Why AI Product Discovery Beats In-Chat Checkout — And Why Shopify May Not Have Gotten It Right (For Now)

The rush to AI checkout

In late 2025 and early 2026, three major moves reshaped online commerce:

The message was clear: the tech giants want you to buy things without ever leaving the AI conversation.

But here’s what happened in between: OpenAI quietly abandoned Instant Checkout. On March 24, 2026, CNBC reported that OpenAI was deprioritizing Instant Checkout and pivoting to product discovery. They stated it “did not offer the level of flexibility that we aspire to provide.”

The pivot was driven by real problems: OpenAI struggled to onboard merchants, show accurate product data, and build multi-item carts. As of February 2026, they hadn’t even built a system to collect and remit state sales taxes. Users simply weren’t using the chatbot to make purchases.

So where does Shopify’s checkout fit? Shopify’s Agentic Storefronts use Shopify’s own checkout infrastructure embedded inside ChatGPT — not OpenAI’s Instant Checkout. When you buy from a Shopify merchant in ChatGPT, it’s Shopify Checkout (via Shop Pay) running inside ChatGPT’s interface, not OpenAI’s abandoned system. This is the key distinction: OpenAI gave up building their own checkout, but Shopify brought theirs.

The trust gap is real

The data is stark. Multiple surveys paint the same picture:

• Only 13-17% of consumers are willing to complete a purchase inside an AI chat without visiting the merchant’s store (Salesforce; McKinsey)
• 68% of consumers want to see a full product page with reviews, size charts, and return policies before buying (Baymard Institute)
• 52% are open to AI-assisted discovery — but not AI-controlled checkout (McKinsey)

The pattern is clear: consumers want AI to help them find things, not buy them. This distinction matters enormously.

Trust in AI checkout is especially low for high-consideration purchases: fashion (where size and fit matter), electronics (where specs matter), and furniture (where seeing is believing). For commodity repeat purchases (phone cables, pet food you’ve bought before), trust is somewhat higher — but these are low-margin categories where commission fees eat into already thin profits.

The returns problem nobody is talking about

When a consumer buys inside an AI chat, they typically see 1-2 product images, a short AI-generated description, and a price. That’s it.

What they don’t see — and why it matters:

• Size charts and fit guides — The #1 reason for fashion returns. A merchant’s product page shows detailed measurements, model height/weight, and fit comparisons. An AI chat shows none of this. The customer guesses, orders, and returns.
• Full product photography — Merchants invest in 10-20 images per product: multiple angles, zoom, lifestyle shots, video. AI chat surfaces 1-2 images. A customer can’t assess color accuracy, texture, or scale from a single thumbnail.
• Customer reviews and Q&A — 93% of consumers read reviews before buying. In an AI chat, reviews are summarized (or absent). The customer loses the ability to read specific feedback about sizing, durability, or quality from real buyers.
• Return policy, shipping costs, and delivery times — These vary wildly between merchants. A 30-day free returns policy is very different from a 14-day returns window with a €7.95 restocking fee. In an AI chat, this information is buried or absent.
• Related products, bundles, and alternatives — A merchant’s product page suggests better alternatives, bundles that save money, and accessories. This is where upselling and customer satisfaction intersect. AI checkout bypasses all of it.

We don’t yet have direct data on AI checkout return rates — the sample sizes from OpenAI’s brief Instant Checkout experiment were too small. But we have strong proxy data from social commerce, where product information is similarly compressed:

The merchant bears the full cost of these returns — shipping, restocking, lost inventory — but in the AI checkout model, they’ve lost the customer relationship. No opportunity to offer an exchange, a store credit, or a personalized discount. The customer contacted OpenAI or Shopify support, not the merchant.

New fraud vectors, old chargeback infrastructure

The authorization problem

“Buy me something nice for Mom’s birthday” — has the user authorized a €50 purchase or a €500 one? When an AI agent interprets a vague instruction and completes a transaction, the boundary between authorized and unauthorized becomes blurry.

