Google AP2 and UCP Market Progress

This is a follow-up to this week’s Google’s Big Day: Gemini Tasks, AP2 and Wallet on iOS. Several readers asked the obvious next question: how does the announcement impact Agent Payments Protocol (AP2) and the Universal Commerce Protocol (UCP)? What of my previous coverage has changed? The biggest questions are commercial. Who owns checkout? Who sees intent? Who takes the risk, and who authorizes the agent? Today I’ll attempt to look for evidence of actual implementations, then answer those questions.

My position has not changed. Google’s engineering is the best in agentic commerce. Its partnership record is the weakest of the large platforms. Standards are how Google creates control and solves problems inside its own domain. A standard cannot solve monetization, risk, or authentication/ authorization for the benefit of a relying party. Those are commercial problems. Note: Thanks to Claude Opus for helping me find relevant references to fact-check today’s post.

Executive Summary

  • UCP is broad with many elements: discovery, cart, checkout, payment and order management. Merchants can adopt most of it and still keep checkout.
  • Within checkout there are many paths: a merchant-owned checkout, UCP Native checkout inside Google’s interface, and AP2-authorized agent purchases are separate things. Counting one as evidence of another overstates progress.
  • UCP checkout adoption is real, and mostly Shopify. An independent test found 88.3% of Shopify stores return a checkout link to agents, against 1.4% of other stores. Shopify switched eligible merchants to Native checkout by default in September. The large retailers are keeping checkout: Walmart and Target have both told me they will own it, to control their own cards on file, and will require the consumer to log in (for EULA/terms (your bot is your responsibility).
  • AP2 moved from specification to a Google surface. As of mid-October, no independent review could find a production AP2 native checkout transaction. On November 17 Google said Gemini “now uses” AP2 on Visa and Mastercard tokens. No issuer, merchant or volume has been named. Issuers have a say in how their cards are tokenized. My read is that this is a Google commitment to abide by network agentic governance (ie rules) to obtain agentic tokens. These include agent registration, originating device/account data and intent.
  • Consumer permissioning is split. Google announced the Wallets new role “secure hub for managing agentic purchases across Google”. Today, spending caps, confirmations and “Always allow” live in Gemini. Cards, receipts and order tracking live in Google Wallet. Google Wallet on iOS does not yet show agent permissions. Wallet will likely evolve as identity credentials move toward wallet (for example, DPCs, a topic for another blog).
  • What a standard will not fix. Merchants still own checkout and fraud loss. Issuers still decide whether to accept an agent’s token. The Gordian Knot of agentic monetization remains unresolved as there is no price paid by merchants for agentic demand. Intent data still sits with the platform (but this may be giving way).

1. UCP Is Broader Than Checkout (101 stuff)

Obviously UCP is more than a checkout button. It is a protocol for the full commerce lifecycle. The specification spans search and lookup, cart building, identity linking, checkout and order management as core capabilities (Universal Commerce Protocol).

StageWhat UCP coversWho holds control
1. Discovery and registrationA business publishes a profile at `/.well-known/ucp` declaring its capabilities, versions and payment handlers. The agent negotiates the capabilities both sides support, then searches the catalog for products, pricing and availabilityMerchant publishes; platform reads
2. Shopping and cartProduct selection, cart creation and updates, discount codesPlatform today: Universal Cart holds the cart until the shopper transfers it
3. CheckoutNative or Embedded. Identity linking (OAuth 2.0) lets the shopper log in to the merchant accountMerchant’s choice; Shopify’s default is Native
4. PaymentPayment instrument selection through payment handlers, credential exchange with the processor, and AP2 mandates as proof of consent. The merchant returns the instruments it will acceptMerchant and its processor; Google Pay supplies the card in Native
5. Order managementOrder confirmation, tracking and fulfillment updates, with return and refund policies exposed to the agentMerchant fulfills; Google Wallet displays

Participation is broad. Google co-developed UCP with Shopify, Etsy, Wayfair, Target and Walmart, with more than 20 endorsers at launch, including Adyen, American Express, Best Buy, Mastercard, Stripe and Visa; Amazon, Meta, Microsoft, Salesforce and Stripe have since joined the UCP Tech Council. The protocol now reaches beyond retail: ucp.dev lists Amadeus, Booking.com, Expedia, Hilton and Marriott for hotel booking, and DoorDash, Square, Toast and Uber Eats for food delivery.

