MediaPost first reported the August 21 release alongside Microsoft’s broader push to deploy AI Max across Advertising accounts and reduce the use of manual cost-per-click limits. Microsoft’s own playbook confirms the three-part approach: make products discoverable, make transactions possible in the conversation, and measure how AI systems surface the brand. It also presents a 90-day activation plan rather than a new standalone Copilot capability.
The important caveat is that this is not a switch that makes any retailer immediately purchasable through Copilot. Microsoft’s current agentic-commerce documentation says Copilot Checkout eligibility is limited to English-language merchants selling to U.S. buyers in U.S. dollars, subject to Microsoft privacy, responsible-AI, and payment-service-provider requirements. The playbook is useful as a preparation document, but it should not be read as proof that agent checkout is broadly available to every Microsoft Merchant Center account.
Product feeds become operational data
For years, retailers treated product feeds as a channel-specific marketing input: submit a title, price, image, availability status, and product identifier to a shopping engine, then fix errors when ads stop serving. Microsoft’s playbook makes clear that an agentic shopping system needs a much higher-confidence version of that data.
An AI agent deciding whether to recommend a product must be able to determine whether an item is in stock, where it can ship, what it costs, whether it carries warnings or restrictions, whether it can be returned, and whether checkout can be completed. Missing or stale details do not merely produce a less attractive listing. They can prevent a product from being selected at all.
Microsoft frames this as “getting discovered,” but the underlying work is data governance. Retailers need reliable feeds and APIs connecting the commerce platform, inventory system, fulfillment rules, customer-service policies, and payment provider. A beautifully written product page cannot compensate for an incorrect inventory count or an absent returns policy when an automated system is trying to compare merchants programmatically.
That has a direct implication for enterprise IT teams. Ownership of product data can no longer sit entirely with paid-search managers or an ecommerce merchandising group. The people operating Microsoft Merchant Center may depend on upstream systems maintained by ERP, PIM, ecommerce, identity, security, and payments teams. If data arrives late, differs across channels, or cannot be audited back to a source system, the organization is building an unreliable foundation for automated recommendations and transactions.
Microsoft’s guidance also raises a less visible problem: its systems use both merchant-provided feeds and information found on the wider web to surface products organically. That means a merchant could have a technically valid feed while still presenting contradictory details elsewhere on its site. Product specifications, policy pages, structured markup, and support content now need the same change-control discipline as the feed itself.
UCP support is real, but the wording matters
The MediaPost report describes Copilot as operating on the same “UPC” used by major partners. The protocol’s name is Universal Commerce Protocol, or UCP, not UPC, and Microsoft’s own documentation uses UCP consistently.
More importantly, “Copilot runs on UCP” compresses several separate pieces of infrastructure into one sentence. Google introduced UCP in January with Shopify, Etsy, Wayfair, Target, and Walmart among its collaborators and supporters. Microsoft has since positioned Microsoft Merchant Center as a route for merchants to prepare UCP-ready feeds and expose richer commerce signals, while payment firms such as Stripe and PayPal offer their own onboarding and transaction paths.
Microsoft says its Merchant Center will support UCP, including richer data around returns and customer-support policies. That language matters. It signals an implementation path and a direction of travel, but it is not the same as declaring every Merchant Center feed, every Copilot surface, and every merchant checkout live under the protocol today.
The relevant technical distinction is straightforward:
- Product feeds help an agent identify and assess an offer.
- UCP-style data and transaction flows are meant to make an offer actionable through discovery, checkout, and eventually post-purchase actions.
- A payment and checkout provider still has to validate the cart, handle authorization, protect credentials, detect fraud, and pass an order to the merchant of record.
Microsoft’s agentic-commerce page states that the merchant, not Copilot, remains the merchant of record. That preserves a retailer’s responsibility for taxes, fulfillment, returns, customer support, and consumer-law compliance. It also means a Copilot transaction does not eliminate the retailer’s existing operational stack; it puts a new AI-mediated entry point in front of it.
For developers, the immediate assignment is to inventory the fields and services required to support that entry point. A workable implementation needs more than a catalog export. It needs stable product IDs, live stock status, clear offer eligibility, validated pricing, shipping and returns data, order-state handling, and payment-provider integration. Retailers should also determine where an agent is permitted to act, what consent it must obtain, and how an agent-originated order is distinguished in support and fraud workflows.
