AI shopping participation is a decision about control
Retailers can let customers discover products through an AI assistant while keeping control of the sale, the customer account, and service after purchase. To participate, they must decide which parts of the customer relationship an external agent may handle and whose rules apply. Meta, Google, Microsoft, and OpenAI are moving discovery and checkout into conversational interfaces, and each has a commercial interest in attracting shopping activity to its services.
Growing traffic makes those decisions consequential. Shopify, which benefits when merchants conduct more commerce through its platform, reported that AI-driven traffic to its stores increased eightfold year over year in the first quarter of 2026, while orders originating from AI-powered searches increased nearly 13-fold. Target, which benefits from attracting shoppers to its retail business, reported a 2,000% increase in AI-driven traffic to its digital properties over the same period, compared with nearly 400% growth across retail sites overall.
Those growth rates show expanding activity, but a retailer’s participation decision also depends on channel size, profitability, and the authority an integration grants. The agreement must define control over checkout, identity, loyalty, fulfilment, and customer service. Retailers are already choosing how much of that work an external service may perform.
Product access and purchase permission are separate decisions
Meta’s expansion of Muse, its personal AI agent, shows how many services can become available through a single interface. Meta announced ten retail connectors, integrations linking Muse to Walmart, Best Buy, Gap, American Eagle Outfitters, Sephora, Ulta, Wayfair, DICK’S Sporting Goods, Fanatics, and Michael Kors. It also announced Shop Pay and PayPal as payment options, added Instacart for grocery shopping, and listed Expedia as forthcoming for travel planning.
With that range of services, the permissions behind each connection matter. A connector can provide access to shopping services while leaving the retailer and platform to agree on which actions the agent may perform. For a retailer, the operational question is what Muse may do after a customer expresses interest in a product.
Shopify gives merchants separate controls over product distribution and checkout. Through Shopify Catalog, merchants can select which AI partners receive their product data. Eligible merchants can also decide, channel by channel, whether customers may complete purchases through direct checkout, so sharing a catalogue and authorising a transaction remain separate choices.
These choices produce different purchase experiences across external platforms. Shopify says some environments provide checkout within the interface, while others open an in-app browser connected to the merchant’s existing commerce systems. The interface can therefore change while the merchant uses its established systems to process the sale.
Shopify says merchants remain the merchant of record across supported channels, retaining the seller’s role in the transaction. Orders return to Shopify Admin with the originating AI channel identified, while Shopify’s infrastructure continues handling payment processing, tax calculation, fraud detection, and fulfilment. Merchants also retain their customer relationships and data, Shopify says, keeping externally initiated purchases connected to existing operations.
OpenAI addresses continuing merchant duties through its Agentic Commerce Protocol, a set of rules for connecting purchasing agents with commerce systems. Under that arrangement, OpenAI says merchants remain responsible for fulfilment, returns, support, and customer communication after a purchase through ChatGPT. Shopify’s channel controls and OpenAI’s assignment of post-purchase duties cover different parts of participation: permission to transact and responsibility once the transaction is complete.
A project in mind?
Schedule a 30-minute meeting with us.
Senior experts helping you move faster across product, engineering, cloud & AI.
Customer continuity requires account and loyalty connections
Continuing responsibility is easier to manage when the retailer’s account information remains available during the interaction. For Walmart, keeping its own assistant involved would also preserve its role in serving the customer and completing the sale. At Morgan Stanley’s Technology, Media & Telecom Conference in March, Walmart Executive Vice President Daniel Danker described plans for Sparky, Walmart’s assistant, to operate within external services including ChatGPT and Gemini.
Danker described a sequence that starts when an external AI service identifies a Walmart product in response to a customer’s request. When the customer selects the product, Sparky can take over the shopping interaction while the customer stays in that service. Customers could remain logged into Walmart, allowing its systems to retain information such as products already in their carts and use it in subsequent shopping decisions.