The breadth matters for the rest of this post. My focus is on Checkout because it relates to payments, risk and is the one merchants guard most.

2. Three Paths, Three Milestones

Where UCP is the commerce protocol. AP2 is the authorization land orchestration layer: signed mandates recording what the consumer allowed the agent to do. The piece of UCP that separates the shopping- and discovery-related aspects of commerce from the payment-related aspects. Today I see three paths to a purchase:

PathWho runs checkoutPayment credentialStatus
UCP Embedded or cart transferMerchantMerchant’s choice (COF if possible)Live; Universal Cart transfer (May 2026)
UCP Native checkoutGoogle Google Pay and Chrome Autofill Credentials not under Network FrameworkLive for selected merchants in the US
Native under Network FrameworkGoogleNetwork tokens on “supported cards”Announced Nov 17; no named issuer or merchant

Google’s own documentation draws the line between the second and third paths. In Native checkout today, “the user now interacts only with the Google UI to fill in sensitive fulfillment and payment details and submit the order. The Agent is not involved in this part, ensuring determinism” (Google for Developers, UCP Native checkout guide). The same page says “future plans” will “support more agentic experiences.” Native checkout is a Google Pay button inside an AI answer. It is not an agent buying on your behalf. But merchants still own the risk of this checkout, thus need additional data from Google to manage it.

Google’s Merchant Center help says UCP checkout “uses standard Funding Primary Account Numbers (FPANs) that users have stored on their Google Wallet” (Google Merchant Center Help). Note this page may not have been updated size this week’s launch, but it is what merchants are evaluating. I would characterize it as an agent autofill of a merchant checkout page.

3. Where UCP Stands

The retailer list is real, but it is a list of logos. On May 20 Google said shoppers could “try these select checkout features soon across retailers like Nike, Sephora, Target, Ulta Beauty, Walmart, Wayfair and Shopify merchants such as Fenty and Steve Madden” (Google, May 20, 2026). The same paragraph offers two paths: Google Pay or “transfer items to the merchant’s site.” Google has never said which retailer uses which.

Independent tests point the same way. Brandon Lazovic checked the standard `/.well-known/ucp` profile for eight retailers Google named. Fenty Beauty and Steve Madden, both on Shopify, published profiles. Nike, Target, Walmart, Wayfair, Ulta Beauty and Quince did not (Brandon Lazovic, reported by PPC Land, Oct 2026). A missing profile does not prove a retailer has no integration, since Google can also onboard merchants through Merchant Center. It does show that the large retailers have not publicly declared native checkout.

ARC runs an open benchmark of whether agents can reach checkout. Its November 19 snapshot covers 11,726 completed store checks. 88.3% of Shopify stores (7,163 of 8,110) returned a checkout link to an agent. 1.4% of other stores (50 of 3,616) did (ARC open dataset, Nov 19, 2026). ARC is careful to add that “a checkout link is not a purchase.” It measures reach, not transactions. Also note that the checkout link is the merchant’s checkout page that could be fed from a UCP “cart”.

Shopify adoption is a default, not a decision. On September 22 Shopify announced that “Eligible products are automatically included” in native checkout across AI Mode and Gemini. with and “opt out” (Search Engine Roundtable). Shopify applies the same opt-out logic to its own agentic storefronts. Stores on its separate Agentic plan, the brands not on Shopify, start ShopPay as the default payment (The Defiant, “Shopify Takes Shop Pay Into AI Agents and Off-Platform Checkouts,” Oct 2026). The long tail is on Native, with ShopPay, because the default put it there. There is obviously some conflict in the payment credential store between ShopPay and Google Wallet, but if Shopify is touching the merchant, it is ShopPay that is used. Which then makes me question what is left for wallet? Receipts? Credentials? Gemini only permissions?

