May 18, 2026

Catalogs Become Ads, Voice Becomes Checkout, BNPL Joins the Agent

May 12-18, 2026. The week the product feed became the ad creative, the search bar became the agent, and buy-now-pay-later became standard equipment in AI checkout.

Last week we covered Shopify’s Q1 numbers, Amazon’s 40-person agentic team, and Europe’s AI Act haircut. This week the operating layer caught up with the slide deck. On May 12, OpenAI shipped product-feed-driven shopping ads in ChatGPT, with Criteo and StackAdapt as ad tech partners. The same day, Google announced Affirm and Klarna integrations for AI Mode and the Gemini app, both built to the Universal Commerce Protocol. On May 13, Amazon retired the Rufus name, put Alexa for Shopping in the main search bar, and made Buy for Me and Shop Direct standard equipment on the marketplace. By Saturday, OpenAI was telling staff that ChatGPT, Codex, and the developer API will merge into a single agentic platform under Greg Brockman, timed to land in front of Google I/O. Most of last quarter’s “agentic commerce is coming” forecasts ran out of headroom this week.

Here’s what happened.

OpenAI turns the catalog into the ad

On May 12, Digiday reported that OpenAI shipped product-feed campaign automation for ChatGPT shopping ads. Retailers connect their existing product catalog (the same structured feed they already send to Google Shopping), set filters for which products are eligible, and the platform generates ads automatically using product names, images, and attributes. Ads still appear below the ChatGPT response, still labeled “sponsored,” but the unit of creation moved from “campaign” to “feed.” The platform can handle up to one million SKUs per advertiser. Onboarding currently starts with a 100-product sample feed before full catalog access is unlocked, per Search Engine Land’s coverage the same day.

Criteo is OpenAI’s first ad tech partner, with one retail brand already live in testing. StackAdapt is the second confirmed partner, and per one of the ad buyers involved, the integration “lowers adoption barriers” by accepting existing Google Shopping feeds without reformatting. The conversion data behind the rollout was published a week earlier: per Criteo’s May 5 update, more than 1,000 brands are now running through the Criteo-OpenAI integration, with AI-referred conversion rates approaching 2x traditional search in consumer electronics, lifestyle, and home and garden, and click-through rates roughly 3x comparable formats elsewhere.

Why it matters for merchants: The catalog used to be the input to your website. Then it became the input to your Google Shopping campaigns. Now it is the input to your ChatGPT ad placements, with the same file. The unlock is that the cost of running ads in ChatGPT just collapsed for any merchant with a clean feed, and the cost of running ads in ChatGPT without a clean feed stayed effectively infinite. The 2x conversion rate is the carrot. The 100-SKU sample feed gate is the stick. If your product attributes are sparse, your titles are SKU-codes, or your images are dim, you do not get into the pilot, and the brand sitting next to you with a properly enriched feed gets there first. This is the same catalog hygiene story we have been telling for six months, only now it has a paid-acquisition price tag attached. The retailer who treats the product feed as a marketing artifact rather than a back-office artifact wins the next two quarters of ChatGPT placements. (See our coverage of the 99% catalog gap for the underlying readiness numbers.)

Amazon retires Rufus, puts Alexa for Shopping in the search bar

On May 13, Amazon announced Alexa for Shopping, the merger of Rufus (the product-expertise model) and Alexa+ (the cross-device personalization layer) into a single assistant. The Rufus brand is being retired from the shopping interface. Rufus the model continues to power parts of the system behind the scenes, but the consumer-facing name in the Amazon app, on amazon.com, and on Echo Show devices is now Alexa for Shopping. The rollout is to all U.S. customers over the week of May 13.

The functional list is the news. Alexa for Shopping ships with Scheduled Actions (set a condition and the agent buys automatically when met), Buy for Me (the agent completes purchases on third-party merchant websites), Shop Direct (over 400,000 external merchants surfaced inside Amazon search), 365-day price history, side-by-side comparisons in the result list, persistent memory across Amazon properties, and conversation-built carts that draw on prior purchase history. Per PPC Land’s deep dive, Rajiv Mehta, VP of Conversational Shopping at Amazon, described it as “an expert personal shopper” with memory across devices. Rufus, the 2024-launched predecessor, was used by more than 300 million customers in 2025, with monthly actives up 115% and engagement up nearly 400% year over year.

The competitive read, per Retail Dive, came from Brad Jashinsky at Gartner: “Alexa has much higher brand awareness than Rufus.” The brand recognition gap matters because Alexa is now occupying the search bar, not a side panel.

