Agentic Commerce Protocol: How AI Agents Are Already Buying on the Internet — and What You Need to Do Now

Published on — Updated on — By

# Agentic Commerce Protocol: How AI Agents Are Already Buying on the Internet — and What You Need to Do Now Imagine asking ChatGPT: "Buy a pet sunscreen with free shipping, up to R$ 80, delivery by Friday." And it simply does it. No opening a new tab, no price comparison, no filling out a form. The agent researches, compares, chooses and completes the purchase — all within the chat. This is not science fiction. It's happening now, in 2026, in the United States. And the protocol that makes this possible — the **Agentic Commerce Protocol (ACP)** — is already open for merchants worldwide to prepare. This post explains what ACP is, how it works in practice, which platforms have already adopted it, and — most importantly — what you need to do today so your site appears (and gets purchased) by AI agents. --- ## What the Agentic Commerce Protocol Is The Agentic Commerce Protocol (ACP) is an open standard of communication between artificial intelligence agents and e-commerce systems. In simple terms: it's the language that allows an AI to "talk" to a merchant's store to check availability, price, process a payment and confirm an order — without human intervention. The ACP was co-developed by OpenAI in partnership with Stripe and publicly released in September 2025, alongside ChatGPT's **Instant Checkout** feature. OpenAI made it open source so any platform, payment processor or merchant can adopt it without relying on a single company. > "The Agentic Commerce Protocol provides the language that allows AI agents and businesses to work together to complete a purchase for a user." > — OpenAI, September 2025 At the same time, Google launched in January 2026 the **Universal Commerce Protocol (UCP)**, its own open standard for agentic commerce, co-developed with Shopify, Etsy, Wayfair, Target, Walmart, Mastercard, Visa, Stripe and more than 20 other companies in the ecosystem. The UCP is compatible with OpenAI's ACP, the Model Context Protocol (MCP) and Agent2Agent (A2A) — forming an ecosystem of interoperable protocols. --- ## The Numbers That Explain the Urgency Before diving into technical details, you need to understand the scale of what's happening. The data is unequivocal: | Metric | Data | Source | |---|---|---| | AI traffic growth to retail (Jul 2025) | **+4,700% year over year** | Adobe Analytics | | Global sales influenced by AI on Black Friday 2025 | **US$ 14.2 billion** | Adobe Analytics | | Consumers who want AI assistants for shopping | **76%** | Capgemini Research | | Retailers with AI implemented or in development | **97%** | Salesforce | | AI traffic growth on Black Friday 2025 | **+805% year over year** | Adobe Analytics | | Conversion lift for buyers coming from AI | **+38%** | Adobe Analytics | | LLM traffic to e-commerce originating from ChatGPT | **~97%** | Kaiser & Schulze, SSRN | | Projection of global sales via AI agents by 2030 | **25% of global e-commerce** | Deloitte | The most revealing figure may be the conversion lift: visitors who arrive at your site from an AI recommendation convert **38% more** than regular organic traffic. They already arrive convinced — the AI did the research and comparison for them. --- ## How It Works in Practice: The Flow of an Agent-Mediated Purchase To understand what needs to be done on your site, it's useful to visualize the full flow of an agentic transaction: **1. User intent** — The user tells the agent (ChatGPT, Perplexity, Gemini) what they want to buy, with constraints (price, delivery time, preferences). **2. Discovery** — The agent crawls the web looking for relevant products. It reads structured data (Schema.org), product feeds and semantic content. Sites without proper structure simply don't appear at this stage. **3. Comparison and ranking** — The agent evaluates price, availability, ratings, semantic relevance and — in ChatGPT's case — whether the merchant has Instant Checkout active. The ranking is organic, not paid. **4. Presentation to the user** — The agent displays cards with image, price, store and purchase option. The user confirms the choice. **5. Checkout** — If the merchant is integrated with ACP/UCP, the checkout happens inside the chat. The agent sends the necessary data to the merchant's backend via a secure protocol. The merchant processes the payment normally and handles fulfillment. **6. Post-purchase** — Confirmation, tracking and support remain with the merchant. The AI only mediated the transaction. The critical point: **the AI does not browse your site like a human**. It doesn't click buttons, doesn't fill forms, doesn't read CSS. It consumes structured data and APIs. If your site doesn't expose that information in a machine-readable way, you are invisible to this channel. --- ## Who Is Already Integrated The ecosystem is forming quickly. In September 2025, OpenAI launched Instant Checkout with Etsy and more than one million Shopify merchants (including Glossier, SKIMS, Spanx and Vuori). In November 2025, Perplexity launched its own shopping agent in partnership with PayPal, enabling direct purchases from over 5,000 merchants. Google, in January 2026, enabled direct checkout in product listings in Search's AI Mode and in Gemini, using UCP. Companies like Petco, Lowe's, Reebok, Poshmark and Macy's are already integrated with Google's Business Agent. For merchants outside Shopify and Etsy, OpenAI maintains a waitlist at [chatgpt.com/merchants](https://chatgpt.com/merchants), where you can register a product feed to appear in ChatGPT searches — even without Instant Checkout active. --- ## What You Need to Do: The Four Layers of Preparation The good news is most of the preparation work is technical and can be done progressively. Here are the four layers, in order of immediate impact: ### Layer 1 — Structured Data (Schema.org) This is the most critical layer and the one that most affects visibility today. Every product page needs JSON-LD markup with: - `@type: Product` with `name`, `description`, `image`, `brand`, `sku` - `Offer` with `price`, `priceCurrency`, `availability`, `url`, `shippingDetails` - `AggregateRating` with real customer reviews - `FAQPage` for frequently asked questions about the product This