The 90-day plan is an onboarding sequence, not a market forecast
Microsoft divides the playbook into two 45-day phases. The first focuses on foundations: preparing feeds and establishing a friction-free checkout path. The second focuses on optimizing content and performance based on what the business learns.
That sequence is sensible, but it exposes where many businesses will struggle. There is little value in measuring “AI visibility” or testing AI-native ads if the underlying item records are incomplete, product availability cannot be refreshed quickly, or checkout exceptions need manual intervention. In traditional paid search, a weak landing page costs conversion rate. In an agentic transaction flow, a weak fulfillment or policy signal can remove the merchant from consideration earlier in the process.
Microsoft also points merchants to Clarity’s AI Visibility capabilities as a way to see which AI platforms cite a brand and how AI-referred visitors behave. For organizations already using Microsoft Clarity, that can produce a new reporting category, but it should not be mistaken for a definitive attribution system. A citation, referral, recommendation, and completed order are different events. Teams will need to reconcile Clarity reporting with Merchant Center, Microsoft Advertising, web analytics, ecommerce orders, and payment-provider records before assigning revenue to an agentic channel.
The playbook cites Microsoft research indicating that consumers expect AI-assisted shopping and Adobe Analytics data indicating stronger conversion from AI-referred visitors. Those figures are directional vendor evidence, not a guarantee that an individual retailer will see the same results. The first useful metric is not gross AI referral traffic; it is whether agent-originated sessions produce profitable orders without higher fraud, support, refund, or fulfillment costs.
AI Max expands automation while Microsoft removes a guardrail
The playbook arrived as Microsoft is also broadening AI Max for Search campaigns. Search Engine Journal reported on August 19 that Microsoft had begun rolling AI Max out globally after its earlier pilot. MediaPost described the wider account rollout on August 20. The two reports differ on the precise rollout timing, but agree on the operational result: AI Max is moving from a test feature toward routine Microsoft Advertising infrastructure.
AI Max expands matching beyond advertisers’ keyword lists, generates or adapts ad text from existing assets and website content, and can route visitors to a page Microsoft judges more relevant than the campaign’s static final URL. For advertisers importing eligible Google Ads search campaigns, Microsoft says the associated AI Max settings can carry into the equivalent Microsoft campaigns.
This will save campaign-management time for organizations with well-maintained sites and conversion tracking. It also increases the consequences of poor site hygiene. If URL expansion can select alternative pages, teams need to review which pages are eligible, whether those pages have current product and legal information, and whether campaign exclusions prevent traffic from reaching obsolete, region-inappropriate, or low-conversion content.
Microsoft is pairing that expansion with a significant bidding change. Beginning October 1, new non-portfolio campaigns using Target CPA, Target ROAS, Maximize Conversions, Maximize Conversion Value, or Maximize Clicks will no longer offer a Max CPC control. Existing campaigns that already use Max CPC will retain it, as will portfolio bid strategies; Target Impression Share and Enhanced CPC are also outside this change.
Microsoft’s stated rationale is that a Max CPC ceiling can conflict with conversion-based automation. That is technically credible: a platform cannot freely bid for auctions it predicts will meet a cost-per-acquisition or return-on-ad-spend target if an independent click-price cap blocks those bids. But advertisers should recognize the tradeoff. They are being asked to place more trust in conversion tracking, goal configuration, attribution windows, and Microsoft’s auction-time model.
The practical response is to review conversion quality before October 1, not after. Campaigns using imported goals, offline conversion uploads, duplicate events, weak lead-quality signals, or delayed revenue data are poor candidates for more autonomous bidding. A Max CPC cap may have constrained performance, but it may also have contained spending while those measurement problems remained hidden.
Microsoft’s new playbook therefore lands as part of a broader shift: product data, web content, checkout integration, analytics, and bidding controls are converging into a single AI-directed commerce workflow. Retailers can start preparing now, but the near-term work is unambiguously operational—clean the data, verify the transaction path, protect the measurement layer, and test automation with clear controls before making agentic commerce a production revenue channel.