Your Next Customer Is an AI

How shopping agents are changing what we buy and why brands will soon have to convince more than just humans.

I have spent an absurd amount of time in my life buying things on the internet that are actually completely mundane. Which cordless vacuum is any good? Which headphones work well with multiple devices? Which monitor fits my setup? You open Google, Amazon, and a few reviews, have twelve tabs open and, surprisingly, after half an hour sometimes know less than before.

We call this online shopping, or at least I call it that. Actually, it is unpaid research work.

I believe that AI will radically change exactly this part. Not because there is now another chat window next to the search field. That already exists as a tool on quite a few websites (I never really use it). But because we increasingly will not do the research ourselves at all. At first, I could not really get comfortable with the idea, but somehow I think to myself, why not actually.

Instead of googling “best cordless vacuum 2026,” I could simply tell my AI agent:

I need a quiet cordless vacuum for an 80-m² apartment. Maximum 350 euros. We mostly have hard floors, spare parts should be easily available, and I do not want a manufacturer where the battery is practically electronic waste after two years. Find the three best options for me. (Okay, I probably would never really write it like that, but it is just an example and there are people who would write it even more descriptively).

And at some point, another sentence will be added: Take the best one and order it.

At that point, we are no longer talking about a better search engine. We are talking about a new player in commerce. We will then probably have to invent another fancy term such as AI Shopping Convenience.

We Are Moving From Search to Delegation

Online shopping is still surprisingly manual today. We formulate a search query, skim results, open product pages, close cookie banners, search somewhere between the hero image and newsletter popup for technical specifications, compare prices and try to figure out whether a product is actually good or simply marketed pretty well. If it is marketed pretty well, most people do not even notice. That is the point.

AI shifts this behavior from search to delegation.

This is no longer a particularly futuristic prediction. OpenAI significantly expanded shopping in ChatGPT in 2026: users can discover products based on budget, preferences and constraints, upload images as references and compare alternatives directly with each other. Behind these shopping experiences is, among other things, the Agentic Commerce Protocol, or ACP for short. For merchants, there are now structured Product Feeds through which, among other things, title, description, images, price and availability can be transmitted to ChatGPT. OpenAI recommends, for example, a complete feed daily and additional updates via an API. [1] [2]

Google has also long been working on the same shift. In May 2026, the company introduced its Universal Cart, an intelligent shopping cart that extends across Search, Gemini and, prospectively, other Google products. The cart can, among other things, monitor price changes, point out products that are available again and even identify problems, such as incompatible components when putting together a PC. [3]

What is interesting about this is less the individual feature. What matters is the direction. AI no longer just shows us information. It is slowly moving toward research, evaluation and action. And that is exactly where shopping actually becomes interesting.

The Most Important Visitor to Your Shop Could Soon Be Not a Human

Let us imagine for a moment that I want to buy running shoes. My agent knows my shoe size, my budget, perhaps even my previous purchases. I tell it that I run between five and ten kilometers three times a week, do not want aggressive stability support and want to spend a maximum of 160 euros.

The agent could then search retailers and manufacturers, compare models, evaluate technical specifications, check availability and take customer reviews into account. In the end, I might see three options instead of 300. For me as a buyer, that is fantastic. For brands, it is more complicated.

Because suddenly, it is not exclusively the human who decides which product page is opened. Between demand and provider sits a system that pre-selects. Perhaps nine out of ten products disappear from my selection before I have ever seen a logo, a product photo or an advertisement. This creates a new form of gatekeeper.

The crucial thing is that this gatekeeper has completely different needs than I do. A good brand can reach me emotionally. I like beautiful design, stories, images and, unfortunately, sometimes completely irrationally certain sneakers. An agent, on the other hand, has to be able to understand what a product actually is, what it costs, whether it is available and why it fits my requirements.

If a product page mainly says that the product “revolutionizes my performance,” but the agent cannot reliably determine how heavy it is, which sizes are available or what the actual price is, the marketing department may have written a great text and still created a new SEO problem.

