What Happens When the Customer Is Software?

A consumer looking for a washing machine no longer has to spend an evening moving between retailer websites, comparison portals and reviews. A prompt can set the budget, noise level, energy rating, dimensions and delivery date, then return a shortlist with links. The consumer may still make the final choice, but much of the work that once exposed competing brands has already happened elsewhere. Brands are no longer competing only for attention. They also need to appear in decisions filtered by an AI system.

McKinsey measured the shift in a survey of consumers in France, Germany and the UK: 38 percent said they use AI to research products and services or decide what to buy. Among those users, 63 percent use it to compare brands, models, prices or reviews, 55 percent to learn about products and 46 percent to discover new ones. Deloitte reached a similar conclusion from a much larger European sample: 56 percent of the 13’500 consumers it surveyed had already used AI while shopping. Only 17 percent of European AI users used it at checkout.

AI is entering the purchase journey well before the transaction. Consumers are using it to narrow the field, compare alternatives and decide what deserves further attention. That changes where the sale begins.

The shortlist is moving away from the website

Online retail spent much of the internet era built around the website. Search brought shoppers in, product pages presented the range, filters narrowed the choice, and reviews and price helped decide what to buy. Checkout completed the sale.

AI removes parts of that sequence from the retailer’s control. A shopper who asks an assistant for three running shoes below CHF 180 for wet winter roads may see only three products from a market containing dozens of credible alternatives. A company excluded from that answer never gets the opportunity to persuade the shopper with a landing page, navigation design or retargeting campaign.

McKinsey estimates that agentic commerce could eventually orchestrate USD 3 trillion to USD 5 trillion of global consumer spending by 2030. The forecast remains uncertain and fully autonomous purchases are still uncommon. The earlier stages are much further advanced: AI already compares offers, interprets reviews and reduces large categories to a manageable list.

Product information now plays a different commercial role. Retailers have traditionally treated structured data as infrastructure for search engines, marketplaces and internal systems. AI assistants also need it to decide whether a product fits an instruction.

A person can interpret a vague description, notice a delivery caveat in small print or infer that two differently named colour variants are probably the same. Software works better when the merchant supplies exact specifications, current prices, stock, delivery terms, identifiers and return conditions in forms other systems can read.

OpenAI and Google are building around the transaction

The technology companies developing conversational interfaces are now building the plumbing that connects recommendation with checkout.

OpenAI introduced Instant Checkout in the United States in September 2025 with Stripe and Etsy. The Agentic Commerce Protocol behind it allows an AI system to pass an order to a merchant while the merchant remains responsible for payment processing, fulfilment, returns and support. Users explicitly confirm actions before a purchase goes through.

Google has taken another route. Its Universal Commerce Protocol connects merchants with AI-led shopping across discovery and checkout, while the Universal Cart announced in May 2026 works across Search and Gemini and supports checkout through Google Pay at participating retailers. The cart also draws on price history, stock information, loyalty data and payment-method benefits.

Neither system requires the conventional visit to a retailer’s homepage. The merchant still fulfils the order and keeps responsibility for the transaction, but the interface where the customer searches, compares and decides sits elsewhere.

Retailers already know a milder version of that problem from marketplaces. Amazon, Zalando and Booking.com trained companies to compete inside someone else’s interface. AI adds a new layer because the interface is conversational and the number of options shown to the customer may be much smaller.

Product data becomes part of marketing

A retailer selling laptops once optimised a product page for two audiences: the prospective buyer and the search engine. AI shopping introduces a third.

Consider a request for a lightweight laptop under CHF 1’500 with at least 16 GB of memory, strong battery life and delivery before Friday. The system needs facts it can compare. Processor, weight, memory, battery data, price, stock and delivery have direct use. A slogan about “performance without compromise” does not.

McKinsey now describes product data, pricing, availability and fulfilment reliability as inputs into automated buying decisions. Companies whose systems expose that information consistently are easier for AI tools to assess.

The commercial work is therefore not confined to SEO or a new discipline with an acronym. Marketing, e-commerce and product teams have to agree on the underlying facts. A model number should not differ between a manufacturer’s website and a reseller feed. A product listed as available should have an inventory status that reflects reality. Warranty and delivery terms need enough precision for another system to compare them.

None of that replaces brand building. It changes what happens when the customer starts with a need rather than a brand name.

A customer who asks for “black Nike running shoes” has already narrowed the decision. One who asks for “the most durable running shoe under CHF 180 for wet roads” gives the AI more influence over which brands enter the conversation. Customers still recognise and seek out brands they know because consumers can instruct the system to include a company they know and trust. The source material points to the same requirement. Customers need to remember the brand. AI systems need product information they can recognise and retrieve.

Advertising does not disappear, but its job changes

Digital advertising has been built around visibility: put the right message in front of a likely buyer, earn the click and move the customer into a controlled environment.

