On this page
What Does Agentic Commerce for Advertisers Actually Mean?
Key Takeaway: Agentic commerce for advertisers means preparing for a buyer who is a model rather than a person. Three protocols — Google's AP2 and UCP, and OpenAI's ACP — now let an agent discover a product, prove the customer authorised the purchase, and complete checkout without anyone opening your site. Your feed becomes your storefront.
Between September 2025 and January 2026, three competing standards launched to solve the same problem: how does an AI agent buy something on your behalf without anyone getting defrauded?
It is tempting to file this under "interesting, not yet urgent." The volume numbers support that instinct. AI-agent shopping is forecast at roughly 1.5% of US ecommerce in 2026 — around $20.9 billion — which is about four times its 2025 level but still a rounding error next to the whole. Bain's projection puts it at 10–25% of US ecommerce by 2030.
But the standards war is happening now, and standards harden fast. The advertisers who understand the mechanics this year will be the ones whose products are reachable when the volume arrives. This post covers what each protocol actually does, why the platforms are fighting over it, and the four preparations worth making regardless of which one wins.
Three Protocols, Three Different Bets
They are frequently discussed as competitors. They are only partly that — each governs a different stage of the same transaction, and they overlap unevenly.
Protocol | From | Launched | What it governs |
|---|---|---|---|
AP2 | Sept 2025 | Authorisation and liability. Signed mandates prove a human approved the purchase, so the payment network knows who is accountable. | |
UCP | Jan 2026 | Product discovery and catalog exchange. Live across Google AI Mode, Gemini and, since May 2026, YouTube Shopping. Integrates via MCP. | |
ACP | OpenAI, with Stripe | 2025–26 | End-to-end checkout inside ChatGPT — "Instant Checkout". OpenAI takes a fee on completed purchases. |
AP2 launched with more than 60 partners including Mastercard, PayPal, American Express, Coinbase and Salesforce. UCP was announced at NRF with Shopify, Etsy, Wayfair, Target and Walmart, and endorsed by Visa, Mastercard, Best Buy and Home Depot. ACP went live with Etsy, then Shopify merchants including Glossier, SKIMS, Spanx and Vuori.
Those partner lists are the actual signal. Payment networks and major retailers do not co-sign specifications speculatively.
Key Takeaway: UCP integrating via MCP is the detail most coverage skips. It means the same protocol layer connecting your ad accounts to an AI assistant is the one being used to expose product catalogs to shopping agents. If you already run a marketing MCP, you are closer to this than you think.
How Does an Agent Prove You Agreed?
The hard problem in agentic commerce is not technical. It is legal: if an agent buys the wrong thing, who pays?
AP2's answer is signed mandates. The customer's intent is captured and cryptographically signed, then the cart contents, then the payment authorisation. Each step produces a verifiable record showing a real human approved this specific purchase within these specific limits.
That chain is what makes agent purchasing insurable. Without it, no payment network would accept transactions where the buyer never saw a checkout page, because there would be no way to distinguish a legitimate agent purchase from a compromised one.
For advertisers the mandate chain matters for a practical reason: it produces a cleaner record of purchase intent than a browser session ever did. Whether that data becomes available to you, and in what form, is unresolved — and worth asking your platform reps about now rather than after the fact.
Why Are the Platforms Fighting About This?
In late 2025 Amazon sued Perplexity to stop its Comet browser agent shopping on Amazon on users' behalf. A federal court granted a temporary hold; as of March 2026 Perplexity was petitioning to have it lifted.
Read the commercial logic rather than the legal filings. An agent that goes straight to the right product does not scroll past sponsored listings. It does not see a display banner. It does not get retargeted. Agentic shopping routes around the placements that fund the marketplace.
Amazon's advertising business is enormous. Google's is larger. Both have every incentive to ensure agent traffic flows through surfaces they control and monetise — which is precisely what UCP being live in AI Mode, Gemini and YouTube Shopping accomplishes, and what Amazon's own Sponsored Prompts placements are designed to do.
So the fight is not really about whether agents can shop. It is about whose agent, on whose surface, with whose ads alongside.
Honest limitation: This is genuinely early. Protocol adoption could consolidate, fragment, or stall — and the volume forecasts above are forecasts, not measurements. Anyone selling you an "agentic commerce strategy" today is selling you preparation, not performance. The preparations below are worth doing because they pay off regardless of which protocol wins, not because we know which one will.
What Should You Actually Do Now?
Four preparations, all unglamorous, all useful even if agentic shopping stays at 1.5% forever.
Treat your product feed as your storefront. When the buyer is a model, your feed is your product page — there is no photography, no layout, no persuasive copy doing work. Attribute completeness, accurate categorisation and clean titles stop being hygiene and start being the entire shelf. Audit Merchant Center disapprovals and missing attributes now.
Make structured data comprehensive. Price, availability, shipping, returns and variant relationships all need to be machine-readable and correct. An agent comparing options cannot infer from a photo what a human infers instantly.
