The short version: OpenAI launched a feature that let ChatGPT complete purchases directly inside the chat window, then quietly retired it five to six months later after sales stayed close to zero. That is not a reason to ignore agentic commerce — the underlying protocols are real, backed by OpenAI, Google, Stripe, and the card networks, and discovery traffic from AI is already growing fast. It is a reason to be precise about which half of this is working today and which half isn't, before spending a budget on either.
We build Webflow sites for a living, so the practical question at the end of this article is one we get asked directly — worth saying upfront.
What "Agentic Commerce" Actually Means
The pitch is straightforward: instead of a person browsing, comparing, and clicking "buy," an AI agent does the full purchase lifecycle on their behalf — discovery, authorization, payment, fulfillment — starting from a stated goal rather than a session of clicks. "Book me a flight to Chicago under $400" or "reorder the dog food I usually get" becomes a single instruction the agent executes end to end.
That is the vision. 2026 is the year it got tested against real customers, real merchants, and real money — and the results were more instructive than the pitch.
What Actually Happened
OpenAI launched Instant Checkout inside ChatGPT, built on the Agentic Commerce Protocol (ACP) — an open standard developed jointly with Stripe. Around 30 Shopify merchants integrated it, letting a shopper complete a purchase without leaving the chat window.
OpenAI retired Instant Checkout roughly five to six months after launch, in March 2026. Reported sales from the integrated merchants stayed close to zero for the whole run. ChatGPT was repositioned toward product discovery and comparison — helping someone decide what to buy — while the actual transaction moved back to the merchant's own website, the way it always worked.
Worth being precise about what did and didn't fail here. The protocol — ACP — is still an open standard and still in use elsewhere. What failed was the specific bet that shoppers wanted to complete a purchase without ever leaving a chat interface. They apparently didn't, in large enough numbers to matter yet.
Why Checkout Specifically Stalled
Not "it's too early," which explains nothing. A few concrete reasons show up consistently in postmortems of this period:
- Trust doesn't transfer the way clicks do. Clicking through to a merchant's own checkout carries decades of learned trust signals — a URL, a padlock, a familiar page. Handing payment details to an agent inside a chat window skips all of that, and shoppers noticed.
- Returns and support get harder to reason about. If an agent bought the wrong size or the wrong flight, who resolves it — the merchant, OpenAI, or the shopper? That ambiguity is a real deterrent at the moment someone is deciding whether to trust the flow at all.
- The economics were thin. Thirty merchants and near-zero sales is not a rounding error; it suggests the demand simply wasn't there yet at any price point tested.
None of this means the idea is dead. It means the "buy without leaving the chat" version of it arrived before the trust infrastructure did.
What's Actually Working Right Now
Discovery, not checkout. Adobe Analytics measured a 4,700% year-over-year jump in generative-AI referral traffic to US retail sites between mid-2024 and mid-2025. People are asking ChatGPT, Perplexity, and Gemini what to buy in large and rapidly growing numbers — they are just finishing the purchase on the merchant's own site afterward, the same way they always did after a Google search.
That distinction is the entire article in one sentence: the "find it" half of agentic commerce is real and growing fast. The "buy it without leaving the chat" half isn't there yet.
The Protocol Landscape, Briefly
Three names are worth knowing, because they will likely outlast this specific stumble:
| Protocol | Built by | What it does |
|---|---|---|
| ACP (Agentic Commerce Protocol) | OpenAI + Stripe | Open standard for agent-initiated checkout; powered Instant Checkout, still used independently of it |
| AP2 (Agent Payments Protocol) | Google, announced Sept 2025 | Uses cryptographic "Mandates" to create verifiable proof of what a user actually authorized an agent to do |
| UCP (Universal Commerce Protocol) | A newer, broader cart/checkout standard; also used by Perplexity's Instant Buy |
Visa and Mastercard have also introduced their own agent-facing trust protocols for card-present-style verification when an agent, not a human, initiates a payment. The infrastructure is being built in parallel by nearly every major player in payments — which is a signal this is a "when," not an "if." It just isn't a "now" for most businesses.
What A Small Business Should Actually Do Today
Nothing expensive, and nothing that would be wasted even if agentic checkout takes another two years to mature — because all of it overlaps with fundamentals that already matter:
- Add Product, Offer, or Service schema to anything you sell or book. This is the same structured-data work that helps conventional SEO and AI citation, and it is exactly what an agent needs to understand what you offer, its price, and its availability.
- Keep your own checkout or booking flow fast and simple. Since the transaction is staying on merchant sites for the foreseeable future, a clean one-page checkout is more valuable now than a speculative agent integration — this is also just good Core Web Vitals practice.
- Don't block the AI crawlers behind discovery traffic. We covered the specific user agents — GPTBot, PerplexityBot, ClaudeBot, and others — and what blocking each one costs you in our guide to AI search optimization. That traffic is the part of agentic commerce that is demonstrably real right now.
- Do not build custom ACP or AP2 integration yet unless you are a platform-scale retailer. That is real engineering effort chasing a checkout flow that just failed its first real-world test. Revisit this specific point in six to twelve months.
The Honest Bottom Line
Treat 2026's agentic commerce story as a preview, not a deadline. The protocols being built by OpenAI, Google, and the card networks in parallel are a strong signal that agent-completed purchases are coming eventually. The fact that the highest-profile attempt at it was quietly shut down after near-zero sales is an equally strong signal that "eventually" is not "now."
The fundamentals worth investing in today — structured product data, a fast checkout, staying open to AI crawlers — are the same ones that already pay off in conventional and AI-assisted search. If agentic checkout matures on the timeline its backers hope for, that groundwork carries over directly. If it takes longer, you have lost nothing by doing it anyway.
If you want help getting your product or service pages properly structured for both today's search and whatever comes next, that is part of what our Webflow website development service covers.
By Faizyab Ahmed