Capability	Before May 13	After May 13
Consumer-facing name	Rufus	Alexa for Shopping
Surface	Side chat window	Main search bar + chat + Echo Show
Buys on third-party sites	Buy for Me (March 2026)	Buy for Me, mainstream
External merchants visible	Shop Direct (March 2026)	Shop Direct, 400,000+ merchants (and growing)
Scheduled buying	Not available	Scheduled Actions (condition-based auto-buy)
Cross-device memory	Limited	Persistent across app, web, Echo

Why it matters for merchants: The Amazon merchant story has two halves now, and they are about to be the same half. For sellers on Amazon, Sponsored Prompts inside Rufus exited beta and became billable on March 25, 2026, and that monetization is now inheriting Alexa’s brand reach and search-bar placement. Cost per click inside an AI-filtered recommendation is going to look very different from cost per click inside a 1990s-style search-results page, and merchants will be bidding on intents (“a quiet treadmill for an apartment under $1,500”) rather than keywords. For brands off Amazon, the Shop Direct number (400,000+ merchants) is the news. Amazon’s search bar now points at the open web by default. Whether you wanted distribution through Amazon or not, you are about to be tested for it. The first action is to find out whether your products are eligible for Shop Direct, and the second is to make sure the listing surfaced inside Alexa for Shopping is the listing you would want a customer to see. (See our earlier coverage of Buy for Me listing products without permission for the background on how Amazon ingests external catalogs.)

Google plugs Affirm and Klarna into AI Mode under UCP

On May 12, per Digital Commerce 360, Google announced simultaneous partnerships with Affirm and Klarna to bring buy-now-pay-later into agentic checkout flows. The integrations span traditional Google Search, AI Mode, and the Gemini app, with payment completion routed through Google Pay. Affirm built its integration to the Universal Commerce Protocol, and Klarna joined UCP at the same time, putting both providers on the open standard Google co-developed with Shopify and the rest of the UCP Tech Council. Affirm’s experience rolls out “in the coming weeks.” Klarna’s is launching in tandem.

Ashish Gupta, VP and GM of merchant shopping at Google, framed the move on payment integrity: “As AI becomes a more active part of how people discover and buy, it’s critical that the payment options remain secure and reliable.” David Sykes, Chief Commercial Officer at Klarna, framed it on consumer behavior: “As shopping moves into conversational and AI-driven environments, flexible payments become essential infrastructure for how people buy.” Vishal Kapoor, SVP of product at Affirm, framed it on user choice: “People deserve transparent, flexible financial options.” The three framings stacked together tell the same story: the agentic checkout was missing the BNPL slot, and the BNPL slot is now filled.

Why it matters for merchants: Two things to read off this announcement. First, the UCP compliance line is the new floor. Affirm and Klarna did not negotiate one-off integrations with Google. They built to the same protocol that Shopify, Amazon, Stripe, Salesforce, Mastercard, Visa, and Microsoft sit on (we covered the council expansion in April). For a merchant, that means the connector you build to UCP buys you BNPL eligibility on the agent surface as a bonus, not a new project. Second, the customer-facing impact is shifting the agent’s “no” rate. A non-trivial share of agentic checkouts get abandoned because the customer wants Klarna or Affirm and the agent surface does not offer it, with the basket spilling back out to a website checkout. That leak is about to plug. The corollary for merchants is to make sure the BNPL terms you offer on your own site are no worse than the terms surfaced inside AI Mode, because Google now controls the comparison. For European merchants specifically, Klarna’s coverage will land first where it already holds the consumer credit and banking licenses (Sweden, Germany, the U.K., France, Spain), while Affirm’s footprint remains North America-centric. The agent BNPL map is going to look different on the two sides of the Atlantic for at least the next year.

OpenAI quietly merges ChatGPT, Codex, and the developer API under Brockman

On May 16, OpenAI told staff that co-founder and president Greg Brockman will permanently lead a unified organization covering ChatGPT, Codex, and the developer API. The reorg was reported by The Next Web and TechTimes the same day, landing four days before Google I/O 2026 opens on May 19. Brockman’s note to staff, per the reports, was that OpenAI will “invest in a single agentic platform and merge ChatGPT and Codex into one unified agentic experience for all.” Codex on mobile (iOS, Android) shipped two days earlier on May 14, and the internal framing is that Codex will expand into general productivity tasks first, with ChatGPT and the company’s research browser Atlas folded in over time.

