How Agentic Commerce Works
From shopper intent to completed purchase — a full breakdown of the technology stack, protocols, and business model behind agentic commerce in 2026.
The Agentic Commerce Stack
Agentic commerce is built on four interconnected layers. Understanding each one reveals how AI agents go from understanding a shopper's intent to completing a purchase.
Intent Layer — Natural Language Understanding
The top of the stack is the AI's ability to understand what a shopper actually wants — regardless of how they phrase it. This is powered by large language models (LLMs) like those behind ChatGPT, Gemini, and Copilot, which can interpret ambiguous, complex, and multi-part requests.
- "Find me waterproof hiking boots for wide feet under $200" → understood perfectly
- Multi-turn conversation: the AI remembers context across follow-up questions
- Disambiguation: AI asks clarifying questions when intent is unclear
- Publisher context: AI reads editorial content to understand what products are being discussed
Product Intelligence Layer — Discovery & Matching
Once intent is understood, the agentic commerce engine searches for matching products. Unlike traditional keyword search, this layer understands semantic meaning — 'reliable brand' isn't a database field, it's an emergent property from reviews, return rates, and reputation signals.
- Product catalog ingestion with full attribute data (specs, variants, pricing, inventory)
- Semantic matching against shopper intent rather than keyword matching
- Real-time inventory and pricing verification
- Cross-network affiliate link selection for publishers (15+ networks)
Commerce Protocol Layer — Checkout & Transactions
This is where the 'agentic' part happens — the AI agent doesn't just recommend, it can act. Two competing open standards have emerged to enable secure AI-driven transactions.
- ACP (Agentic Commerce Protocol): OpenAI + Stripe standard, powers ChatGPT & Copilot checkout
- UCP (Universal Commerce Protocol): Google's protocol for AI Mode and Gemini shopping
- Shared Payment Tokens: Stripe handles credentials — AI never sees actual card numbers
- Webhook-based confirmation: orders are verified and tracked in real time
Monetization Layer — Revenue Distribution
The final layer routes revenue to the right parties. In a two-sided network like Brambles.ai, this includes both publisher earnings (affiliate, CPC, retail media) and retailer attribution, all tracked and distributed through Stripe.
- Affiliate commissions: tracked per-click and per-purchase across 15+ networks
- CPC revenue: pay-per-click contextual ads served within AI responses
- Retail media: brand-sponsored product placement in publisher AI results
- Direct checkout: retailer revenue from purchases completed inside AI conversations
How It Works for Publishers vs. Retailers
For Publishers
- 1Publisher installs WordPress plugin or JavaScript snippet (2 minutes)
- 2Brambles.ai AI crawls and indexes all editorial content
- 3AI agent appears on site as a floating chat or inline embed
- 4Reader asks a product question — AI matches to relevant products from 15+ affiliate networks
- 5Reader clicks or purchases — publisher earns affiliate commission, CPC fee, or retail media revenue
- 6Revenue flows to Stripe account automatically
For Retailers
- 1Retailer connects product catalog via API or data feed
- 2AI indexes full catalog: names, descriptions, specs, inventory, images, variants
- 3AI widget deploys to retailer site with custom branding
- 4Shopper asks a question — AI searches catalog and returns personalized recommendations
- 5Shopper tries on (virtual try-on), views in room, gets service questions answered — all in chat
- 6Shopper adds to cart or checks out directly from conversation
The Open Standards Powering
Agentic Checkout
Two competing protocols have emerged to enable AI agents to securely complete purchases. Understanding both helps retailers choose the right integration path.
Agentic Commerce Protocol
By OpenAI + Stripe
Open-source protocol under Apache 2.0. Uses Stripe Shared Payment Tokens so AI agents can initiate checkout without ever handling actual card numbers.
Powers
Universal Commerce Protocol
By Google
Google's proprietary protocol with broader scope than ACP — covering discovery, not just checkout. Integrates with Google Pay. No transaction fee currently charged to merchants.
Powers
Why the Two-Sided Network
Changes Everything
Most agentic commerce platforms solve for one side: either publishers or retailers. Brambles.ai is the only platform that powers both — creating a network effect that benefits everyone.
Unified Platform
One platform. One integration. Whether you're a publisher earning commissions or a retailer increasing conversions — same codebase, same AI engine.
Network Effects
Publishers bring reader traffic. Retailers bring product catalogs. The more of both join, the smarter recommendations become for everyone.
Retailer Discovery via Publishers
Retailers listed in Brambles.ai can be discovered and recommended across our entire publisher network — not just on their own site.
Technical Questions Answered
How does agentic commerce work technically?
Agentic commerce works through a layered system: an AI language model interprets shopper intent in natural language, a product intelligence engine matches that intent against a retailer's catalog, and a commerce protocol (like ACP or UCP) handles the transactional layer. The AI can operate across publisher sites and retail stores simultaneously, with all monetization flowing through a unified platform.
What is the difference between ACP and UCP?
ACP (Agentic Commerce Protocol) is an open standard built by OpenAI and Stripe that powers checkout within ChatGPT and Microsoft Copilot. UCP (Universal Commerce Protocol) is Google's own protocol, built independently to power agentic shopping within Google AI Mode and Gemini. Both allow AI agents to initiate and complete purchases, but they operate through different payment and authentication mechanisms.
Does agentic commerce require changes to my website?
For publishers, agentic commerce requires only a single JavaScript snippet or WordPress plugin — no structural changes to your site. For retailers, you need to connect your product catalog via API or data feed, then deploy the AI widget. The entire process typically takes 24–48 hours.
How does the AI know what products to recommend?
For publishers, the AI reads and indexes your editorial content to understand what products you cover. For retailers, it ingests your product catalog with full attribute data. In both cases, the AI uses natural language understanding to match shopper intent against available product data in real time.
Ready to Deploy Agentic Commerce?
Publishers and retailers — get started with the only platform that powers both sides of agentic commerce.