The hype machine behind agentic commerce is at its apex. Every week brings another protocol announcement, another pilot, another prediction that autonomous agents will soon do our shopping for us. For payment executives trying to separate signal from noise, here is the signal: the first real agentic business model is not autonomous checkout. It is advertising.
What Is Actually Working: AI Inside the Retailer
AI is making great strides within the retailer’s own domain. Amazon’s Rufus, Walmart’s Sparky, and Michaels’ “Mike” are conversational agents built by retailers, trained on their own assortment and category expertise, driving basket sizes up over 35% in the best implementations. Amazon told investors that Rufus helped generate nearly $12 billion in incremental annualized sales, with monthly active users up 115% and users 60% more likely to complete a purchase (Bain & Company, “Agentic AI in Retail: How Autonomous Shopping Is Redefining the Customer Journey,” May 2026). Michaels took “Ask Mike” from concept to production in six weeks on Google Cloud, as I covered in Google Pulls Back From “Buy For Me”.
Note what these successes have in common: the retailer owns the agent, the data, and the checkout. When I surveyed top merchants, as related in Retailer Actions in Agentic Commerce, the answer was consistent and blunt. One CMO quipped “There is no way we will enable a new Google to disintermediate us, we want to use our data for our own benefit.” Retailers are not thrilled with the idea of handing over their data or their customers to AI platforms, and they are firmly in control of checkout. Bain’s research confirms consumers agree: shoppers trust retailer owned agents three times more than third party agents to complete a transaction.
The Three Problems Holding Agentic Back
As I have written since Agentic Commerce Economics and Governance, agentic commerce has three major problems:
- Lack of retailer support. Retailers control the checkout, the fraud loss, the returns, and the customer relationship. No agentic platform scales without them, and they are not volunteering to be disintermediated.
- Lack of governance. Who pays when things go wrong? I have blogged on this at length (see Target’s consumer terms, “your bot is your responsibility”). The liability question remains unresolved for nearly every platform and network.
- The economic model. What construct encourages a virtuous cycle where platforms, retailers, networks, and consumers all gain? This last point is the focus of today’s blog.
Remember the first rule of payment economics: merchants pay for everything. Any agentic model that adds a new toll without delivering measurable value to the merchant will fail. Which brings us to the one model where merchants are voluntarily writing checks today: advertising.
OpenAI Goes All In on Advertising
An excellent Lengow analysis, ChatGPT Ads: first results are in, lays out how OpenAI arrived here, and the first results.
OpenAI tried to own the transaction. Instant Checkout launched in September 2025 with Shopify, Etsy, and Walmart, promising payment completion without leaving the conversation (see my take in Walmart and Open AI). The results were grim. Walmart reported that conversion for payments made inside ChatGPT ran at roughly a third of what the same visitor delivered when sent to walmart.com. OpenAI pulled the plug in early March 2026, rebranding the retreat as a “shift to Apps.” The conversation stays in ChatGPT, the transaction goes back to the merchant’s own site, and OpenAI earns per click rather than per sale. In other words, a search engine.
From there the ad machine assembled quickly: an ad pilot on February 9, 2026, a self serve Ads Manager with click based bidding on May 5, automated product feeds (up to one million SKUs per advertiser) on June 2, and a UK launch on June 6 with Zalando, Tesco, Hilton, and Hiscox. Advertising is no longer a side feature at OpenAI. It is the main economic engine.
And the early data validates the channel. Per Adobe Analytics, shoppers who reached retail sites via an AI chatbot during Prime Day 2026 were 40% more likely to complete a purchase than those coming from search, email, or social (cited in the Lengow analysis). AI referred traffic is small (under 1% of total traffic for most retailers per Bain) but it converts.
More Than One Strategy: Visibility vs. Full Stack
Not everyone wants the click to buy ad. Some just want the visibility.
Amazon is the paradox. It has been blocking ChatGPT’s crawlers for months, sued Perplexity for shopping on behalf of its users, and holds the lowest share of agentic traffic in the industry (around 0.4% per J.P. Morgan). Yet Amazon bought sponsored placements in ChatGPT to promote Prime Day. Every sponsored click redirects to Amazon.com, never to a payment completed inside the conversation. Amazon buys the click, never the transaction. It pays to be seen and refuses to be read, keeping control of the transaction, the customer data, and the relationship.
Target is playing the full stack. Testing with its ad arm Roundel since February 9, Target claims a 40% increase in ChatGPT driven traffic to its site every month, has rolled out a dedicated app inside ChatGPT, and partnered with Google on the Universal Commerce Protocol (see UCP Enables a New Economy). Organic, agentic, and paid all move together.
Meanwhile, the poster child for autonomous purchasing has quietly exited the stage. Google has dropped “Buy for me,” as I detailed in Google Pulls Back From “Buy For Me”. No case studies, no conversion disclosures, no merchant testimonials. The feature that took the wheel away from the shopper is the feature nobody talks about anymore.
Desperation Makes OpenAI a Great Partner
Here is the strategic irony payment executives should note: OpenAI’s desperation for a business model has made them much more flexible and aggressive in working with retailers and payment networks. They are the platform where others want to partner.
A great example this week is Resy’s partnership with OpenAI to launch Resy Reservations in ChatGPT. ChatGPT users can now discover and book US Resy restaurants without leaving the conversation, with American Express (Resy’s owner) sitting behind the experience. Note that Amex is the only network with a live and operable agentic protocol in ACE, as I covered in American Express Breaks the Agentic Commerce Deadlock. As a three party network, Amex controls the credential, the message, and the participant, so it can move alone. Visa and Mastercard both have solid alternatives, but their four party models require both issuers and platforms to accept the solution before anything operates at scale. That is a much longer road.
Google, on the other hand, seems to be building a closed system where partners take it or leave it. As I described in Agentic, Intent and the New “Data Games”, Google has no desire to share consumer intent mandates, and securing even basic intent data from Google is a battle. Its protocols arrive with a house wallet first and negotiability later.
The business unit inside Google that is making real progress is Google Cloud, enabling retailers to build their own AI. Ask Mike is the model: Google supplies the infrastructure and the speed, the retailer supplies the judgment, the data, and the checkout. This supports a world of multiple proprietary “specialist agents” rather than one omniscient assistant, a structure I have long argued is the realistic end state.
Wrap Up: The Bridge Between Engagement and Fulfillment
The business model of agentic commerce is evolving. Advertising is where it landed first because advertising is the one construct where the merchant keeps checkout, keeps the data, keeps the customer, and pays only for measurable demand. It will not be advertising forever. But whatever comes next must bridge the domain where customers engage (today that is ChatGPT and, increasingly, retailer agents) with the domain that can fulfill (the retailer’s site, inventory, and payment rails).
Right now the key enablers of that bridge are OpenAI and Google Cloud: one aggregating consumer engagement and selling access to it, the other selling retailers the picks and shovels to build agents of their own. For payment executives, the takeaway is patience. The transaction has not moved. The checkout still belongs to the merchant, the liability question is still open, and as I argued in Commercial Models for AI Agents, the networks remain the best positioned entities to build the value exchange layer when the economics finally get negotiated. Until then, watch the ad dollars. They are the most honest signal in agentic commerce.
Feedback appreciated.