An AI assistant may compress that journey into a recommendation. The consumer supplies a brief; the software does part of the filtering. A brand can spend heavily on awareness and still lose a category-level query because its product information does not match the criteria as clearly as a competitor’s.

Algorithms do not become neutral procurement officers simply because they sit inside an AI interface. Platforms still have commercial incentives, and retail media, sponsored placement and merchant offers will move into these environments just as advertising moved into search and social.

Disclosure gets harder once shopping moves into conversation. Search pages gave users a visible split between paid and organic results. AI answers often offer a shorter list with fewer clues about why a product appears. Platforms, retailers and regulators now have to show where commercial influence begins when a recommendation reads like an answer, not an advert.

The response goes beyond buying another media format. Paid exposure still creates demand, brand advertising still builds recognition, and product information has to hold up once AI starts comparing criteria rather than impressions.

Consumers trust AI more with research than with money

The payment step remains the boundary. Sopra Steria surveyed 8’400 consumers across eight European countries in 2026. Forty-five percent said they were prepared to delegate purchases of electronics and technology products to an AI agent. Only 16 percent said the same for healthcare and groceries. Banks were the provider consumers most often named as a legitimate operator of shopping agents, but only 27 percent chose them.

Ecommpay found an even clearer gap between interest and authority. Its survey of 1’756 consumers in the UK, France, Germany, Spain and Italy found that 73.4 percent expect AI shopping agents to become common within five to ten years. Only 13.9 percent wanted an agent to complete purchases for them, while 50.1 percent said they would not trust one with their card details.

You can see the difference in everyday buying decisions. “Find me the cheapest direct flight to Zurich next Friday” delegates research. “Book whichever flight you think is best and spend up to CHF 500” delegates judgement and money.

Payment systems therefore need more than technical access to a card. They need to record what the customer authorised, the permitted amount, the merchant involved and the point at which human confirmation remains necessary. OpenAI’s current checkout design keeps the user in that loop and restricts payment authorisation to a specific merchant and amount.

How far customers hand over the purchase will depend on what they are buying and what is at stake. A recurring order of household detergent carries little risk. A CHF 2’500 laptop, an airline ticket or a financial product demands more scrutiny. Delegation is likely to spread first where the cost of a poor decision remains low, and the survey evidence already points in that direction.

The first contest is already under way

Retailers do not need to prepare for a world in which autonomous agents buy everything. The current market gives them a more immediate problem: software already influences which products customers consider.

Deloitte’s 56 percent AI-shopping adoption figure and McKinsey’s 38 percent figure measure different populations and behaviours, so they should not be treated as directly comparable. Both point to the same location in the journey. Consumers have adopted AI faster for discovery, research and comparison than for checkout.

Companies can test their position without waiting for autonomous payment to become common. Ask the leading assistants to find one of the company’s products under realistic customer constraints. Check which facts they retrieve, which competitors appear, whether the price is current and whether availability, delivery and specifications are correct. Then trace each error back to its source.

A retailer that cannot supply consistent product information today will struggle when software does more of the buying work tomorrow. A brand that customers remember by name enters the prompt with an advantage. A merchant that combines both has more control over a journey in which the website is no longer guaranteed to be the first stop.

Online marketing spent twenty years trying to win the click. AI is moving part of the contest one step earlier: into the system deciding what deserves to be clicked at all.

The first job is not to prepare for a future in which software buys everything. It is to make sure software can understand what the company sells today. Clean product data, accurate availability, clear commercial terms and a brand customers still ask for by name give businesses the best chance of staying on the shortlist.

 

What companies need to know now

The customer journey is starting to change from search → website → browse → compare → buy to prompt → AI comparison → shortlist → choice → purchase. AI is taking over part of the research and comparison work before the customer reaches a brand or retailer website.

1. The shortlist is becoming more important than the click
   A customer may ask an AI assistant to find the best product under a set price, compare specifications, check reviews and narrow dozens of options to three. If a product does not appear there, the company may never get the website visit.
2. Product data becomes part of marketing
   Price, stock, specifications, delivery times, warranty terms and product identifiers need to be accurate and consistent across websites, marketplaces, reseller feeds and third-party sources. AI systems use those facts to compare offers.
3. Brand recognition still gives companies an advantage
   A customer asking for “the best laptop under CHF 1’500” gives the AI wide discretion. A customer asking specifically for a Lenovo, Apple or Dell product narrows it. Brand building still matters because customers can bring the brand into the prompt themselves.
4. Reviews and third-party sources matter more.
   AI assistants do not rely only on the company’s own website. Reviews, comparison sites, retailers and independent publications can influence which products the system recommends.
5. Companies should test the journey themselves.
   
   Run realistic customer questions through the leading AI assistants. Check whether your products appear, which competitors are recommended, what facts the AI retrieves and where the information is wrong or incomplete.

The immediate task is to give AI clear, reliable product information so it can find the right item, compare it and put it in front of the customer before they reach the website.