Universal Cart keeps the intent. I wrote about Universal Cart in June (Universal Carts and Mandates) as a single cart across Search, Gemini, YouTube and Gmail. The October rollout to US shoppers covers Search and Gemini only. The Cart API sends the merchant nothing while the shopper saves items. Google calls the merchant once, with every item, when the shopper presses transfer. Updates and syncing are not supported in this release (Google UCP Cart API documentation, Oct 2026). A product saved on Monday and abandoned on Tuesday never reaches the retailer. Abandoned carts are the most valuable intent signal in eCommerce, and in Universal Cart, Google keeps them.

My read. Merchants want agentic discovery. The large ones are keeping checkout. Walmart and Target have both told me directly that they will own checkout, to control their own cards on file, and will require consumers to log in before they check out. Large retailers require log in before checkout today, and that paradigm is carrying into agentic. UCP accommodates it. Identity linking lets the shopper log in to the retailer account from inside the agent, and Google’s Merchant Center hub offers it as one of three choices alongside cart transfer and native checkout. Walmart’s own agent design works the same way, with agent profiles and spending limits inside the customer’s Walmart account (Walmart Agent Gateway).

This is the pattern I described when OpenAI’s Instant Checkout folded: Walmart saw conversion inside ChatGPT at roughly a third of what the same visitor delivered on walmart.com (Agentic Reality: Advertising). Bain found that consumers trust retailer-owned agents three times more than third-party agents to complete a purchase, and told retailers not to become “a ‘dumb fulfillment pipe'”. Today’s PR from Target also demonstrates this point.

All of this is consistent with my previous blogs covering embedded checkout and why earlier efforts of AI platforms to own checkout and the economics failed (see UCP Enables a New Economy, Explaining the Death of OpenAI’s Instant Checkout).

4. Where AP2 Stands

Technically speaking, AP2 is an elegant, well-designed standard and is not making progress across standards and partners. But it is broad, and which elements of it are supported by which partners remains unclear. Frequent readers know I have characterized Verifiable Intent as a self attestation. This cryptographic chain of custody means nothing to a relying party that wants to validate what the consumer permissioned the agent to do, and compare it to what the agent actually is proposing to do. This is more than intent, but also about whose credentials are signing it, how their credentials are stored and how the consumer signs the transaction. This is what “governance” is, and a platform attempting to monetize the transaction can’t be given all control over how this works.

MilestoneStatus
AP2 launched with 60+ partnersSep 2025
v0.2 with human-not-present (autonomous) mandatesApr 28, 2026
AP2 donated to the FIDO Alliance; Mastercard contributes Verifiable IntentApr 28, 2026
A2A x402 extension for stablecoin payments between agentsSep 2025
Independent review finds no confirmed production AP2 transactionOct 18, 2026 (Protofire)
Google says Gemini “now uses” AP2 on Visa and Mastercard tokensNov 17, 2026
Named issuer validating AP2 mandates in productionNot found
Named merchant or published AP2 volumeNot found

The FIDO Alliance describes the split well. AP2 is “the mandate and coordination layer” and Verifiable Intent is “the evidence layer” (Nishant Kaushik, FIDO Alliance, “Building the Trust Layer for Agentic Payments with AP2 and Verifiable Intent,” May 2026). I covered why FIDO was the right home in AP2 Donation to FIDO.

Protofire wrote (Oct 2026) that AP2 “is a published protocol with no confirmed production deployment” and that claims it already runs inside Gemini should be treated as unverified “until the operator says so”. A month later the Google confirmed: : “Gemini now uses the Agent Payments Protocol (AP2) to make purchases like this. We’re working with American Express (coming soon), Mastercard, and Visa to keep your purchases secure with digital tokens on supported cards”. That is progress, and the network token choice is the right one. Network agentic tokens bring network rules: agent registration, consumer authentication, disputes, liability allocation, data sharing and issuer authorization. That is the commercial construct AP2 lacked on its own (Agentic Governance: “Standards” Are Worthless Without a Commercial Construct).

Look at where AP2 went live. The first production AP2 surface is one where Google controls every party except the issuer. Universal Cart, which Google previewed at I/O alongside AP2, launched in May with no mention of AP2. AP2 runs where Google needs no partner’s permission. I made the same point about the three-mandate design in Carts and Mandates: when the platform signs, holds and attests the mandate, it is grading its own homework.

