AI may soon do your shopping for you. But whose side will it be on?

The next big change in online shopping may be that you don't visit an online store at all.

Instead, you might tell an artificial-intelligence assistant, “Find me a good cordless vacuum under $300 with replaceable batteries and free returns,” and let the software search, compare and recommend what to buy. Increasingly, the industry envisions going one step further: the AI could select the merchant and initiate the purchase as well.

Payments company Adyen says retailers are already preparing for that world — and worrying about what happens when an AI system rather than the retailer becomes the shopper's main point of contact, according to a Reuters report.

For consumers, however, the more important question isn't whether retailers can preserve “customer loyalty.”

It's whether the AI shopping assistant is loyal to you.

Shopping could get dramatically easier

There's a lot to like about the idea.

Conventional online shopping can require opening dozens of browser tabs, sorting through sponsored listings, deciphering nearly identical model numbers and comparing prices, shipping charges, warranties and return policies.

A capable AI shopping assistant could theoretically do much of that work in seconds.

And consumers are already embracing AI as a shopping research tool. Adobe Analytics data cited by Reuters found that 41% of U.S. consumers used generative AI for online shopping in June. AI-referred shoppers also generated 41% more revenue per visit than consumers arriving at retail sites through conventional channels, according to Reuters.

Earlier Adobe data showed the same shoppers were spending more time researching products and converting to purchases at substantially higher rates than other visitors.

That suggests AI may be especially useful for purchases that require research rather than simple replenishment.

But the convenience comes with a new set of questions.

Why did the chatbot recommend that product?

Google search results, Amazon listings and social-media feeds have taught consumers an important lesson: what appears first isn't necessarily what's best.

The same issue could become much harder to see when recommendations arrive conversationally.

Ask an AI assistant for “the best washing machine under $800” and it may confidently produce three choices. But consumers will increasingly need to know:

Why those three?

Did the system examine the whole market? Only retailers with which its operator has agreements? Products whose data were easiest for the AI to retrieve? Merchants that pay referral fees? Companies offering the AI platform favorable commercial terms?

Or were they genuinely the products that best matched the shopper's request?

Those distinctions could become one of the central consumer-protection issues of AI commerce.

The Federal Trade Commission has long held that material commercial relationships affecting endorsements and recommendations should be clearly disclosed, and its truth-in-advertising principles apply to online marketing as well as older forms of advertising.

How those principles will work when an algorithm is simultaneously search engine, product adviser and purchasing agent is still developing.

Retailers want your data — but so will the AI

Reuters' reporting shows another struggle developing behind the scenes.

Retailers including Walmart, Ulta Beauty and Wayfair are trying to make their products easier for AI systems to discover while also encouraging shoppers to finish purchases on the retailers' own websites.

There's a reason.

When shoppers buy directly from a retailer, the company can learn what they searched for, what they put in their cart, what they bought and how frequently they return.

That information helps retailers personalize promotions and build loyalty programs.

Adyen co-CEO Pieter van der Does told Reuters that one merchant he recently spoke with gets about 70% of its volume through direct channels and wants to keep it that way as chatbot shopping grows.

“Loyalty becomes way more important,” he said, according to Reuters.

But from a consumer standpoint, moving the transaction to an AI intermediary doesn't necessarily eliminate tracking. It may simply change who possesses the most valuable information about you.

A shopping agent could eventually know far more than any individual store: your clothing sizes, preferred brands, household needs, budget, previous purchases and perhaps even the price at which you're willing to buy.

Researchers have already begun examining the possibility that information supplied to shopping agents could inadvertently reveal a consumer's willingness to pay — potentially undermining the very shopper the agent is supposed to represent.

The crucial question: Who does the agent work for?

Consumer Reports has been exploring this question through its “Loyal Agents” initiative.

Its premise is straightforward: AI agents increasingly stand between consumers and the marketplace, so their design should ensure that they act in consumers' interests rather than quietly favoring platforms, advertisers or sellers.

That sounds obvious, but today's internet offers plenty of examples showing why it isn't.

Search engines make money from advertising. Marketplaces earn commissions from sellers. Credit-card companies collect transaction fees. Retailers make more money when customers spend more.

An AI shopping service may eventually have several of those incentives simultaneously.

The best consumer agent would instead behave almost like an extremely diligent personal shopper: compare widely, reveal conflicts, respect a budget, protect private information and explain why it recommends one choice over another.

Whether commercial AI shopping systems ultimately work that way will depend heavily on how their business models develop.

And what happens when the AI makes a mistake?

Delegating research is one thing. Delegating authority to spend money is another.