Prompt injection

This is speculative but worth considering: security researchers have demonstrated that AI agents can be manipulated via prompt injection — hidden instructions embedded in web content that trick the AI into unintended actions. In theory, a malicious product listing could contain hidden instructions that influence an AI shopping agent’s recommendation or purchasing behavior. While no real-world AI commerce exploits have been publicly documented, the attack surface exists and grows as AI agents gain purchasing authority. (For background on prompt injection risks, see OWASP Top 10 for LLM Applications.)

Chargebacks don’t fit

The existing chargeback system (Visa, Mastercard) was designed for bilateral disputes: “I didn’t buy this” or “This isn’t what I ordered.” AI-mediated purchases create a new category: “I asked the AI to buy something but this isn’t what I meant.” Neither the merchant, the AI platform, nor the payment processor has a clean process for handling this.

EU regulation says not yet

For European merchants, AI checkout faces regulatory headwinds that aren’t going away soon:

Consumer Rights Directive — Article 6

Merchants must provide comprehensive pre-contractual information before a purchase: product characteristics, total price, return rights, delivery details, trader identity. A compressed AI chat interface arguably does not satisfy these requirements. If regulators agree, the withdrawal period extends from 14 days to 12 months. (Directive 2011/83/EU)

AI Act (August 2026)

AI systems that influence purchase decisions must be transparent. Users must know they’re interacting with AI, and commercial arrangements that influence recommendations must be disclosed. AI checkout that appears to neutrally recommend Product A while having a commercial deal with the brand faces scrutiny. (EU AI Act — Regulation 2024/1689)

PSD3 and Strong Customer Authentication

European payment regulation requires Strong Customer Authentication (biometric/2FA) for electronic payments. This reintroduces the very friction that AI checkout was designed to eliminate. An AI can’t complete biometric authentication on your behalf. (European Commission PSD3 Proposal)

The “Third Actor Problem”

“AI platforms must not be allowed to profit from facilitating transactions while externalizing all risk to consumers and merchants.” — BEUC (European Consumer Organisation), 2025 position paper on AI and consumer protection

EU consumer law is built on a two-party relationship: merchant and consumer. AI checkout introduces a powerful third party — the AI platform — that influences product selection, controls the purchase interface, and may store payment credentials, but takes zero liability for the product, delivery, or customer satisfaction.

This structural gap will need legislation to resolve. Until it does, operating in this space carries regulatory risk.

What the industry is saying

We’re not alone in this assessment:

• Forrester Research: “What It Means That The Leader In Agentic Commerce Just Pulled Back” — analyzing OpenAI’s retreat from checkout
• Ecommerce Europe: Has consistently called for merchants to retain the direct customer relationship in AI-mediated commerce
• Benedict Evans: “AI in commerce will look more like a better search engine than a new point of sale” — Annual Presentation 2025
• OpenAI (by actions): Abandoned Instant Checkout in March 2026 and pivoted to discovery-first

Discovery-first: the better model

The discovery-first approach is simple: AI agents help consumers find the right product. When they’re ready to buy, they click through to the merchant’s own store and complete the purchase there.

The department store analogy

Think of AI as a knowledgeable sales assistant in a department store. They listen to what you need, walk you to the right section, explain the differences between options, and make a recommendation. But you go to the checkout counter. You see the price, confirm the purchase, and get the receipt. The assistant doesn’t take your wallet and pay on your behalf.

Discovery-first AI follows this same natural pattern. It’s how humans actually want to shop.

What this means for merchants

The window is now. AI shopping traffic to e-commerce stores grew 7x since January 2025, and AI-attributed orders are up 11x. The consumers using AI to discover products are high-intent shoppers. You need to be visible to these agents — and you need to be visible now, before your competitors are.

But you don’t need to hand over your checkout, your customer data, or your revenue to do it. You just need your products in the right format for AI agents to find them.

That’s what we do.