For most of the web’s history, the buyer was a human pointing a browser at a product page. In 2026, that assumption is breaking down. ChatGPT, Perplexity Comet, Google Gemini and a wave of vertical agents are doing the research, picking the product, and increasingly clicking the checkout button — sometimes via a bespoke protocol, sometimes by literally driving a headless browser. This shift has a name: agentic commerce.
If you sell anything online, this matters. Even if you don’t, the same forces are reshaping how AI agents reach your content, train on it, and decide whether to recommend you. Below is a practical primer on what agentic commerce is, who the players are, and the concrete steps brands can take today.
Table of Contents
What is agentic commerce?
Agentic commerce is the broader pattern of AI agents acting on behalf of users to complete commercial actions — typically across four phases:
- Discovery: an LLM surfaces brands, products, or services in answers and recommendations.
- Comparison: the agent gathers reviews, prices, and specs from multiple sources, often crawling sites in real time.
- Checkout: the agent (or a partnering checkout endpoint) places the order, sometimes via a structured protocol, sometimes by clicking through a normal checkout flow.
- Post-purchase: tracking, returns, and renewals that an agent can manage on a user’s behalf.
What’s new is not that AI is involved — that has been true since recommendation engines. What’s new is that the AI is the buyer’s delegated agent, with permission to spend money, share credentials, and interact with merchants directly.
The protocols racing to standardise it
Three open standards dominate the conversation in 2026:
Agentic Commerce Protocol (ACP)
Co-developed by OpenAI and Stripe, ACP is an open spec for how an AI agent can call a merchant’s checkout endpoint, share payment credentials securely (via a Shared Payment Token), and receive a confirmation. It powers ChatGPT’s Instant Checkout flow and is the protocol most ecommerce builders are integrating first. The spec is open-source under Apache 2.0 and is maintained jointly by OpenAI and Stripe.
Universal Commerce Protocol (UCP)
Google’s open standard for agent-to-merchant interoperability — covering checkout, identity linking, order tracking, and secure payment token exchange. UCP launched with a sizeable partner roster including Walmart, Target, Etsy, Shopify, and others. It’s tightly coupled with Google’s Gemini agent surfaces and Search.
Agent Payments Protocol (AP2) and x402
AP2 is a Google initiative aimed at the payments leg of agentic transactions. x402 is Coinbase’s micropayments-friendly approach for machine-to-machine payments. Both target slightly different use cases (subscriptions and crypto-rail transactions, respectively) but share the assumption that the customer of the future is software.
Why this is happening now
- Capable agents: LLMs are good enough to read product pages, follow checkout flows, and reason about return policies.
- Browser-driving agents: tools like Perplexity Comet effectively log in as the user in a controlled browser, click through checkout, and use saved credentials. No protocol needed — but also no merchant control.
- Distribution: ChatGPT alone has roughly 800M weekly active users. If even a small share of them ask the agent to buy, that’s an enormous new commerce surface.
- Money: McKinsey estimates agentic commerce will drive $3–5 trillion in spend by 2030. Morgan Stanley projects 10–20% of ecommerce ($190–385B) routed through AI agents by the late 2020s.
What this means for brands
Three big shifts:
- Your product page is now an API for agents. The clarity of your titles, the structure of your descriptions, the speed of your pages, and the cleanliness of your structured data all matter more than ever. Agents extract — they don’t browse.
- Reviews and citations are the new SEO links. Agents weigh third-party signals heavily. Showing up in trusted publications and getting cited inside AI answers becomes a primary growth lever.
- You need a new analytics layer. Most analytics products are blind to agents. GA4, Plausible, Mixpanel, PostHog — they all rely on JavaScript that AI bots don’t run. If you only look at your normal analytics dashboards, you’ll undercount agent traffic by an order of magnitude.
How to prepare for agentic commerce: a practical checklist
- Audit your structured data. Make sure product schema, FAQ schema, and Organization schema are clean and complete. Agents lean on this.
- Watch your AI visibility. Track how often your brand is mentioned and cited inside ChatGPT, Perplexity, Gemini, Google AI Mode, and AI Overviews. This is where AI agents start their research.
- Decide your protocol stance. Will you integrate ACP for ChatGPT Instant Checkout? Join UCP via Shopify or Stripe? Allow Perplexity Comet to drive your existing checkout? Each has tradeoffs around control, fees, and customer relationship.
- Manage your bot policy at the edge. Decide which agents you welcome (most), which you rate-limit, and which you block. This requires knowing who is hitting you in the first place.
- Measure agent traffic separately. Treat AI bot traffic as its own funnel — separate from organic, paid, and social. The questions are different (which pages are training, which agents are recommending) and the tools to measure it are different.