Fix inventory accuracy. An agent that completes a purchase for an out-of-stock item creates a refund, a support ticket and a trust problem. Feed-to-warehouse accuracy matters more when nobody eyeballs the listing first.
Decide how you will measure it. Agent-completed purchases may attribute strangely or not at all. Establish a baseline of your current channel mix now, so that when a slice of revenue starts arriving through a surface your ad platforms cannot see, you notice the shape of the gap rather than just the shortfall.
That last one connects to a problem that already exists. Platform-reported attribution is imperfect today; a buyer who never touches your site makes it materially worse.
Read this next → Amazon Sponsored Prompts: Ads Inside Alexa for Shopping
The first concrete example of an ad placement built specifically for conversational, agent-mediated shopping — how it works, what it costs, and how to tell whether it's earning anything.
The Uncomfortable Question
If agents mediate a meaningful share of purchases, what happens to advertising as a discipline?
The pessimistic reading: brand persuasion stops mattering, everything collapses into specification-matching, and the winner is whoever has the cleanest feed and lowest price. The optimistic reading: agents still need to select between comparable options, and brand, reviews, return policy and fulfilment speed all remain inputs to that decision — they just get evaluated more consistently than a distracted human would evaluate them.
Both readings are plausible and neither is settled. What is safe to say is that the inputs are shifting from persuasion toward verifiable fact. A model does not respond to a compelling headline; it responds to complete, accurate, well-structured data about what you sell and how reliably you deliver it.
Which is a strange kind of good news for advertisers who have been doing the unglamorous work all along.
What Should You Take from This?
Three protocols now exist: AP2 for payment authorisation, UCP for catalog and discovery, ACP for checkout inside ChatGPT. They govern different stages, not identical ground.
UCP integrates via MCP, which puts agentic commerce on the same protocol layer as marketing data access.
Platform resistance — including Amazon's litigation against Perplexity — is about ad revenue, not technology. Agent shopping bypasses sponsored placements.
The preparations that matter are feed completeness, structured data, inventory accuracy and measurement baselines. All four pay off whether or not agentic volume materialises.
Know Your Baseline Before the Mix Shifts
Brandlio connects your ad platforms, store, payments and Merchant Center to Claude or ChatGPT, so feed health, inventory and channel mix are one prompt rather than four exports. See pricing or connect your first account.
Frequently Asked Questions
What is agentic commerce?
Agentic commerce is online purchasing where an AI agent completes the transaction on a person's behalf rather than the person browsing and checking out themselves. The customer expresses intent, the agent finds the product, and a protocol handles authorisation and payment. It requires new standards because payment networks need proof that a human authorised the specific purchase.
What is Google's AP2 protocol?
AP2, the Agent Payments Protocol, was announced by Google in September 2025 with more than 60 partners including Mastercard, PayPal, American Express and Coinbase. It uses cryptographically signed Intent, Cart and Payment mandates so an agent can prove a real person authorised a specific purchase within specific limits, which is what makes agent transactions insurable.
What is the Universal Commerce Protocol?
UCP is Google's product discovery and catalog exchange standard, announced in January 2026 at NRF with Shopify, Etsy, Wayfair, Target and Walmart. It is live across Google AI Mode, Gemini and, since May 2026, YouTube Shopping. Notably it integrates via the Model Context Protocol, the same layer used to connect marketing data to AI assistants.
How is OpenAI's Agentic Commerce Protocol different?
ACP, developed with Stripe, powers Instant Checkout inside ChatGPT — the purchase completes in the conversation rather than on the merchant's site. It launched with Etsy and then Shopify merchants, and OpenAI takes a fee on completed purchases. Where AP2 governs authorisation and UCP governs discovery, ACP covers the checkout itself.
Why did Amazon sue Perplexity over shopping agents?
Amazon sought to block Perplexity's Comet browser agent from shopping on Amazon on users' behalf, and a federal court granted a temporary hold. The commercial logic is straightforward: an agent that navigates directly to a product does not scroll past sponsored listings, so agent-mediated shopping bypasses the placements that fund the marketplace.
How big is agentic shopping actually?
Forecasts put AI-agent shopping at roughly 1.5% of US ecommerce in 2026, around $20.9 billion, which is about four times the 2025 level. Bain projects 10–25% of US ecommerce by 2030. These are forecasts rather than measured results, so treat them as directional. The standards work is further along than the volume.
What should advertisers do to prepare?
Four things, all of which pay off regardless of how agentic commerce develops: make your product feed complete and accurately categorised, since it becomes the storefront when the buyer is a model; ensure price, availability and variant data are machine-readable; fix inventory accuracy so agents do not complete purchases for unavailable items; and baseline your current channel mix so you can spot revenue arriving through surfaces your ad platforms cannot see.
Does agentic commerce make advertising irrelevant?
Unlikely, but it changes the inputs. Agents still select between comparable options, and brand, reviews, return policy and fulfilment speed remain factors — they simply get evaluated more consistently than a distracted human would evaluate them. The shift is from persuasion toward verifiable fact, which favours advertisers who have kept their data clean.
