Agentic commerce is shopping done by an AI agent on a person's behalf. Instead of a shopper searching, opening ten tabs, comparing and checking out, they tell an assistant what they want and the agent finds options, narrows them down and, increasingly, completes the purchase.
The term has been around for a while, but September 2026 is when it stopped being a demo. Meta launched an agent that can buy things. Shopify switched AI checkout on by default for eligible US stores in Google's AI Mode and Gemini. Amazon was reported to have blocked Meta's agent. This guide explains what agentic commerce means, who is building it, and what a brand should do about it.
Definition
A working definition: agentic commerce is any purchase journey where an AI system, not the human, does the searching, comparing or transacting. It sits on a spectrum:
| Level | What the AI does | Example (as of Sept 2026) |
|---|---|---|
| Assisted discovery | Recommends products; shopper clicks out to buy | ChatGPT product answers after OpenAI dropped Instant Checkout |
| Native checkout | Recommends and lets the shopper pay without leaving the AI | Google AI Mode and Gemini checkout for eligible Shopify stores |
| Delegated purchase | Acts in a browser, fills forms, pays | Meta's Muse using Link by Stripe |
Every level shares one feature: the AI decides which products the shopper sees. At the far end, the shopper may never see a list at all.
How is this different from ordinary e-commerce? In a normal online shop, the customer does the work. They type a search, scan results, open product pages, read reviews and decide. Your job as a merchant is to rank, attract the click and convert the visit. In agentic commerce, much of that work moves to software. The agent reads the product pages and reviews, weighs the options against what the shopper said they wanted, and presents one or a few choices. The "visit" to your site may be an agent reading your page in a fraction of a second, with no human ever seeing your homepage. That changes what good merchandising looks like: clear, specific, machine-readable facts matter more than persuasive design.
Who is building it
Four companies shaped the market this month.
Meta. On 8 September 2026 Meta introduced Muse, a personal AI agent in the US that can open a browser, fill in forms and buy things using Link by Stripe, with Shop Pay listed as "coming soon" (Meta, 8 September 2026).
Shopify. Shopify and Meta then partnered to bring Shop Pay checkout to Muse. PYMNTS reported that Shop Pay has passed $400 billion in lifetime accelerated GMV, and that Shopify says shoppers who start in an AI chatbot are 2.5 times more likely to reach a product page, with roughly an 80% conversion uplift. Those are Shopify's own figures, not independent measurements (PYMNTS, 22 September 2026).
Google. Search Engine Land reported that eligible US Shopify stores are being automatically enrolled in native checkout inside Google AI Mode and the Gemini app (Search Engine Land, 22 September 2026). Google says its Universal Commerce Protocol already powers direct checkout for "hundreds of thousands of brands and retailers" (Google, 16 September 2026).
OpenAI. OpenAI went the other way. In March it dropped Instant Checkout in ChatGPT, letting merchants use their own checkout so it could "focus our efforts on product discovery" (CNBC, 24 March 2026). ChatGPT still shapes what people buy. It just hands the payment step back to the merchant. Our post on ChatGPT shopping after Instant Checkout covers what replaced it.
Walled gardens vs open rails
There are two competing visions, and September showed both.
Open rails are shared protocols that any agent and any merchant can use. Google's Universal Commerce Protocol (UCP) lists Google, Shopify, Etsy, Wayfair, Target, Walmart, Amazon, Microsoft, Meta, Salesforce and Stripe as co-developers (ucp.dev, accessed September 2026). The Agentic Commerce Protocol (ACP), developed by Stripe and OpenAI, is open source under Apache 2.0 (agenticcommerce.dev, accessed September 2026). We compare them in Google UCP vs ACP vs AP2.
Walled gardens are platforms deciding which agents may shop on them. GeekWire reported, around 21 September 2026, that Amazon blocked Meta's Muse from Amazon.com (GeekWire). Amazon appears on the UCP co-developer list at the same time. Supporting an open standard and letting a rival's agent loose on your own store are clearly different decisions.
For merchants, the lesson is that there will not be one agentic checkout. There will be several, with overlapping rules, and each will change. Plan for plurality rather than picking a winner.
Why recommendation comes before checkout
Most agentic commerce coverage focuses on the payment step, because that is where the protocols and partnerships are. But the payment step is the last step. Before any agent pays, it has to decide what to buy.
Think about what a delegated purchase looks like. A shopper says "get me a waterproof hiking jacket under $200 that fits over a fleece." The agent searches, reads product pages and reviews, compares a handful of options and picks one. If your jacket was not in that shortlist, the quality of your checkout integration is irrelevant.
That is why OpenAI's March decision is telling. It chose discovery over checkout. And it is why Google is pushing merchants to improve product data: in lululemon tests, Google said conversational attributes submitted by the brand were used 50% of the time in relevant AI Mode product recommendations (Google, 16 September 2026).
Put simply:
- Recommendation decides whether you are considered.
- Checkout decides how smoothly the sale completes.
Most brands are working on step two. Step one is where the sale is won or lost.
What merchants should do now
A practical list for September 2026:
- Know what is switched on. If you are on Shopify in the US, check whether your store has been enrolled in Google AI Mode and Gemini checkout. Our Shopify agentic storefronts guide walks through the settings.
- Fix your product data. Agents read structured data. Complete titles, specific attributes, accurate prices and stock, and clear return policies all help an agent match your product to a request.
- Keep your own site readable. Agents and AI answers still pull from your product pages, FAQs and policies. Clear, specific copy beats vague brand language.
- Earn third-party mentions. Reviews, comparison articles and forum discussions are part of what AI systems use to decide which brands to recommend.
- Decide your stance on native checkout. Native checkout may convert well, but it also moves part of the experience off your site. Make a deliberate choice rather than accepting a default.
- Watch the protocols, don't chase them. If you are on Shopify or another large platform, much of the protocol work is handled for you. Custom builds can wait until the picture settles.
How to measure whether agents recommend you
No dashboard yet tells you how often an AI agent chose your product over a competitor's. What you can measure is the upstream signal: when people ask AI assistants for recommendations in your category, are you named?
Google now offers part of this. Merchant Center's AI performance insights show share of voice across AI Mode and AI Overviews for businesses in the US, Australia, Canada, India and New Zealand (Google, 16 September 2026). That covers Google's surfaces only.
For the other engines, run structured tests: ask the questions your buyers ask, across ChatGPT, Claude, Perplexity and Google AI Overviews, and record who is recommended. Citedify does this monthly. It asks 20 buyer-intent questions about your brand across those engines, scores each answer for mention, position, sentiment and recommendation strength, and shows which competitors are named instead of you and which sources the AI cited. It does not track Meta Muse or complete purchases. It tells you whether you are in the shortlist that agents and AI answers draw from. See e-commerce AI visibility and AI search optimization for more on the approach, and ChatGPT Shopping for how product answers work in ChatGPT.
Key takeaways
- Agentic commerce means AI agents doing the searching, comparing or buying for a shopper.
- In September 2026 Meta launched Muse, Shopify turned on AI checkout by default for eligible US stores in Google AI Mode and Gemini, and GeekWire reported Amazon blocked Muse.
- OpenAI dropped in-chat checkout in March to focus on product discovery.
- Open protocols (UCP, ACP) and walled gardens are developing side by side. Expect several agentic checkouts, not one.
- The agent picks the product before it pays. Being recommended is the first problem to solve.
- Measure recommendation share across engines, not just checkout settings.
Ready to see where you stand? Get your AI Visibility Audit — track your AI visibility score, competitor share, and a prioritized action plan every month.
