Agentic Checkout: Why the Online Shopping Cart Dies in 2026

Since September 29, 2025, ChatGPT users in the US can buy products directly inside the chat window — without ever opening a merchant's website. No cart, no checkout form, no order confirmation page. OpenAI calls the feature Instant Checkout, and Visa and Mastercard are simultaneously building the payment infrastructure meant to make agent-driven purchases work at scale.

For e-commerce managers at DTC brands, that's an uncomfortable headline. For years, budget has flowed into checkout optimization, A/B tests on button colors, exit-intent popups, and cart-abandonment flows. If purchases increasingly close out inside an agent's chat window, those investments stop paying off — not because the work was bad, but because the brand simply loses visibility into the moment that matters most.

This article breaks down what Instant Checkout, the Trusted Agent Protocol, and Agent Pay actually mean on a technical level, where these systems fall short — and which steps e-commerce teams should take now on product feeds and attribution, before the rollout reaches Europe.

Three initiatives currently define the infrastructure for agent-driven purchases, and they mesh together like gears — even though they come from competing companies.

OpenAI Instant Checkout launched in September 2025 with Etsy as its first commerce partner. Since then, US users have been able to buy Etsy products directly inside the ChatGPT window. At the same time, OpenAI announced its expansion to Shopify merchants — according to the announcement, more than 1 million merchants from the Shopify ecosystem are set to be connected, including brands like Glossier and SKIMS. For brands, that means ChatGPT's reach — OpenAI cited roughly 700 million weekly active users at the time of the launch announcement — is turning into a direct sales channel, not just a research surface.

Visa's Trusted Agent Protocol (TAP), unveiled in October 2025 and developed together with Cloudflare, tackles a different problem: how does a merchant or card-issuing bank know whether a requesting AI agent is a legitimate purchasing party — or a scraping bot? TAP cryptographically verifies agents to issuers and merchants while transmitting purchase intent and customer context. The agent proves who it is before any money moves.

Mastercard Agent Pay, announced back in April 2025, pursues a related concept with its own tokenization layer: so-called Agentic Tokens, built on top of Mastercard's existing tokenization infrastructure. This gives an agent a payment-capable but limited authorization — comparable to a virtual card tied to one specific order.

The technical thread connecting all these initiatives is the Agentic Commerce Protocol (ACP), which OpenAI released as an open standard together with Stripe. ACP defines how agents query product data, submit orders, and delegate payments — regardless of which card network settles the transaction behind the scenes.

Visa's chief product officer, Jack Forestell, compared this shift to the leap from brick-and-mortar retail to online shopping: "Soon, people will have AI agents browse, select, purchase, and manage on their behalf" — his framing at the launch of Visa's Intelligent Commerce initiative. Who's building this infrastructure is now clear. The more interesting question for practitioners is what a purchase inside it actually looks like, step by step.

An agentic checkout purchase differs fundamentally from a classic shop session — and it differs at every single point in the chain.

1. User request in chat: The user states a purchase intent in natural language – something like "Order the same dog food I got last month" or "I need a gift for someone who loves ceramics, under $40."
2. Product search via feed or API: The agent scans connected merchant feeds or pulls product data through the ACP interface. It compares price, availability, shipping time, and product attributes – not landing pages.
3. Product selection and confirmation: The agent presents options in the chat. The user confirms with a click or a short reply. With Instant Checkout, the user stays in the loop – the agent doesn't buy fully autonomously, it secures final approval first.
4. Agent verification: Before payment is authorized, the agent identifies itself to the card network and issuer via the Trusted Agent Protocol (Visa) or Agentic Tokens (Mastercard). The bank confirms: this transaction comes from a verified agent acting on user authority, not an unknown bot.
5. Tokenized payment: Payment runs through stored, tokenized payment credentials – no redirect to the merchant's site, no login screen, no cart icon. Stripe handles payment processing behind OpenAI's Instant Checkout and settles the transaction server-side with the merchant.
6. Order confirmation in chat: The order confirmation shows up directly in the conversation thread. The merchant remains the merchant of record, handling shipping, returns, and customer service – but the entire visible purchase journey happened off their domain.

Here's the part that matters most for anyone owning conversion: between purchase intent and order confirmation, there is zero traditional brand touchpoint. No product detail page, no cross-sell banner, no newsletter opt-in moment. When the entire purchase flow bypasses these touchpoints, one question becomes unavoidable: what's actually left of the classic funnel?

The classic e-commerce funnel rests on a quiet assumption: the brand controls every stage from awareness to checkout. It runs the ad, builds the landing page, optimizes the product detail page, designs the checkout flow. Every stage is measurable, every stage is tunable. That control disappears the moment an agent stands between brand and customer.

