AI shopping agents promise to take the hard work out of buying things online: compare prices, filter specifications, judge alternatives and sometimes complete an order. A consumer study released in Germany on 8 October 2026 suggests shoppers are keen on the earlier steps but remain less comfortable surrendering the final decision and payment.
The KPMG consumer snapshot, conducted and analysed by the EHI Retail Institute, found that 73% of respondents could imagine using an AI agent to compare products, prices or offers. The response changed markedly when the imagined task moved from researching a smartphone to paying for it. The gap matters because the most ambitious visions of agent-led commerce depend on people trusting software with the part of shopping that is hardest to undo.
AI shopping agents meet a payment boundary
In the smartphone example described by KPMG, 72% of respondents were willing to use an AI agent to search for information. The share fell to 45% when the task involved a binding payment step. These are answers to survey scenarios, not measured rates of actual agent transactions, and they should not be presented as sales forecasts.
The difference is easy to understand. Research can usually be checked and repeated. A customer can ask another question, compare a review or reject an unsuitable suggestion. Payment immediately introduces money, delivery details and the possibility of a difficult return. If the AI has misunderstood a preference, chosen an incompatible accessory or missed a condition, the mistake becomes more costly.
The distinction also helps explain why a shopper might welcome assistance from a conversational system without wanting it to control a bank card. Delegation is not an all-or-nothing decision. Someone may be happy for software to find three suitable washing machines but want to see the warranty, seller and total cost before buying.
This is a meaningful constraint on how businesses should design AI shopping agents. Removing friction is useful only when the friction was pointless. A deliberate moment of confirmation can be a feature rather than a failure of automation.
Personalisation can help and mislead
An AI assistant that knows a customer’s budget, preferences and previous purchases can make recommendations more convenient. It can narrow a crowded catalogue and reduce repeated searching. Yet the same information can also make a commercial recommendation harder to assess. Shoppers may not know whether a result reflects their needs, an affiliate arrangement, advertising or a retailer’s interest in selling a particular item.
KPMG identified concerns including retailer favouritism, unauthorised use of personal data, unreliable product information and loss of control during a transaction. These concerns are not proof that every existing assistant behaves badly. They identify the conditions that could prevent people from trusting wider automation.
The wider topic of AI persuasion while shopping shows why explanations matter. A friendly conversational answer may feel less like an advert than a traditional sponsored listing, even when it is guiding the user towards a sale. The buyer should still be able to identify who sells the item, how alternatives were chosen and whether commercial influence affected the order.
Advertising transparency online therefore becomes a practical product requirement, not simply a legal disclosure buried in the interface. If the agent cannot explain why it favoured one option, the shopper lacks information needed for an informed choice.
The big platforms are already moving
Amazon shopping assistant details illustrate how the technology is being brought into ordinary online retail. Alexa for Shopping can answer questions within the Amazon app, compare products, use shopping preferences and help with reminders or purchases. The company has also described alerts for price changes and functions that automate buying when an item reaches a specified target.
Those functions operate within the design and limits of Amazon’s own system; they should not be confused with an independent purchasing agent searching every retailer on equal terms. An assistant embedded inside a shop may be particularly convenient for that shop’s catalogue while still lacking a comprehensive view of the market.
That is a reason to understand the boundaries of a comparison. Does it include competing merchants? Are delivery charges included? Are refurbished or older models considered? Does the recommendation favour items with faster shipping rather than better long-term value? A good tool should allow these criteria to be inspected and adjusted.
The appeal is obvious. A person shopping for a laptop or replacing a broken appliance may find natural-language comparisons more approachable than complicated filters. But convenience alone is not evidence that the selected product is the cheapest, most reliable or best suited to the user.
A purchase has more than one price
When an agent compares products, the obvious number is the sticker price. Real purchasing decisions also involve tax, postage, subscription terms, accessories, repair options, warranties, returns and the likelihood that a seller will honour an agreement. An automated checkout may have access to some of these details but not necessarily a complete picture.
Consider a family choosing a printer. One model may cost less today while requiring expensive replacement ink. Another might include a subscription that is inexpensive at first and less attractive later. The correct recommendation depends on how much printing the household actually does. A persuasive assistant could overlook this if its comparison is driven by short product descriptions rather than the full cost of ownership.
The lesson from AI retail purchasing assistants is not that comparison software should be rejected. It is that the buyer needs a clear way to specify priorities and inspect assumptions. An agent can ask whether durability matters more than price, whether a refurbished item is acceptable and whether delivery speed is essential.
These questions may sound less magical than an instant purchase. They are also what make a recommendation genuinely useful. A buyer who understands the trade-offs is better equipped to challenge mistakes before money changes hands.
The survey also highlights a product-design issue: an assistant must know when to ask instead of acting. An ambiguous request to find a good deal should result in comparisons or a clarifying question, not a surprise transaction. Customers should be able to decide whether shopping history influences suggestions, whether commercial partnerships are disclosed and whether private preferences are retained. If a retailer wants shoppers to trust autonomous checkout, it needs to make the boundary between advice and authorisation obvious.
International differences complicate the picture. Payment protections, delivery rights, product availability and the habits of online shoppers vary across markets. Findings from this German consumer study may suggest a broader pattern, but businesses should not assume the same percentages will apply elsewhere. Good implementations will test confidence in the markets they actually serve and report failures as carefully as successful purchases. An assistant should help a consumer understand trade-offs, not simply move them down a sales funnel faster.
What trustworthy shopping automation would need
A credible shopping agent should separate discovery from consent. It could search and shortlist products freely within a user’s instructions, then pause before entering a binding transaction. At that point, the user should see the exact item, seller, total price, delivery arrangement, refund conditions and payment method. A later change in any material term should require renewed approval.
Clear records matter too. A shopper should be able to discover why an order was placed, which instruction authorised it and how to correct a mistake. Automated purchases at a target price can be sensible for well-defined products, but a simple price threshold is not enough when product quality, seller identity or specifications have changed.
The German findings are not a worldwide census of consumer opinion, and acceptance may move as people gain experience with particular products. Even so, the pattern gives companies a useful design constraint. People want assistance that reduces the effort of evaluating options without removing their ability to decide.
The next phase of online retail may not be an agent that buys everything invisibly. It may be a carefully staged relationship in which AI does the comparison work and the buyer retains the final say. That outcome would be less spectacular than fully autonomous commerce, but potentially more attractive to the people actually spending the money.
About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, cybersecurity and the social impact of emerging technology. LiveAIWire is an independent, human-led technology publication using AI-assisted research, editorial production and original AI-assisted editorial illustrations under his direction.