The day-to-day product the merchant interacts with does not change this week. The strategic frame for the next twelve months does. OpenAI’s stated direction is a single application where a consumer can have a shopping conversation, kick off a multi-step agent task, run a price comparison, and have the agent complete the checkout, all without leaving the ChatGPT surface. That is the same shape Google is building for Gemini, and the same shape Amazon is building for Alexa.

Why it matters for merchants: Three years ago, your channels list was Google, Amazon, Meta, and email. This week’s announcements collapse three of those into a smaller number of agentic surfaces (ChatGPT, Gemini, Alexa, Copilot) plus a smaller number of payment-and-discovery protocols (ACP, UCP). The implication is that “AI strategy” stops being a parallel workstream to commerce strategy and starts being commerce strategy. The concrete decision this week is sequencing: a merchant who has a finite ecommerce roadmap for Q3 should put the ChatGPT presence ahead of the Codex-API integration, because the unified app is going to put the shopping conversation in front of the largest consumer surface before it exposes any new developer SKU. Treat the surfaces as channels, audit your presence on each of them quarterly, and assume the merger of ChatGPT and Codex will surface as a more capable agent-shopper inside ChatGPT before the end of 2026.

Stellagent launches Agentic Commerce Studio (Japan / APAC angle)

On May 14, Yokohama-based Stellagent launched Agentic Commerce Studio, a browser-based environment that lets merchants and platforms test AI agent shopping flows end to end. The Studio runs natural-language product search, recommendations, cart creation, shipping quotes, and checkout preparation against a connected test environment, and validates the merchant server’s response to product-feed queries, inventory endpoints, shipping calculators, checkout sessions, and webhooks. The protocol coverage list is the interesting part: ACP, UCP, AP2 (Google’s Agent Payments Protocol), Visa TAP, and Mastercard Agent Pay all in scope. The launch is targeted at the Japan and Asia Pacific market first.

Why it matters for merchants: The story behind the story is that Japan now has a domestic agentic-readiness vendor. Until this week, the readiness tooling ecosystem (Mirakl, Crossing Minds, Bloomreach, the U.S. and European auditors we have covered) was almost entirely Western. Stellagent is the first APAC-headquartered entrant. For European merchants, the read is that the regional readiness layer is going to fragment by jurisdiction faster than the protocols themselves, which means the agentic readiness tool you pick in 2026 will likely be replaced by a regionally tuned one in 2027.

Where the AI commerce stack stands this week

Layer	Update this week
Ads on agent surfaces	OpenAI ships ChatGPT product-feed campaigns with Criteo and StackAdapt
Marketplace agent	Amazon retires Rufus brand, ships Alexa for Shopping with cross-web Buy for Me
Agent payment options	Google adds Affirm and Klarna BNPL to AI Mode under UCP
Platform consolidation	OpenAI merges ChatGPT, Codex, and developer API into one agentic platform
Regional readiness tooling	Stellagent launches Agentic Commerce Studio for Japan and APAC

What merchants should do this week

1. Treat your product feed as marketing inventory, not just operational inventory. OpenAI’s product-feed ad format is now live, Criteo and StackAdapt are accepting integrations, and the same Google Shopping feed you already maintain is the on-ramp. If your titles are not descriptive, your attributes are not complete, or your images are inconsistent, you do not get the 2x conversion benefit even if you get into the placement. Run a feed audit this week. Promote it from “data ops” to “growth.”

2. If you sell on Amazon, find out today whether your listings are eligible for Shop Direct and Buy for Me. With Alexa for Shopping in the main search bar, the difference between “indexed but not surfaced” and “indexed and surfaced” is the difference between a quarter of growth and a quarter of decline. Check your seller dashboard, audit your top 20 SKUs against the Alexa for Shopping result for the obvious customer query, and identify which ones are not winning the slot.

3. Make sure your BNPL offer matches what AI Mode is about to show. Once Google’s Affirm and Klarna rollout reaches your customers, the BNPL term shown next to your product inside AI Mode will be the term the customer expects on your site. If your direct-checkout BNPL terms are worse (longer installments, less attractive interest, smaller eligible basket), you will lose the comparison at the moment the agent surfaces it. Reconcile the terms now.

4. Build to UCP, not to one-off integrations. Affirm and Klarna did not negotiate a Google-specific integration. They built to UCP and inherited Google. Stripe, Mastercard, Visa, Amazon, Microsoft, and Salesforce are now all in the same protocol. The merchant who builds a UCP connector this quarter gets the BNPL slot, the Alexa Shop Direct slot, and the AI Mode slot in one project. The merchant who builds three bespoke connectors gets none of them on time.