Agentic commerce after Instant Checkout: what online stores should actually do

OpenAI launched Instant Checkout in ChatGPT on 29 September 2025 and wound it down on 24 March 2026, saying the first version "did not offer the level of flexibility that we aspire to provide". Six months in, a Forrester analyst told CNBC roughly 30 Shopify merchants were live. AI shopping did not fail. Buying inside the chat did.

The pattern that replaced it is simpler: discover in AI, buy on the merchant's site. Here is what that means for an online store, and what to do about it.

What happened to ChatGPT Instant Checkout?

OpenAI moved checkout back to merchants and refocused ChatGPT on product discovery. Since March 2026, merchants send product feeds and promotions to ChatGPT, shoppers find products there, and the purchase completes in the merchant's own checkout or in a retailer's app inside ChatGPT, not in a generic OpenAI flow.

The original launch, built on the open Agentic Commerce Protocol developed with Stripe, supported single-item purchases from US Etsy sellers, with Shopify merchants "coming soon". OpenAI's March announcement expanded the same protocol to cover product discovery instead, and named Target, Sephora, Nordstrom, Lowe's, Best Buy, The Home Depot and Wayfair as integrated.

The behavioural evidence was blunt. CNBC, citing Walmart's data, reported that conversion was three times lower for products bought directly in ChatGPT than for those that sent shoppers to the retailer's site. Walmart's Daniel Danker had already called Instant Checkout "a very temporary moment in time".

Why did buying inside the chat stall?

Because shoppers already have accounts, saved payment details, loyalty points and order histories with the stores they trust, and a chat window offered none of them. People were happy to research and compare in an AI assistant, then preferred to finish the purchase somewhere that already knew them.

That is consistent with how OpenAI described the change. Merchants could now "use their own checkout experiences while we focus our efforts on product discovery". Walmart's response was its own app inside ChatGPT, with account linking, loyalty and Walmart payments — the retailer's checkout, surfaced in the assistant.

Shopify launched Agentic Storefronts on the same day as OpenAI's announcement, letting merchants sell through ChatGPT, Microsoft Copilot, Google's AI Mode and Gemini. On desktop, ChatGPT links out to the merchant's site; orders arrive carrying ChatGPT referral attribution, with no fees beyond standard payment processing.

Is agentic checkout dead everywhere?

No. Google's Universal Commerce Protocol, launched in January 2026 with Shopify, Etsy, Wayfair, Target and Walmart, powers checkout inside AI Mode and Gemini for approved US retailers, and PayPal runs Instant Buy in Perplexity. What survived is checkout where the merchant remains the merchant of record and the shopper's existing payment wallet comes along.

Google's UCP announcement listed more than 20 endorsers, including Visa, Mastercard, Stripe, Adyen and American Express. By May, Google had added a cart that works across Search and Gemini, with Google Pay checkout at retailers including Nike, Sephora, Target and Walmart.

PayPal and Perplexity's Instant Buy launched in November 2025, again with the retailer as merchant of record. Payment networks have built the rails underneath: Visa Intelligent Commerce and Mastercard Agent Pay were both announced in April 2025.

The distinction that matters: checkout that carries the shopper's existing trust works. Checkout that asks them to start from scratch inside a chatbot did not.

How much shopping traffic does AI actually send?

Not much yet in absolute terms, but it is growing faster than any other channel. Adobe measured traffic from generative AI tools to US retail sites rising 693% year on year over the 2025 holiday season, while noting the "base of users remains modest". It forecasts a further 130% rise for holiday 2026.

Adobe's holiday analysis draws on more than a trillion visits to US retail sites. Its September 2026 forecast adds a consumer survey in which 77% of AI shoppers said they felt more confident in their purchases.

Salesforce puts a larger frame around it. Its 2025 holiday data attributes $262 billion, or 20% of global online holiday sales, to AI and agents. That figure includes on-site recommendations and service agents, not just chatbot referrals — read it as the size of AI's influence, not of AI search traffic.

The traffic that does arrive is well qualified. Salesforce reports AI-search shoppers converted nine times more often than those arriving from social media referrals.

What decides whether an AI assistant recommends your products?

Your product data. Every AI shopping surface — ChatGPT, AI Mode, Gemini, Perplexity, Copilot — builds its recommendations from structured product feeds and the pages behind them. A product with a vague title, missing attributes or a stale price is less likely to be matched to a specific shopping question.

Google has published the clearest numbers. In its September 2026 shopping update, it says merchants adopting core Merchant Center feed best practices see an average 5% increase in conversions the following month. In testing with lululemon, brand-supplied conversational attributes were used in 50% of relevant AI Mode recommendations.

OpenAI's commerce specification requires nine fields per product: ID, title, description, URL, brand, seller name, image URL, availability and price. Search is enabled by default, checkout is disabled by default, and the full catalogue is refreshed at least daily. Feeds are open to approved partners; merchants apply through ChatGPT's merchant programme.

The shift is from keywords to questions. A shopper asks an assistant for "a waterproof trail shoe for wide feet under $150". The product that gets recommended is the one whose data states it is waterproof, comes in wide fit, and costs $139 — explicitly, in fields, not implied in marketing copy.

What should an online store do now?

Treat AI assistants as a discovery channel and make your catalogue legible to them. Get into the feeds each platform reads, enrich product attributes so they answer real shopping questions, keep your own site fast and server-rendered, and measure AI referrals separately so you can see what they are worth.

A practical checklist for an ecommerce business:

1. Claim the feeds. Merchant Center for Google surfaces; Shopify's Agentic Storefronts or Catalog if you are on Shopify; ChatGPT's merchant programme for OpenAI.
2. Enrich attributes. Material, fit, dimensions, compatibility, use case — the facts a shopper asks about. Write titles that state what the product is, not a campaign name.
3. Keep price and availability live. A recommended product that turns out to be out of stock costs you the sale and the assistant's trust.
4. Mark up product pages. Product, Offer and Review structured data on pages that render as HTML, because some AI fetchers do not run JavaScript.
5. Answer comparison questions on your own site. Buying guides and comparison pages are what assistants cite when they explain why a product fits.
6. Track AI referrals as their own channel. Separate ChatGPT, Perplexity, Gemini and Copilot in analytics, and compare conversion against search and social.

Items one to four are technical SEO work extended to product data. Item five is generative engine optimization, and item six is analytics and conversion — without it, you cannot tell whether any of the rest is working.

Should you build for agentic checkout?

Only where your platform makes it close to free. If you are on Shopify, Agentic Storefronts and UCP are largely switched on through the platform. A bespoke checkout integration with each AI assistant, for most mid-sized retailers, costs more than the share of sales it is likely to carry today.

The evidence so far points the same way: shoppers buy where they already have an account. The durable investment is in being recommended — accurate feeds, specific attributes, citable pages — which pays off whichever assistant, and whichever checkout model, wins.

For how one major assistant chooses what to cite, see how Claude chooses which websites to cite; for measuring your visibility across all of them, see running an AI citation baseline.

The short version

Checkout inside a chatbot stalled because shoppers prefer stores that already know them; discovery inside AI assistants is growing fast and converting well. Product feeds and structured attributes now decide which products get recommended. If you want to see where your catalogue stands across ChatGPT, Google AI Mode, Gemini and Perplexity, that baseline is where our GEO engagements start.