The engineering test for AP2 is a documented purchase in which a mandate is generated, independently verified and used to authorize payment. The commercial test is harder: which issuer relied on the mandate, under which rules, and who paid when it was wrong. This is why I deride Verifiable Intent and the numerous MA PR announcements. They were lab efforts to demonstrate “it works”. One Mastercard EU Issuer told me last week “Tom Verifiable Intent is completely unworkable, what on earth is MA thinking of in supporting this”? My answer? They want a big partner that drives VAS and will do anything to gain a lead here.

5. What the Launch Changes for Consumers

Google’s description of the consumer controls is specific:

  • Spending cap. “You can also set a hard spending cap upfront, and Gemini is designed to stay within your budget.”
  • Confirmation. Gemini “is designed to ask for your confirmation or pre-approval before clicking ‘Buy.'”
  • Standing permission. “For routine stuff you already trust, like adding those Cheerios to your grocery cart, you can set that to ‘Always allow.'”
  • Receipts. “All of your receipts and tracking info land straight in Google Wallet.”

Google Wallet on iOS lets consumers “stay in control of agentic purchases made across Google” by viewing “order confirmations, receipts, and delivery updates” (Google, Nov 17, 2026). That is a record of what happened. The controls that decide what may happen live in Gemini. Logins live in Password Manager. But all of this is likely to change (in my view) depending on where credentials are stored and how consumers authorize agents. In a 2 yr view I certainly see Google pushing for DPCs here, and way that will work across all platofrms and surfaces.

There is a disconnect that may just be e delay in updating public facing documentation, relating to the permission experience across three Google products. In the prior post I argued that Wallet on iOS is the natural credential store and agent control panel. The capability is not there yet. A consumer who wants to know what Gemini may buy, with which card, up to what limit, opens Gemini settings, not Wallet.

The agent control “gateway” is a very active space. Banks, merchants, payment providers (ex Link, Shopify) and Issuers are all building this agent control dashboard. For example, Shopify lets a shopper authorize an agent to place orders within a weekly, monthly or total budget and revoke it. Walmart puts agent profiles and spending limits inside the customer’s Walmart account (Walmart Agent Gateway). Stripe made Link balance the agent control dashboard (Stripe Link Balance).

Will Google’s wallet only control Gemini permissions? Google has the consumer reach to make Wallet the cross-agent permission layer. I can’t imagine other AI platforms (or merchants) buying into a model where Google wallet sees everything across all platforms, that would create the same virtuous cycle and centralization effect that Google Analytics has today.

6. What It Changes for UCP and AP2

UCP becomes Gemini’s merchant interface, eventually. Google says, “Up next, we’ll integrate the Universal Commerce Protocol (UCP) into Gemini.” Today Gemini Tasks completes purchases in other ways. It hands groceries off to Instacart. On some sites, it navigates the merchant’s web checkout and “can make it most of the way through and then hand the screen back to you to click the final button” (Google, Nov 17, 2026). That is an agent using the merchant’s checkout without the merchant’s integration. UCP is the cleaner path, and merchants will decide whether it arrives as Native or Embedded.

AP2 gets a reference implementation. Until this week AP2 was a specification with partner logos. Now it has a consumer product, a token path on two networks and a spending cap that works like an open mandate. Engineers and standards bodies can now point to a working deployment. Issuers can now ask concrete questions: what does the mandate contain, and does it reach the authorization message? What additional Data will I receive from Google to authorize? What rules must Google abide by to use an agent token?

The partnership question is unchanged. Google’s history here is long. Shopping Actions launched in 2018 with Target, Walmart, Ulta, Costco and Home Depot, became Buy on Google, dropped its commission in 2020, and closed in the US in 2023. Google’s explanation: “Buy on Google was a small feature that a very limited number of merchants used”. Buy for me was the I/O 2025 headline and then went quiet (Google Pulls Back From “Buy For Me”). Each time the technology worked. Each time Google needed merchants to give up something they valued, and allow them to act a the new trusted intermediary. Merchants are not exactly thrilled with being disintermediated, with a new partner set on owning the standard, the consumer and the economics. Imagine that!

