Shopify is opening its checkout to browser agents. Amazon is setting conditions on outside access. Walmart and Google describe a more controlled handoff.
An AI assistant can recommend a product without being able to complete the purchase where it is sold. That distinction is turning retailer access into a new question for brands: who controls the shopper after a recommendation has been made?
Retailers are giving different answers. In reporting published on 2 October, <em>The Wall Street Journal</em> said Gap, Walmart and Best Buy had publicised partnerships with Meta's shopping agent Muse. Its accessible report also described QVC as welcoming agents, Kate Spade as allowing browsing but not buying, and Amazon as blocking agents. Those descriptions cover different kinds of access; a partnership announcement does not demonstrate that a complete purchase flow is live. [1]
Shopify, meanwhile, announced browser-agent tools for its checkout on 28 September. Amazon's seller-facing terms assert a right to restrict some automated access. Walmart and Google have described ways for an outside assistant to connect with a retailer-controlled shopping experience. These announcements are not a controlled comparison of retailer performance. They do reveal a strategic choice: how much of the discovery and transaction journey should be handed to an agent the retailer does not own? [2] [4] [7] [9]
<h2>Shopify is making its checkout accessible</h2>
Shopify's incentives differ from those of a retailer that owns a large discovery platform. A merchant on Shopify may benefit when a customer finds its products through an independent assistant, even if that assistant sits outside the merchant's own site. That is the strategic logic behind making the merchant's commerce infrastructure accessible in more places, not a measured claim that more agent traffic will produce more sales.
Shopify's 28 September developer announcement added WebMCP support for the checkout that is open in a buyer's browser. Its tools let a compatible browser agent read checkout state, update supported checkout fields and submit an order after the buyer confirms it. If a payment challenge or other buyer action is needed, the agent hands control back to the shopper. Shopify says the tools use the checkout page's existing state rather than a new merchant API. [2]
The distinction between <em>cart</em> and <em>checkout</em> matters. Shopify separately documents storefront and cart tools, but its checkout documentation says <code>update_checkout</code> ignores line-item changes. It also requires the agent to show the current order and total and get the buyer's permission before completing the order. Web Bot Auth applies to browser requests, and the technical page does not specify a universal eligibility or geographic rollout. The documentation establishes a route for agent participation, not proof that every store, browser agent or purchase already uses it. [3]
<h2>Amazon can restrict access while building its own agent</h2>
Amazon's amended Business Solutions Agreement and Agent Policy, effective 4 March 2026, set requirements for automated software or AI agents accessing Amazon services. Amazon says an agent must identify itself, comply with its policy and stop accessing the services if asked. It also says it may restrict agent access in certain circumstances. The seller announcement does not itself show that every outside shopping agent is blocked, nor does it describe every enforcement decision. [4]
That position has a commercial logic. Amazon controls search and product rankings, customer accounts, reviews, fulfilment and a substantial retail-media business. McKinsey's August analysis illustrates a possible risk for commerce media networks: if an assistant narrows a shopper's options before they browse a retailer's category or product page, some conventional digital-shelf advertising opportunities could be bypassed. McKinsey explicitly describes its opening example as a scenario, not a forecast or measured loss of advertising revenue. [5]
Amazon is not rejecting shopping assistance itself. In June, Amazon said its Rufus assistant had been used by more than 300 million customers <strong>in 2025</strong> and described combining Rufus with Alexa+ into Alexa for Shopping. It listed product comparison, price history, cart building and some automated shopping tasks, and positioned the assistant for shopping on Amazon and other stores. The usage figure belongs to Rufus in 2025, not to the new combined product. Amazon did not disclose a comparable usage or transaction figure for Alexa for Shopping in that announcement. [6]
The resulting asymmetry is an interpretation, not a finding about user behaviour: Amazon can encourage its own shopping assistant while setting terms for other agents on Amazon services. That could protect a valuable marketplace experience. It could also create friction if more shoppers begin their journeys in assistants Amazon does not control. Neither effect is quantified by the cited sources.
<h2>Walmart and Google describe a governed handoff</h2>
Walmart says it is building a multi-agent ecosystem in which customers retain choice and retailers maintain control of their experiences. In a January 2026 investor-conference transcript, a Walmart executive described a Gemini interaction: a product discovered through Google's assistant could be passed to Walmart's agent, making a familiar Walmart experience available within Gemini and adding the item to an existing basket. The same executive called the integration 'very early days'. It should be read as Walmart's account of an emerging flow, not evidence of general availability or incremental sales. [7] [8]
Google's Universal Commerce Protocol addresses a related handoff. At NRF in January, Google said it had built the protocol with Shopify, Etsy, Wayfair, Target and Walmart. Its stated design kept the retailer as merchant of record and the customer relationship with the retailer, including possible loyalty enrolment and personalised offers. Google described UCP as available at that point but said a buy button directly in AI Mode and Gemini would come 'soon'. The remarks did not establish that every partner, merchant or Google checkout surface was live in January. [9]
The common strategic idea is <strong>controlled interoperability</strong>: an outside assistant can help surface intent, while the retailer preserves a recognised commercial relationship, a governed transaction and the parts of the basket or service experience it can actually support. This is a proposed architecture and editorial interpretation. The two announcements do not demonstrate that Walmart's and Google's implementations are identical or already operate at scale.