Except that perhaps soon we will no longer call it SEO.

From SEO to AI Visibility and Eventually to Agent Visibility

That is exactly why I find shopping agents so exciting from the perspective of AI Visibility. We are already observing today how part of search is shifting from traditional result lists to generative systems. For companies, this creates an additional question: Not only “How do we rank on Google?” but also “How are we understood, classified and recommended by AI systems?” Shopping agents consistently take this logic further.

SEO deals, very simply, with being found (not only that, of course). AI Visibility deals with being visible and relevant in AI answers. Agentic Commerce adds a third layer:

Will I be selected?

The distinction sounds semantic at first, but it is fundamental. A language model can mention my running shoe in an answer today. Tomorrow, a shopping agent could decide whether my running shoe is even one of the three candidates that a user gets to see. And at some point, an agent could even prepare or carry out the transaction.

The technical infrastructure for this is currently being created. At the beginning of 2026, Google, together with commerce companies such as Shopify, Etsy, Wayfair, Target and Walmart, introduced the open Universal Commerce Protocol. UCP is intended to provide a common language through which agents, merchants and also payment systems can work together across different steps of a shopping journey. The protocol is also compatible with other agent standards such as A2A and MCP. [4]

The interesting thing is: We do not really have to speculate about the theoretical future for much longer. We can currently observe how the interfaces for it are being built. I think that new forms of shopping can emerge here as well. I cannot yet imagine everything that will be possible in the near future and how people will accept it or how quickly it will be accepted.

Product Data Suddenly Becomes Marketing

In traditional e-commerce, product data was often the unloved backend topic. Marketing wanted campaigns, exciting landing pages and beautiful creatives. Somewhere alongside that existed a feed with SKU, GTIN, color, availability and price that hopefully someone kept up to date. In an agentic shopping world, the importance of this inconspicuous data is reversed.

If an AI system is supposed to compare products with each other, it needs information that is as precise as possible, very precise information. This is already visible today: OpenAI's commerce documentation explicitly provides structured Product Feeds so that ChatGPT can understand catalogs and display current product information. The core data includes, among other things, title, description, images, price and availability. [2]

That does not mean that brand management or great product pages become unimportant. Nope. Not at all. On the contrary. As soon as the human makes the final decision, trust, aesthetics and emotion remain crucial. But before this human decision, a machine pre-selection may emerge. Somehow this reminds me of HR systems that have been given AI miracle cures so that they can make a pre-selection.

I consider an old SEO wisdom (hearsay) relevant to Agentic Commerce as well: What is not technically cleanly accessible and semantically unambiguous can be difficult to consider. Simpler: Miserable input, miserable output.

Only this time, the stakes are higher. In the past, poor discoverability might have meant position 14 instead of position 4 on Google. In an agentically curated selection, it could mean not being suggested at all.

The New Conversion Might Happen Before Your Shop

This also calls today's e-commerce metrics into question. We optimize CTR, sessions, add-to-cart rate, checkout completion and conversion rate. That works as long as the user largely goes through the journey themselves.

An agent, on the other hand, could complete a considerable part of this journey outside the shop. It identifies a need, researches different products, eliminates unsuitable candidates, compares prices and then shows its user a very small selection. It may know behavior from previous purchases and constantly learn, so that the agent may know better what the user needs and wants than the user themselves does. Sounds far-fetched, but many people are simply predictable. Of the original 50 potential products, perhaps only three ever receive a human visit.

The most valuable “conversion” may then have taken place before the user enters the website: The agent has decided that product A is considered and product B is ignored.

Google shows how concrete this direction has become. Its Universal Cart is intended to manage products across merchants and enables purchases via Google Pay for participating brands or routing to the merchant for completion. Nike, Sephora, Target, Walmart, Wayfair and Shopify merchants were mentioned, among others. [3] [3]

At the same time, the payments industry is building a trust layer for such scenarios. Visa describes its Intelligent Commerce platform, for example, as infrastructure through which agents can initiate and complete transactions within defined user permissions. Among other things, the company is developing a directory of verified agents and merchants. Mastercard argues similarly: In Agentic Commerce, it must be traceable who is acting, on whose behalf, with what authorization and within what limits. [5] [6] [6]

This is important because there is a huge difference between “Find me the best coffee machine” and “Buy it for me.”

