside AI Mode and Gemini, without leaving the conversation.

What already works, and where

According to Shopify's documentation, the Google AI Mode and Gemini channel lets a customer complete a purchase in a Shopify-powered checkout without leaving the chat. For eligible stores, that direct checkout is active by default, though the merchant can deactivate it and send customers to their own store instead. Products reach these experiences through the Google & YouTube sales channel or Google Merchant Center.

And here is the caveat that changes everything for us: to use direct checkout, the store must be based in the United States and sell to US customers. A Portuguese store cannot switch this on today.

It would be a mistake, though, to ignore the shift simply because it has not arrived here. The infrastructure is already scaling. Behind these experiences sits the Universal Commerce Protocol, which Google presented in September as already enabling direct checkout for hundreds of thousands of brands and retailers, a figure that comes from the company itself. The same protocol now allows a cart to be transferred to the merchant's site and different checkout flows to be tested.

Google is also rolling out tools showing merchants how their products appear in AI experiences, and how they compare with competitors across surfaces such as AI Mode and AI Overviews. For years there was SEO to understand how we appeared in search. The equivalent layer is now emerging for when an AI helps someone buy.

The visitor may no longer be human

This is the conceptual shift that matters.

Until now, improving a store meant thinking about the person browsing. There may now be a second consumer of information: the agent working on that person's behalf.

To decide, that agent needs to know what the product is, what it costs, whether there is stock, which variants exist, when it arrives, what shipping costs, what the returns policy is and what specifications it has. If it cannot determine with confidence that the product answers the request, it will not recommend it.

Imagine someone asking for a desk under 120 centimetres wide, delivered assembled. If your store sells exactly that desk, but the width is not stated and there is no information about assembly, the agent has no safe way to conclude it fits. It recommends another. The product exists, the price is competitive, the store works. The information did not.

A nice photo is not enough

Picture two stores with the same product. The first has a name, one photo and two lines copied from the supplier's catalogue. The second has a clear name, brand, reference, material, sizes, colours, measurements, weight, stock, delivery time, shipping costs, returns policy, frequently asked questions and reviews.

Which is easier for a person to understand? The second. And for a machine? Also the second.

That is why preparing a store for artificial intelligence does not start by installing anything. It starts with the data. The catalogue has stopped being a content task and become infrastructure.

One number that shows the direction

There is not much public data on this, but one figure is worth keeping. British retailer John Lewis said in September that searches originating from AI agents had risen from 0.3% to 2.5% in a year.

It remains a small percentage, and that should be said. But it is more than an eightfold increase in twelve months, and the company describes the trend as accelerating.

The awkward part: measurement and trust

If part of the purchase happens inside another platform, part of your visibility disappears.

Shopify's documentation is clear about it: in the Google AI Mode and Gemini direct checkout, Google Analytics and client-side custom pixels do not run, with server-to-server events instead for checkout start and completion. After years of installing pixels everywhere, measurement will have to change. We wrote about that work in Google Analytics is already installed on your website, and it still holds.

There are also checkout customisations that do not carry over to this channel, and categories that are not supported, such as subscriptions, bundles and customisable items. This is not taking your checkout and dropping it inside a chatbot. It is a new sales channel, with its own rules.

And there is a larger question, raised this week. On September 22, Reuters reported that a group of banks including NatWest, Bank of America, ING, Capital One, ASB Bank and Commonwealth Bank of Australia published a set of principles warning that the technology is advancing faster than industry standards and consumer protections.

The risks they list are concrete: agents asking for card details and entering them on third-party sites, steering towards payment methods with weaker protection, fraud, and uncertainty over who answers when something goes wrong. Among other measures, they propose that disclosure be required when an agent takes part in a transaction.

None of this invalidates the model. It shows it is early, and that trust will be part of the infrastructure.

Does the website lose importance?

No, but its role changes.

People will keep visiting stores, wanting to know the brand, see photos and build confidence. Many purchases will carry on exactly as today. The mistake is to frame this as website or AI. It will be website and AI.

Some purchases are about exploring, others are about solving a problem. When someone asks for the right charger for a laptop, cheapest one arriving tomorrow, the best experience is probably not twenty minutes in a shop.

What a Portuguese store would do today

I would not try to switch on a checkout that is not available here. I would prepare the groundwork, which pays off anyway, because it also improves traditional search and conversion.

Tidy the catalogue. Clear, consistent attributes on every product, not only what the page needs to look good, but everything that helps someone, or something, decide whether it fits.

Rewrite product pages. Descriptions should answer the questions a customer asks before buying, rather than repeating the supplier's catalogue.

Ensure structured data. Product, price, availability, brand, identifiers and reviews in a format engines and agents can interpret.

Treat Merchant Center as infrastructure, not as a campaign requirement. A clean, current, attribute-rich feed matters more every month.

Connect stock, price and logistics. An agent can only promise what your systems confirm. Stale stock and unrealistic delivery times get worse when a machine decides based on them.

Start measuring traffic from AI assistants, even if the numbers are small. The point today is to build history for a year from now.

Think about integrations. A store stops being only a visual interface and becomes a system other systems talk to. Anyone with disorganised data and manual processes will struggle far more.

The next contest may not be for the click

For years we competed to get a person to the store, through search, ads, email and partnerships. Agent-mediated commerce opens another possibility: perhaps it is enough for the right information to reach the right agent at the right moment.

We wrote recently that the next advert may not open a website, but a conversation. This article closes the reasoning: after discovery, the purchase can sit inside there too.

The online store will not disappear. It will be, at the same time, an experience for people, a catalogue machines can read, a source of data, and a system connected to stock, payments and logistics.

For twenty years we asked how to bring more people to the store. The next question is different: how to make sure the agents buying on those people's behalf can find, understand and choose our products.

Frequently asked questions

What is agentic commerce? Commerce in which AI agents carry out parts of the buying process, such as discovering products, comparing options, checking conditions and, when authorised, completing the transaction.

Can people already buy inside Gemini? Yes, in eligible stores. For the Shopify integration with Google AI Mode and Gemini, direct checkout requires the store to be based in the United States and selling to US customers.

Can a Portuguese store sell this way yet? Not through that integration. But the protocols are expanding and it is worth preparing catalogue, feeds and structured data now.

Does the website stop being necessary? No. It remains the central asset of the brand and the operation. What changes is that its data now has to be understood by external systems too.

Does Google Analytics measure purchases made inside Gemini? According to Shopify's documentation, traditional client-side pixels do not run in direct checkout, with specific server-to-server events instead.

How Digito helps

We build ecommerce as a system rather than a set of attractive pages: catalogue, product data, stock, payments, invoicing and logistics connected to one another. That foundation is exactly what starts to count when the buyer may be an agent reading data rather than a person browsing.

We treat feeds and Merchant Center as infrastructure, along with structured data and the SEO and GEO work that decides whether your product is read as the right answer. We build the integrations connecting the store to internal systems, and leave measurement ready for traffic arriving from AI assistants.

And we are straightforward about what does not exist here yet. Checkout inside Gemini is not available to Portuguese stores, and we do not sell it as though it were. What can be done today is the preparation, and that pays off even if none of this arrives quickly.

If you sell online and want to know what state your catalogue is in for this scenario, let's talk about it.

Sources