7. What a Standard Will Not Fix

Monetization. Crickets…. nothing here has changed it is still the biggest problem.

Risk. In the US, merchants own the risk, and there is no 3DS protocol adoption. Issuers are not thrilled with the growth of Network VAS, and are attempting to build their own in Paze. The merchants just laugh at Issuer efforts here; obviously, the only way to gain traction is through existing network mechanics and distribution. For example, a top 5 merchant told me last month, “Tom, we need the Issuers to resolve the current issues in Disputes, chargebacks and refunds before we will ever entertain an expansion of their role in anything”.

A merchant will not accept a checkout it cannot score with its own device and behavioral data. Credential choice has a direct cost too. Replacing a buyer’s regulated debit card with a commercial virtual card raises interchange by about 242 basis points, and the merchant pays it (Jesse Chang, FT Partners, “Muse Be the Money: Three Rails, Different Economics,” Sep 2026). UCP and network tokens avoid that. They do not answer who pays when the agent buys the wrong thing. A UBS panel put the answer at “the source of failure”: the customer, the agent developer, the issuer or the merchant (Timothy Chiodo, UBS, “Identity, Trust, and Fraud in Agentic Commerce,” Oct 2026). Network rules can settle that. A protocol cannot.

Authentication and authorization by the relying party. The party that takes the risk must have a lead role in authenticating the consumer and how the agent is authorized. A signed mandate tells the issuer the platform recorded consent. It does not let the issuer compare what the consumer asked for with what the agent bought. As one EU issuer told me last week, “We must be able to compare what the consumer permissioned and what the agent did” (Google’s Big Day). This is why I argued the bank should hold intent (Why Banks Are Best Positioned in Intent). Europe has a legal version of the same gap. David Birch notes that under EU law “mandates flow between legal persons,” and “nobody has formally decided whether a credential issued to an agent amounts to a legally enforceable authorisation” (David Birch, “Trust Me, I’m an Agent: Europe’s Next Infrastructure,” Oct 2026).

Intent data. Universal Cart shows Google’s default: the platform keeps the saves, the comparisons and the abandoned carts, and the merchant gets the transfer. In AP2, the platform signs the open mandate. The data the issuer needs and the data the merchant wants both start at Google. I have written more than ten posts on this (Agentic Data Battle: Intent, EMVCo Enters the Intent Wars). A standard defines the format of intent. It does not decide who receives it.

What to Watch

  • Named issuers. Which issuers support Gemini’s agentic tokens, and their approval and decline rates on agent purchases.
  • Mandate in the authorization. Whether AP2 or Verifiable Intent data reaches the issuer’s authorization message, under which network program.
  • UCP in Gemini. Whether Gemini’s merchant integration offers embedded checkout and the merchant’s own card on file, or Native only.
  • Retailer log in. Whether Gemini supports UCP identity linking, so a shopper logs in to the retailer and pays with the retailer’s card on file.
  • Native opt-outs. Whether Shopify merchants switch off default Native checkout, and whether any large retailer reports Native conversion.
  • Wallet permissions. Whether Google Wallet adds per-agent limits and revocation, and whether it ever permissions an agent Google does not own.
  • Pricing. Lead fees, ads inside Gemini Tasks, or transaction fees on qualified orders.

Wrap Up

UCP and AP2 are fantastic engineering. Network tokens put AP2 on rails with rules, and Gemini Tasks gives it a real product. UCP reaches thousands of Shopify stores.

Be skeptical of PR here and read the evidence of usage closely. My current assessment is that the pattern from my various posts remains. Adoption is strongest where Google sets the default: Shopify’s opt-out Native checkout, AP2 inside Google’s own agent and wallet. Where Google needs a partner’s decision, the large retailers are quiet, issuers are unnamed and pricing is undefined. Standards extend Google’s control inside its own domain. Monetization, risk and relying-party authorization sit outside it, with merchants, issuers and networks. That is where the next year of this market will be decided.

Please Login to Comment.