The missing layer: agent-aware analytics
The biggest practical blind spot for ecommerce and content teams entering agentic commerce is simple visibility. You can’t optimize what you can’t see, and most teams can’t see AI agent traffic at all.
That’s exactly the gap LLM Pulse Agent Analytics closes. It reads server-side logs (Cloudflare, nginx, Apache, plus integrations on the way for Vercel, Fastly, AWS CloudFront, AWS ALB, Akamai, GCP Load Balancer, Azure Front Door, Netlify, Bunny CDN, KeyCDN), identifies AI bots against a curated catalog (GPTBot, ClaudeBot, PerplexityBot, OAI-SearchBot, Google-Extended, Bytespider, Amazonbot and more), and shows you:
- Total AI agent requests over time
- Share of traffic by company (OpenAI, Anthropic, Google, Perplexity, Meta, ByteDance, Amazon)
- Top bots by volume, with category breakdowns (AI search, AI assistant, AI crawler, reference)
- On Enterprise: per-URL drilldown — which specific pages each AI agent is reading
Agent Analytics is live on the Scale plan (€299/month) for aggregated bot stats, with per-URL drilldown available on Enterprise. Connecting Cloudflare takes about 30 seconds with a read-only API token; CSV upload supports Cloudflare, Apache, nginx and custom log formats.
It pairs naturally with the rest of LLM Pulse AI — brand visibility tracking across ChatGPT, Perplexity, Gemini, Google AI Mode and AI Overviews; sentiment analysis on every mention; citation tracking; share of voice; competitor benchmarking. The same platform that tells you what AI says about you also tells you which AI agents are reading you.
How to start today
- Map your AI visibility. Track your brand and competitors across the AI surfaces where agentic discovery happens. LLM Pulse covers ChatGPT, Perplexity, Gemini, Google AI Mode and Google AI Overviews on every plan, starting at €49/month with a 14-day free trial.
- Turn on Agent Analytics. On Scale or Enterprise, connect Cloudflare or upload a log CSV and start measuring AI agent traffic separately from human traffic.
- Pick a protocol path. Decide whether you want to integrate ACP, UCP, or simply optimize for browser-driving agents like Perplexity Comet.
- Audit your content for agent-readiness. Clean structured data, clear product copy, fast pages, fewer modal interruptions — agents reward simplicity.
FAQ
Is agentic commerce going to replace traditional ecommerce?
Not in a binary sense. Most analysts (McKinsey, Morgan Stanley) expect 10–20% of ecommerce to flow through AI agents by the late 2020s, with the rest still happening through human-driven sites and apps. But that 10–20% is exactly the high-intent slice — research-heavy purchases, repeat buys, and routine reorders.
Do I need to support ACP or UCP to be visible to AI agents?
No. Agents like Perplexity Comet can drive your existing checkout flow without any integration, and ChatGPT can recommend you in answers regardless of whether you’ve integrated Instant Checkout. Protocol integration improves the experience and gives you more control over the transaction; it isn’t a prerequisite for being recommended.
How is agent traffic different from regular bot traffic like Googlebot?
Googlebot crawls to index pages for search results — there’s a long-established playbook for that. AI agents do something more varied: some crawl to train models (GPTBot, ClaudeBot), some crawl to power live answers (OAI-SearchBot, PerplexityBot), and some operate as user-delegated agents (ChatGPT-User, Perplexity Comet) that act in real time on a single user’s behalf. Agent Analytics groups these by category so you can see each behavior separately.
How do I know which AI agents matter most for my business?
Look at two signals together: which agents are citing you in their answers (visible inside LLM Pulse), and which agents are crawling you on the server side (Agent Analytics). The intersection tells you which agents are both reaching you and recommending you — those are the ones worth optimizing for first.
Can I block agents I don’t want?
Yes — at the edge (Cloudflare, robots.txt, custom rules). The tricky part is deciding which to block. A bot that trains a model you don’t like still might be the same bot powering recommendations that drive sales. Start with measurement, then make policy choices once you know the numbers.
How accurate is server-log-based bot detection?
Very accurate for verified bots that publish their IP ranges and user agents (most of the big AI bots do). For UA-spoofing scrapers, you need a layer like Cloudflare Bot Management to disambiguate. Agent Analytics combines Cloudflare’s verifiedBotCategory signal with our own user-agent classifier to maximise accuracy.
Want to see which AI agents are already crawling your site? Try Agent Analytics on the Scale plan, or compare LLM Pulse plans starting at €49/month with a 14-day free trial.