<h2>Three bets, not three fixed camps</h2>
Shopify illustrates <strong>merchant-access infrastructure</strong>: expose supported storefront, cart and checkout capabilities so independent agents can reach merchants, with buyer consent at purchase. Amazon illustrates <strong>proprietary-agent control</strong>: develop its own assistant while retaining policy rights to condition outside access. Walmart and Google illustrate <strong>governed handoffs</strong>: connect assistants and retailer-controlled systems while preserving merchant-of-record and relationship boundaries. These are useful lenses on product strategy, not a classification of every live retailer implementation. [2] [3] [4] [6] [7] [9]
Controlled interoperability may be the strongest long-term design, but that is a judgement, not a proven winner. A shopping assistant is more useful if it can compare options across sellers. Retailers, in turn, have reasons to protect accurate pricing, loyalty, fulfilment, returns and their customer relationships. A permissioned handoff could serve both interests. Whether shoppers prefer it, whether merchants can implement it reliably, and how platforms set access rules remain open questions.
Amazon may also evolve towards approved, rules-based access while retaining substantial control over which agents qualify. That is a scenario, not a reported policy change or a prediction of when one will occur.
<h2>A new AEO question: can an agent act on the recommendation?</h2>
Imagine two hypothetical products. An assistant recommends Product A, but the retailer selling it does not expose current offers or a supported agent checkout. It recommends Product B second, and the retailer provides accessible product information and a buyer-confirmed handoff. The first recommendation is still a recommendation; the second purchasing path may be easier for that particular agent. Nothing in the cited material establishes that such access changes the assistant's independent ranking or which product the shopper ultimately buys.
The scenario adds a layer to the <a href="/articles/who-owns-ai-decision-layer-2030-scenarios">decision-layer framework</a> and the <a href="/articles/google-flipkart-ai-checkout-agentic-commerce-dtc">earlier Google and Flipkart checkout analysis</a>. <strong>Recommendation, agent access, transaction availability and human choice are different outcomes.</strong> A brand should not treat them as one visibility score or assume that a checkout integration buys better organic recommendations.
A useful agent-accessibility diagnostic asks whether an agent can retrieve accurate product information, inspect relevant variants and offers, complete or hand off a buyer-approved transaction, and preserve identity, loyalty and service where those matter. Each dimension should be tested for a defined agent, retailer, market and use case. Shopify's documentation supports some checkout mechanics; Google's and Walmart's announcements describe intended retailer continuity. None of the cited sources shows that greater accessibility increases independent recommendation frequency, completed purchases or incremental revenue. [3] [8] [9]
<h2>AEO Updates Takeaway</h2>
The AI shopper is not owned by one party across the whole journey. Assistants may shape discovery and recommendations; retailers can still set conditions for inventory access and preserve their role in checkout and service. The balance is being negotiated through product design, policies and commercial agreements, not settled by the mere existence of an agent.
For AEO practitioners, the practical step is to audit the handoff as carefully as the answer. Where does the assistant obtain product facts? What can it verify about current offers? When must the customer intervene? Who is merchant of record, and which retailer benefits survive? Record recommendation and purchasing results separately, then test whether changes in access actually affect choice. Access to the transaction may matter as much as visibility in the answer, but its effect has not yet been demonstrated by these announcements.
<h3>References</h3>
[1] <em>The Wall Street Journal</em>, <a href="https://www.wsj.com/cmo-today/ai-agents-aim-to-change-shopping-some-retailers-are-locking-the-doors-666a1ed4" target="_blank" rel="noopener noreferrer">AI agents aim to change shopping; some retailers are locking the doors</a>, 2 October 2026. Accessible opening and headline reviewed; remainder subscriber-restricted.
[2] Shopify Developer Changelog, <a href="https://shopify.dev/changelog/posts/webmcp-support-for-checkout" target="_blank" rel="noopener noreferrer">WebMCP support for checkout</a>, 28 September 2026.
[3] Shopify Developers, <a href="https://shopify.dev/docs/agents/carts-and-checkout/checkout-webmcp" target="_blank" rel="noopener noreferrer">Checkout WebMCP</a>, accessed 4 October 2026.
[4] Amazon Seller Forums, <a href="https://sellercentral.amazon.com/seller-forums/discussions/t/84e3f6b1-42f7-4cf3-a189-a5cc8d78d838" target="_blank" rel="noopener noreferrer">Business Solutions Agreement and Agent Policy update</a>, effective 4 March 2026.
[5] McKinsey & Company, <a href="https://www.mckinsey.com/capabilities/growth-marketing-and-sales/our-insights/next-in-growth/rewiring-advertising-the-ai-imperative-for-commerce-media-networks" target="_blank" rel="noopener noreferrer">Rewiring advertising: the AI imperative for commerce media networks</a>, 5 August 2026.
[6] Amazon Ads, <a href="https://advertising.amazon.com/library/news/agentic-shopping-advertising" target="_blank" rel="noopener noreferrer">Agentic shopping and advertising</a>, 11 June 2026.
[7] Walmart, <a href="https://corporate.walmart.com/about/technology" target="_blank" rel="noopener noreferrer">Technology: agentic commerce future</a>, accessed 4 October 2026.
[8] Walmart, <a href="https://corporate.walmart.com/content/dam/corporate/documents/newsroom/events/2026-icr-conference/ICR-Conference-2026-Transcript.pdf" target="_blank" rel="noopener noreferrer">ICR Conference corrected transcript</a>, 13 January 2026.
[9] Google, <a href="https://blog.google/company-news/inside-google/message-ceo/nrf-2026-remarks/" target="_blank" rel="noopener noreferrer">NRF 2026 remarks on the Universal Commerce Protocol</a>, 11 January 2026.