Anyone Who Gives an AI Their Wallet Wants Control

Technically, autonomous shopping is one of the more fascinating use cases for agents. Psychologically, it is one of the more difficult ones.

I would have few concerns if my AI searched twenty online shops overnight and said the next morning: “These three monitors match your requirements.” I would have significantly more questions if one costing 1,300 euros was already standing at my front door the next morning. As things stand today, I would not want to do that or give up control to that extent.

The future of Agentic Commerce will therefore not be decided by whether AI can technically press a buy button. We can solve that ourselves. What will matter is how much freedom of decision we actually want to give an agent. Well, remember the Tesla drivers in all those social media videos who were filmed asleep in the driver's seat of a Tesla.

Maybe my agent is allowed to reorder consumable products under 30 euros on its own. For purchases up to 200 euros, I have to confirm. For a new laptop, it can research, but not order. Maybe it is only allowed to buy from certain merchants or only if returns, warranty and delivery date meet certain conditions.

Exactly these kinds of limits are what make agents practical in the first place. If prices, identities, permissions and payment limits can be reliably transferred between user, agent and merchant, Agentic Commerce changes from a cool demo into infrastructure. The fact that Visa, Mastercard and Google are currently explicitly working on this trust and payment layer is therefore at least as interesting as the progress of the actual AI models. [5] [6] [3]

What Brands Can Learn From This Now

I would therefore still not recommend renaming the entire e-commerce team “Agent Optimization” tomorrow. The market is changing far too quickly for that, and we will probably invent enough new three-letter acronyms over the coming years. Companies no longer have to make their offerings understandable exclusively for humans and traditional search engines. Products increasingly also have to be reliably interpreted, compared and placed within a specific usage context by AI systems.

It starts surprisingly unspectacularly. Are product name, manufacturer and variant unambiguous? Are price and availability up to date? Are important characteristics available in structured form? Are statements on the website, product feed and external sources consistent? Can an AI clearly recognize which use case a product is suitable for and which it is not?

For me, that is exactly the connection between Agentic Commerce and AI Visibility.

Many companies today still primarily view AI Visibility as a new reach problem: “Do we appear in ChatGPT?” That makes sense, but it is probably only phase one. The question is rather whether AI systems understand a company correctly and develop enough trust in its information to include it in a decision or not. In e-commerce, this abstract question suddenly becomes revenue.

Maybe the Online Shop Was Only an Intermediate Step

When online commerce became big, we did not simply copy the physical store digitally. We invented completely new things: search functions, reviews, recommendation engines, one-click checkout, retargeting, marketplaces and comparison portals. At some point, these things seemed completely normal to us.

The same could happen with agentic shopping.

Maybe in ten years we will look back at the way we shop today and find it absurd that we compared twenty nearly identical products ourselves. That we checked prices in multiple tabs.

The AI agent does not necessarily take the purchasing decision away from us. At best, it takes away the boring part before it.

For merchants and brands, however, this convenience comes at a price: They are no longer just competing for our attention. Increasingly, they also have to get through those systems that are supposed to protect our attention.

And that is why one of the most important questions in digital marketing could change surprisingly in the coming years.

Today, companies ask:

How do I get found?

Then came:

How do I become visible in an AI?

The next question could be much more uncomfortable:

Why should an AI choose me of all things?

Having a good answer to that is likely to become more valuable than the next perfectly optimized buy button.

Sources

[1] OpenAI: Powering Product Discovery in ChatGPT
[2] OpenAI: Commerce Product Feeds and ACP
[3] Google: Universal Cart and agentic shopping
[4] Google: Universal Commerce Protocol
[5] Visa: Intelligent Commerce
[6] Mastercard: Agentic Commerce Vision