Three established tools lose their bite in this new model:

Cart-abandonment campaigns hit a dead end. There's no abandoned cart, because there's no cart. The agent narrows down the options, the user confirms or rejects — and the in-between state of "item sitting in the cart, shopper on the fence," which an entire industry of recovery emails and retargeting ads was built on, simply doesn't exist in this flow. Retargeting pixels never fire, because there's no page visit for them to catch.

A/B tests on checkout UX become pointless. Whether the buy button is green or orange, whether the form has three fields or five, whether trust badges sit above or below the price — none of it matters when the purchase happens inside an agent interface that OpenAI controls, not the brand.

The purchase decision shifts from the UX layer to the data layer. An agent doesn't respond to urgency banners or clever page layouts. It compares structured attributes: price, availability, delivery time, reviews, product specs. Landing at the top of an agent's recommendations comes down to the quality and completeness of your product data — not the polish of your product page.

This isn't a gradual shift, it's a change in category: the brand doesn't lose the customer. It loses direct access to the moment of purchase. That distinction becomes especially clear in conversations with e-commerce teams that have spent years fine-tuning their checkout flows — reactions swing between "this doesn't affect us yet" and genuine concern that entire optimization teams could lose their purpose overnight. Both reactions miss the point. It's a subtle but critical difference — and it explains why so many commerce teams oscillate between panic and denial. Before declaring the funnel dead, though, it's worth taking a clear-eyed look at where this technology actually falls short.

The idea that the funnel is dead entirely doesn't hold up under closer scrutiny – for at least four reasons.

High-involvement purchases don't work in chat. Someone buying a $2,600 sofa, configuring an e-bike, or selecting a luxury watch goes through a visual, often emotional decision-making process: material previews, configurators, sizing guidance, showroom visits. A text-based agent can compare specs, but it can't convey the feeling a brand has spent years building into its product presentation. For these categories, the brand's own platform remains the central conversion point.

The liability question is still unresolved. What happens when the agent misinterprets a request and orders the wrong product? The Trusted Agent Protocol and Agent Pay govern the transaction side – verification, tokenization, authorization. They don't address who's liable for bad purchases caused by agent misinterpretation: the user who gave the instruction, the agent operator who misread it, or the merchant who fulfilled the order. Until regulation and case law bring clarity here, a residual risk remains – one that's especially likely to slow things down for high-ticket orders.

The rollout is deliberately conservative. Instant Checkout didn't launch broadly; it launched with a single partner – Etsy – and a catalog made up mostly of low-priced niche products. That's not a coincidence, it's risk management: small cart values, clear product descriptions, manageable return complexity.

That points to a conclusion that tends to get lost in the current debate – and one worth stating as a contrarian take: Agentic checkout isn't replacing the funnel. It's replacing the illusion that brands control every touchpoint. The agent is primarily taking over convenience purchases, reorders, and low-involvement decisions – exactly the transactions where users aren't looking for a brand experience in the first place, just efficiency. The emotional first purchase, brand discovery, and consultation-heavy buys will stay on owned channels for the foreseeable future. Once you understand that, you stop fighting the agent and start optimizing for it.

Because the technology only applies selectively, the real task for e-commerce teams shifts to a layer that used to be treated as a box-checking exercise: product data.

When agents don't visit landing pages but query structured data instead, the product feed becomes the storefront. Four technical requirements determine whether a brand even shows up in the agent channel at all.

First: structured product data following Schema.org and merchant feed standards. Complete GTINs, precise categorization, and machine-readable attributes for material, size, color, and compatibility replace the SEO landing page as the primary visibility driver. An agent can only recommend what it understands—and it only understands what's structured. Flowery marketing copy with no attribute structure behind it is simply invisible to an agent. In feed audits we've run for DTC brands, this is consistently the biggest gap: attributes aren't missing entirely, they're just buried in body copy instead of being properly structured. If you're already rewriting product copy, be realistic about the lift involved—our analysis on AI-generated product descriptions breaks down what this actually takes at scale across larger catalogs.

Second: integration with the Agentic Commerce Protocol or Shopify feeds. Shopify merchants will eventually get this through the platform integration itself; anyone running a custom stack has to implement the ACP endpoints themselves. Without this baseline technical requirement, there's no participation—full stop.

Third: real-time availability and pricing accuracy. In a traditional shop, the customer double-checks price and delivery time right before buying. An agent buys without that intermediate check. A feed reporting stock that doesn't actually exist, or a price that's been stale for days, generates cancellations, complaints, and lost trust almost immediately—both with the shopper and with the agent operator, who will start deprioritizing merchants with unreliable data.

Fourth: tokenization compatibility. For an agent to even register as payment-authorized with the card network, the transaction chain needs to support Visa TAP or Mastercard Agent Pay. This falls mainly on the payment provider—but merchants should be checking now whether their PSP has these protocols on the roadmap.