Researchers studying autonomous purchasing have found that AI agents can produce materially different outcomes depending on the system being used, and that automated negotiations and transactions can sometimes result in overspending or poor deals.

That raises practical questions consumers rarely face with today's search engines.

Suppose you tell an agent to spend “around $500” and it spends $650.

Or it chooses a seller with a restrictive return policy.

Or it buys a slightly different model because the desired one is unavailable.

Or a price falls substantially the next day.

Who is responsible — the consumer, the AI company, the retailer or the payment provider?

Those questions become much more important once AI moves from advice to authority.

Loyalty programs could become the battleground

Retailers aren't giving up easily.

Ulta Beauty told Reuters that customers arriving through Gemini and ChatGPT were showing roughly twice the conversion and purchase intent of other shoppers. Yet company executive Josh Friedman also described the cost of relying on outside platforms:

“There's Retailers are preparing for a shopping journey in which a chatbot may recommend a product, select a merchant and initiate a payment before a customer ever visits a store’s website. Payments company Adyen said the shift is pushing merchants to protect direct relationships and repeat purchases, according to a Reuters’ report.

The stakes are visible in the traffic. A merchant that Adyen co-CEO Pieter van der Does spoke with recently generated 70 percent of its volume through direct channels and wanted to preserve that share as more shoppers began searches through chatbots, Reuters reported. “Loyalty becomes way more important,” van der Does said.

Earlier research showed why retailers want the traffic even while they resist losing the customer relationship. AI agents were expected to direct $8 billion in retail spending this year, 41 percent of U.S. consumers used generative AI for online shopping in June, and visitors referred by AI services generated 41 percent more revenue per visit than visitors arriving through traditional channels, according to Reuters.

Walmart, Ulta Beauty and Wayfair were among the retailers updating their websites to appear in chatbot recommendations, while encouraging shoppers to complete purchases on their own sites so the companies could retain browsing, basket and purchase data, Reuters reported. Ulta’s Josh Friedman said shoppers finding products through Gemini and ChatGPT showed “double the conversion and intent,” but also said, “There’s always a tax for engaging customers on other people’s platforms,” according to Reuters.

For shoppers, the change may look like convenience: a natural-language request can narrow products, prices and features faster than a conventional search bar. For retailers, it raises a fundamental question about ownership of the customer. The company that controls the final checkout may also control the data used to shape the next recommendation, loyalty offer and purchase.always a tax for engaging customers on other people's platforms.

That helps explain why loyalty programs are likely to become deeply intertwined with AI shopping.

Retailers may offer special prices, points or perks to persuade customers — or their AI agents — to purchase directly.

Those discounts could benefit consumers.

But they could also complicate comparisons. The cheapest advertised price may not be the cheapest price after membership discounts, credit-card rewards, shipping fees, subscriptions or loyalty points are included.

A genuinely consumer-oriented shopping agent would need to calculate all of those factors rather than merely compare sticker prices.

Consumers aren't surrendering the decision yet

Despite predictions of autonomous shopping, consumers still appear cautious about handing an AI complete control.

A recent survey of 1,463 U.S. online shoppers by Product.ai found that 43% had used AI to research products during the previous 90 days — but among those users, 86% said they checked the AI's recommendation against another source before purchasing.

(That may be a healthy habit for some time.

AI can be an extraordinarily efficient research assistant. But today's systems can still make mistakes, overlook alternatives and provide answers without making all of the economic incentives behind those answers obvious.

For now, consumers may be better served by treating AI as a comparison-shopping assistant rather than an autonomous buyer.

What to check before letting AI shop for you

Before allowing a chatbot or shopping agent to make — or eventually complete — a purchase, consumers should consider a few basic questions:

Ask why it recommends something. A useful system should be able to explain which specifications, prices or other criteria produced its recommendation.

Ask what stores it searched. “Best price” means little if the system checked only a limited group of sellers.

Look for commercial relationships. Pay attention to sponsored products, affiliate arrangements or other disclosures that could affect rankings.

Verify expensive purchases independently. Check the manufacturer's site, reputable reviews and at least one conventional shopping source before spending substantial money.

Compare the entire transaction. Include shipping, membership requirements, return fees, warranties and loyalty rewards — not merely the advertised price.

Limit purchasing authority. If an agent can eventually buy without asking permission, use spending limits or require approval above a specified amount whenever those controls are available.

Know who handles returns and disputes. Before an AI places the order, determine which merchant is actually selling the item and whose refund policy applies.

AI may eventually become the most powerful comparison-shopping tool consumers have ever had.

But that advantage depends on something the retail industry's emerging fight over customer loyalty makes increasingly clear:

The most important loyalty program in AI shopping may be whether the machine is loyal to the company selling the product — or to the person buying it.