AI as buyer or consultant

In e-commerce we could’ve heard a lot about artificial intelligence taking over buying decisions for customers. People in the industry were speculating whether clients will automate the whole customer journey, from awareness and discovering the product, through picking the best alternative to buying the product.

However recent research has shown that this might not be our closest future. So what is? How AI will change the path to purchase?

What are the current trends of AI use in e-commerce?

Let’s start with the obvious. AI is no longer a niche tool. According to Eurostat, almost a third of people aged 16 to 74 in the EU (32.7%) used generative AI tools in 2025. Among those aged 16 to 24, it was nearly two in three.

Shopping is quickly becoming one of the ways we use it. In a December 2025 McKinsey survey of consumers in France, Germany and the UK, 38% said they use AI tools to research products or decide what to buy. Capgemini’s 2026 consumer study, covering 12 countries, seven of them European, points in the same direction. One in four shoppers used generative AI shopping tools in 2025, and almost a third more plan to start.

So what exactly are we asking it to do?

Imagine you need a new robot vacuum. You have a budget, a flat with a few rugs and a dog that sheds. A few years ago, that meant an evening of open tabs, spec sheets and review sites. Now you can describe your situation to a chatbot and get a shortlist with the differences explained in plain language.

That is exactly where AI is gaining ground. McKinsey found that comparing brands, models, prices and reviews is the most common shopping task people hand over to AI, followed by learning about a product category in the first place. Use drops off sharply as shoppers move closer to the basket and the payment page.

In other words, AI has made itself at home at the beginning of the journey, where information is scattered and comparing it takes time. The question is whether it will move further down the funnel, and whether shoppers will let it.

The industry is ready to let AI pay

If you only followed the news from the payments and tech industry, you could easily conclude that agentic shopping is already here.

In September 2025, Google announced its Agent Payments Protocol (AP2), an open standard for payments made by AI agents, developed with more than 60 payments and technology companies.

Europe followed quickly. By June 2026, Mastercard reported that all of its European issuers were enabled for Agent Pay and that banks had completed live purchases made by AI agents. In July, Visa announced that AI agents had browsed, selected and paid for products at European merchants such as lastminute.com and Frasers, with 31 banks taking part, including Barclays, Revolut and Klarna.

So the rails exist. An agent can find a product, pick it and pay for it, within limits the shopper has set.

The industry itself, however, is less certain than the headlines suggest. The European E-commerce Report 2026 describes agentic AI as still at an early stage and notes that national e-commerce associations disagree on how fast it will develop and how much it will matter. Among the open questions it lists are trust, visibility in AI-driven interfaces and who will own the relationship with the customer.

The technology is answering “can it be done?”. The more interesting question is whether shoppers want it done for them.

Where European shoppers draw the line

The consumer data tells a calmer story. Shoppers are happy to let AI think with them. They are much less keen to let it act for them.

McKinsey’s European survey shows this pattern clearly. Trust in AI is highest for tasks such as summarising reviews or suggesting the best option. It falls as soon as AI moves towards action: filling the basket, completing checkout or reordering automatically. The authors sum it up neatly: consumers are “more willing to outsource cognition than control”.

Data from Poland shows the same picture. In Gemius’s “E-commerce in Poland 2026” report, 71% of Polish online shoppers said they would not let an AI assistant choose and buy products on their behalf, even everyday items. For bigger, more expensive purchases, resistance was only slightly higher.

Willingness also depends on what we are buying. A Sopra Steria study of 8,400 consumers in eight European countries found that 45% would let an AI agent handle electronics purchases, but only 16% would do the same with groceries. Germany and Norway turned out to be the most open markets, while shoppers in France, the UK, Spain and Belgium were the most reluctant.

Even those who are open to it want conditions. In Capgemini’s 2026 study, 76% of consumers said they want to be able to set limits on AI assistants.

The picture is consistent across countries and studies. AI is welcome as a consultant. As a buyer, it is still on probation.

Is it about trust, or about control?

The easy explanation would be that we simply do not trust AI yet. But that does not quite fit the data. We already trust it enough to compare products, read reviews for us and suggest the best option. And the Polish results show that resistance barely changes between a cheap household item and an expensive purchase, so financial risk is not the main barrier either.

Academic research points to something more personal: the need to feel that the decision is ours.

A 2026 systematic review on agentic AI in e-commerce, by S. Balaskas, points in this direction. When an AI system signals too much independence, users can react with resistance and a sense of powerlessness. According to the review, acceptance depends less on how autonomous the system is and more on whether users keep oversight, give consent and have a way to undo or challenge its actions.

Put simply, shoppers do not need AI to be less capable. They need to stay the author of the decision.

What does this mean for online stores?

If AI is a consultant rather than a buyer, the priorities for e-commerce look a little different from what the headlines suggest.

Make your products easy for AI to explain. Comparison is the task shoppers hand over most often, so the quality of an AI answer depends on what it can find about your offer. Complete descriptions, consistent specifications and clear information on price, delivery and returns decide whether a product makes the shortlist at all.

Present AI as help, not as a replacement. Research on agentic AI suggests that tools signalling too much independence can leave users feeling powerless. An assistant that suggests, explains and leaves the choice to the shopper is easier to accept than one that “knows best”.

Build delegation with guardrails. Where stores do offer automated actions, such as reordering or prefilled baskets, McKinsey’s research suggests shoppers are most comfortable when those actions are explicitly approved, easy to check and easy to reverse. Spending limits, a final confirmation step and a clear explanation of why a product was recommended, all help. Two thirds of consumers said they trust AI more when it explains its recommendations.

Treat transparency as part of the experience. Since 2 August 2026, the transparency obligations in Article 50 of the EU AI Act have applied, including telling people when they are interacting with an AI system. Shoppers’ expectations may go further than the legal minimum, so clear labelling is also a way to build trust.

So, buyer or consultant?

For now, the answer is fairly clear. European shoppers are happy to let AI do the hard work: search, compare, summarise and explain. The final decision, and the click that spends their money, they prefer to keep.

That may change. The payment rails are already in place, and as tools improve and people get used to them, delegation will probably grow, starting with selected categories and with clear limits. McKinsey captures this in the title of its report: decision influence is here; execution is coming.

Until then, the stores that win will not be the ones that remove the shopper from the journey fastest. They will be the ones that help shoppers decide faster and with more confidence, while leaving the last word to them.