Target connects a similar commercial interest in retaining customers to its loyalty programme. The retailer says shopping experiences through Google’s AI Mode and Gemini, Microsoft Copilot, and OpenAI’s ChatGPT support product discovery and multi-item baskets. Customers can also connect their Target Circle accounts, giving Target’s existing loyalty relationship a role in shopping that begins outside its own digital properties.
Account connections require systems to exchange information, which is the focus of Wayfair’s work with Google on the Universal Commerce Protocol, or UCP. UCP standardises how agents and retailers exchange the information needed for product discovery, carts, checkout, and payments. For Wayfair, connecting those steps creates another route to sales; for Google, it brings more shopping activity into its services.
Wayfair says UCP would let shoppers move from discovery to checkout on eligible Google listings while Wayfair remains the merchant of record. The retailer would continue controlling pricing, fulfilment, and customer service. Under this proposal, customers would interact in Google’s environment while Wayfair retained those commercial decisions and obligations.
Google has also added identity linking to UCP so participating retailers can connect shoppers with existing accounts and loyalty benefits. Agents can retrieve current prices and inventory directly from retailer catalogues, bringing customer information together with the retailer’s current offer. Each capability meets a specific integration need: Walmart’s plan preserves account and cart context, Target connects loyalty, and UCP provides a shared method for exchanging commerce information.
Amazon’s Muse block shows how retailers can withhold permission
These integrations depend on agreement between the retailer and the external service. According to Bloomberg, Amazon asked Meta to remove Amazon from Muse’s shopping experience, Meta declined, and Amazon subsequently blocked Muse from Amazon.com. Blocking began on September 20, placing Amazon outside Meta’s announced group of retail partners.
Amazon and Meta have competing interests in who controls access to Amazon shoppers: Meta wants Muse to serve shopping requests, while Amazon wants authority over how external software uses its store. In statements reported by other publications, Amazon said it had not authorised Muse to access its site and argued that applications purchasing on customers’ behalf should respect a provider’s participation decision. Customers attempting to use Muse on Amazon encountered messages explaining that access by an unauthorised AI agent violated Amazon’s conditions of use.
Amazon put its refusal into practice. Where Shopify offers channel settings and Walmart, Target, and Wayfair describe agreed forms of participation, Amazon requested removal and then enforced its refusal through blocking. The dispute concerns access by Muse; Amazon’s response to other possible AI-shopping agreements remains a separate decision.
Product data makes authorised access useful
Once access is authorised, an agent needs usable product information to help a customer shop. Agents read and compare structured information, data organised into fields software can interpret, alongside the webpages retailers design for people. Participating retailers therefore need catalogues that support software-led discovery as well as customer browsing.
Microsoft identifies product specifications, prices, and availability as important inputs for AI shopping. It has added support for UCP-ready product feeds, structured files of catalogue information, in Microsoft Merchant Center, giving retailers a way to supply that information to Copilot. Microsoft benefits when that information makes its shopping service more useful, while retailers gain a way to present product facts to the system interpreting a shopper’s request.
Google Merchant Center plays a similar role in discovery within Google’s shopping experiences and operates separately from Microsoft Merchant Center. Google says its conversational shopping uses submitted product information and warns that incomplete feeds can prevent customers from finding products through AI-assisted tools. Feed completeness therefore affects whether a product appears among a shopper’s options.
Google’s conversational product attributes extend that requirement to longer, more specific requests. Retailers can provide richer descriptions that match how people explain their needs in conversation. These give Google’s systems more relevant facts to evaluate when deciding which products fit a detailed query.
Stripe extends the preparation work from product descriptions to transaction systems. As a payments provider, Stripe has a commercial interest in making agent-led purchases workable; it advises retailers to make product information, pricing, availability, and policies machine-readable and directly queryable, so agents can request the information they need. Stripe also recommends systems that can accept authorised actions customers delegate to agents, connecting the ability to evaluate an offer with permission to purchase it.