1. Check data completeness: Audit GTINs, attribute coverage, and Schema.org markup across your entire catalog — not just your bestsellers.
2. Measure update frequency: How fast do price and inventory changes hit your feed? Anything beyond a few minutes of latency is a liability in an agent-driven context.
3. Confirm interface readiness: Check your shop system's ACP compatibility and ask your payment provider about their TAP/Agent Pay roadmap.
4. Structure your returns and policy data: Return terms, delivery windows, and shipping costs need to be machine-readable — agents factor these directly into ranking decisions.

For a real-world example of what a clean feed stack looks like in practice, take our Shopify project with Papas Shorts — structured product data was the foundation for scalable visibility there long before agentic commerce entered the conversation. But even with a perfectly optimized feed, one question remains: how do you measure success when the classic click path is gone?

The entire attribution toolkit e-commerce relies on rests on one assumption: customers move through measurable sessions on measurable domains. That assumption disappears in the agent channel — and with it, decades of hard-won certainties.

UTM parameters and session tracking stop working. There's no referrer URL, no cookie, no click from ad to landing page. From the merchant's perspective, the purchase materializes as a server-side order with no history attached. Anyone trying to find agent-driven purchases in Google Analytics will find, at best, a direct order with zero context.

Server-side order confirmations become the primary data source. The merchant API through which the agent submits the order is now the only reliable data point. E-commerce teams need to rebuild their reporting so order origin gets captured and tagged at the API level: did this order come through the brand's own store, a marketplace, or an agent channel? That tagging has to go into the data pipeline before meaningful volume starts flowing — it can't be reconstructed retroactively. That's the lesson from previous attribution overhauls: teams that only start tagging after the first revenue bump lose exactly the baseline they'll need most later on.

Success gets measured at the SKU level, not the campaign level. Because there's no session history, the question "which campaign converted?" loses its foundation. In its place: "which SKU is performing in the agent channel — and why?" Products with complete data, competitive pricing, and clear policies get recommended more often. Reporting needs to reflect that reality: agent-order share per SKU, recommendation frequency, cancellation rate within the agent channel.

Payment provider reporting interfaces become a supplementary data source. Visa and Mastercard track, via TAP and Agent Pay, which transactions were agent-mediated. That data — where merchants get access through reporting interfaces — adds a network-level perspective to a brand's own API view and helps cleanly separate agent-driven revenue from the rest of the business.

Brands that pair this shift with automation save themselves manual data consolidation — a space where AI-powered automation of reporting pipelines tends to deliver measurable value fastest, based on what we've seen. The early pilots emerging across the industry show exactly what these adjustments look like in practice.

The rollout decisions OpenAI, Visa, and Mastercard have made so far tell you more than any forecast could — because they reveal exactly where these companies see the lowest risk and the fastest learning curve.

Etsy as the launch partner was a deliberate category choice. Low-price, decisive niche products — handmade goods, vintage finds, craft items — fit chat-based buying almost perfectly: the shopper already knows roughly what they want, cart values stay modest, and Etsy listings have always been detail-rich. It's precisely the profile where an agent can recommend confidently and a bad purchase costs little.

The Shopify expansion signals the rollout pace. With the announced integration of more than one million Shopify merchants, OpenAI is scaling step by step — and the priority clearly sits with convenience categories: beauty, supplements, everyday essentials, reorders. These are categories where purchase intent is easy to articulate and the decision needs little hand-holding.

Visa and Mastercard are testing region by region. Both networks are piloting their protocols with select issuers in select markets first, not rolling out globally at once. For merchants in Europe, that buys some lead time — but it's not a reason to relax. Brands that wait until the channel goes live in Europe will be starting with unstructured feeds while competitors who moved early bank a substantial head start. The parallel to the shift in search visibility triggered by Google AI Overviews is hard to ignore: there, too, "this doesn't affect us yet" was the common refrain — right up until it affected everyone.

Clean data beats big brand names. The most consistent finding from the first few months: brands with complete product data, clear return policies, and reliable inventory come out ahead — regardless of brand recognition. An agent has no concept of emotional brand loyalty; it recognizes data quality, price-to-value ratio, and policy transparency. For smaller DTC brands, that's a rare opening: in agent rankings, media budget doesn't matter — the quality of your own data infrastructure does.

The fallout from all this isn't just procedural — it hits team ownership directly. Feed quality has traditionally sat with SEO or catalog managers, attribution with performance marketing, payment protocols with IT. Agentic checkout forces these three functions to the same table — brands that keep managing product feeds, order tracking, and payment roadmaps in separate silos will miss the exact intersection where the agent channel takes shape, and they'll end up measuring neither reach nor revenue share accurately.

The next concrete step doesn't take a week: check whether your product feed meets the baseline technical requirements for agentic commerce protocols. Complete GTINs, structured attributes, real-time pricing, and reliable availability data are the entry ticket — before agent visibility, ranking, or revenue share can even enter the conversation. The shopping cart isn't disappearing overnight. But the brands that outlast it will have already brought their feed team, attribution, and payment roadmap together.