data is what AI agents read when crawling your site. Without it, the agent cannot extract reliable information about price, availability, or product features — and simply ignores the page. ### Layer 2 — The `/llms.txt` File Just like `robots.txt` instructs search crawlers, `llms.txt` instructs AI agents about what your site is, which pages are relevant and how to interpret your content. The standard was proposed in 2024 and is being progressively adopted by sites that want to maximize their readability by LLMs. A basic `llms.txt` for a pet products e-commerce might be: ``` # Vetfleur — Aromatherapy and Natural Wellness for Pets > Natural aromatherapy products for dogs and cats. Focus on wellness, > without artificial ingredients. Delivery throughout Brazil. ## Products - [Natural Bioprotector](/produtos/bioprotetor): natural sunscreen for pets - [Relaxing Oil](/produtos/oleo-relaxante): aromatic blend for canine anxiety ## Policies - [Shipping & Delivery](/frete): times and conditions - [Exchanges & Returns](/devolucoes): full policy ``` ### Layer 3 — Product Feed The product feed is a structured file (usually in the Google Merchant Center format, which you likely already have if you use Google Shopping) that lists the entire catalog with prices, images, availability and attributes. This feed is what ChatGPT and Perplexity use to display your products in search results. If you use Shopify, you are already automatically integrated with ChatGPT Shopping. For other platforms, you can register directly at [chatgpt.com/merchants](https://chatgpt.com/merchants). ### Layer 4 — Rich Semantic Content AI agents cite authoritative sources on a topic. A well-written article about "aromatherapy for pets: benefits and precautions" is much more likely to be cited in ChatGPT's responses than a generic product page. This is what we call **GEO (Generative Engine Optimization)** — the practice of structuring content to be understood, indexed and cited by generative systems. --- ## The Opportunity of Exclusive Offers by AI Channel One of the most interesting questions in this context is: "Can I create a channel-exclusive condition for users coming via AI?" The answer is yes — and there are three practical ways to do this today: **UTM parameters in the product URL.** When ChatGPT or Perplexity displays your product and the user clicks, you can set the URL in the feed/schema with `?utm_source=chatgpt`. On your site, detect that parameter and show a banner or special discount. Simple, effective and implementable today. **Dedicated landing page for AI traffic.** Create a clean URL like `/ai/product-name` that you use in Schema.org and in the feed. This page has copy optimized for users coming from an AI recommendation — already convinced, needing only convenience — and a direct CTA to checkout. **Direct Offers in Google.** Google launched in January 2026 the **Direct Offers** feature, which allows merchants to set up discounts to appear in Search's AI Mode. Imagine a user searching for "easy-to-clean rug for dining room" and receiving a 20% discount directly in Gemini's answer. That's the channel. What does not yet exist natively is a mechanism for the AI to negotiate price in real time with the merchant. That's on ACP's roadmap, but it's not available yet. What exists today is the ability to **pre-configure offers** that the AI can present to the user at the right moment. --- ## What This Means for SEO 5.0 The Agentic Commerce Protocol is not a threat to SEO — it's its natural evolution. The same practices that make a site perform well on Google (technical structure, semantic content, authority, structured data) are the ones that make a site perform well with AI agents. The difference is that traditional SEO optimizes for an algorithm that ranks pages. SEO 5.0 optimizes for intelligent systems that **understand context, intent and trustworthiness** — and make purchasing decisions on behalf of users. The IAS™ Score of SEO 5.0 already measures 10 dimensions that directly affect this visibility: technical infrastructure, semantic organization, contextual presence, coherence of messaging, interaction with AI, multichannel presence, digital reputation, E-E-A-T, GEO Score and presence in the Common Crawl (the training datasets for AIs). A site with a high IAS™ score is, by definition, better prepared to be found, cited and bought by AI agents. --- ## Where to Start: Practical Checklist If you want to start preparing your site for agentic commerce today, here is the most direct path: | Action | Urgency | Complexity | Impact | |---|---|---|---| | Implement Schema.org Product on all product pages | High | Low | High | | Create `/llms.txt` on the domain | High | Very low | Medium | | Register product feed with ChatGPT Merchants | High | Medium | High | | Configure UTM `?utm_source=chatgpt` in feed URLs | Medium | Very low | Medium | | Create dedicated landing page for AI traffic | Medium | Low | Medium | | Produce rich semantic content about the niche | High | High | High | | Conduct IAS™ diagnostic to identify gaps | High | Very low | High | The IAS™ diagnostic is the recommended starting point because it identifies exactly which of the 10 dimensions are compromising your visibility — both on Google and on AI agents. In 2 minutes you get a score and an action plan. --- ## Conclusion: The Window of Opportunity Is Open Agentic commerce is not a future trend — it's a present reality. AI traffic to retail sites grew 4,700% in one year. buyers who arrive via AI convert 38% more. And 25% of all global e-commerce is expected to be mediated by agents by 2030. The window of opportunity for those who act now is significant: most sites are still unprepared. Only 16% of brands systematically track their performance in AI searches. That means those who implement the preparation layers in the coming months will occupy positions that will be much harder to win later. The Agentic Commerce Protocol is the infrastructure. SEO 5.0 is the strategy to use it. --- *Want to know if your site is ready to be found by AI agents? Take the free IAS™ diagnostic in 2 minutes — no signup, no meeting.* --- > **E-commerce architecture for AI** — see how Product Schema, llms.txt and product pages help AI agents understand your store: [explore the strategy for e-commerce](/para-ecommerce).