Discovery, orders, and advertising need separate measures
Once an agent can find and evaluate products, retailers need to measure what happens next. Shopify Admin identifies the originating AI channel for supported orders, connecting a purchase to the service that generated it. Google’s AI performance insights in Merchant Center measure a different stage: brand and product appearances across AI Mode and AI Overviews.
Google also compares a retailer’s share of voice, its relative presence, with similar brands across Google’s AI services. Microsoft’s AI Visibility features, still under development, are intended to show how AI systems find, present, and cite business products and information. These measures help retailers observe discovery, while order attribution records a later commercial event; determining whether an appearance caused a sale requires a separate assessment.
Paid placement adds another event to measure because platforms and advertising businesses benefit when brands buy exposure. Roundel, Target’s retail media business, began a ChatGPT advertising test in February that allows Target and participating Roundel brands to place sponsored ads beside relevant shopping conversations. The ads are separately labelled, and their relevance follows the subject the customer is discussing.
Target’s existing referral growth explains its commercial interest in the test. At the announcement, Target said traffic from ChatGPT to its site had been growing by about 40% per month on average. Evaluating the advertising requires separating that growth in visits from the additional effect of sponsored placement.
Google is also testing shopping ads in AI Mode, where sponsored retailers can appear beside organic product recommendations. Its Direct Offers format allows participating brands to present offers during AI-led shopping interactions. Google’s and Target’s tests give brands paid ways to reach customers researching products in external conversational services, extending retail media beyond retailer-owned websites and apps.
Agent permissions need technical enforcement
As discovery, advertising, and purchasing move through external services, commercial agreements need matching technical controls. Retailers and platforms must establish which product information an agent may receive, where checkout can occur, how customer identity will be verified, and which company carries responsibility for the transaction. Shopify, Google, Microsoft, and OpenAI are developing protocols to connect AI interfaces with merchant systems, making those permissions and responsibilities part of communication between separate organisations.
Enforcing the agreement also requires identifying the software making a request. A customer’s intention to shop leaves questions about which agent is acting, what authority it has, and whether its actions stay within that authority. Retail infrastructure therefore needs to distinguish legitimate customer-delegated software from unwanted automated traffic.
Akamai’s network observations show why that distinction matters in commerce. Between July and December 2025, commerce accounted for 47.9% of the AI bot traffic Akamai observed across its network. The denominator covers observed AI bot traffic across sectors, so the figure describes commerce’s share of that activity rather than the share of ecommerce traffic generated by AI bots.
Akamai has a commercial interest in demand for traffic-management and security services, which is relevant context for its findings. Its observation groups AI bot activity without classifying every bot as malicious or as an authorised shopper’s agent. Retailers still need to assess individual agents and requests when granting access to product, account, and checkout systems.
Stripe identifies authorisation, audit trails, fraud detection, and trusted-agent identification as necessary controls for that assessment. Audit trails record actions for later review, while trusted-agent identification helps systems recognise software acting within agreed permissions. Stripe plans to adapt its fraud tools to distinguish authorised shopping agents from unwanted bots, allowing permitted activity to proceed while automation outside those permissions can be restricted or blocked.
In conclusion
Retailers can participate in AI shopping without handing over every part of the customer relationship. Product discovery, checkout, account access, and service are separate decisions, and each integration needs to make those boundaries explicit. Remaining the merchant of record preserves responsibilities as well as control.
For business leaders, the starting point is the authority granted to each agent. Commercial agreements should define what the agent may access and do, while technical systems enforce those permissions and record its actions. Catalogue quality, identity connections, and fraud controls determine whether an agreed shopping experience can work reliably.
Performance also needs to be assessed beyond traffic growth. Product appearances, referrals, completed orders, and paid placements measure different stages of the purchase process. Decisions about further investment should account for channel size, profitability, and the cost of serving those customers.
The practical choice is which agents to admit, under what terms, and for which tasks. Those terms should be specific enough to enforce and flexible enough to revise as customer behaviour and channel economics become clearer.
A project in mind?
Schedule a 30-minute meeting with us.
Senior experts helping you move faster across product, engineering, cloud & AI.


