ORIGINAL RESEARCH

The Guide to
Agentic Commerce

AI agents are starting to shop. Is your store ready?

Based on scanning 1,000+ e-commerce brands

CHAPTER 01

AI agents are now shopping
on behalf of consumers

This is not a future prediction. It is happening right now. AI shopping agents are live, processing millions of product queries every week, and making purchase decisions for real consumers.

84MPER WEEK

shopping queries processed by ChatGPT alone. That's 84 million times per week a consumer asks an AI agent to find, compare, or buy a product — instead of going to your site directly.

10 major AI shopping agents, split into two types

Understanding the difference between these two types is the key to understanding agentic commerce.

Feed-Based AgentsREAD YOUR DATA

These agents read your product data through APIs, feeds, and structured markup. They never open a browser. They need machine-readable data to discover and recommend your products.

ChatGPT Shopping
Google AI Mode
Perplexity Shopping
Microsoft Copilot
Klarna AI
Browser-Based AgentsUSE YOUR SITE

These agents open a real browser, see the screen with AI vision, click buttons, and fill forms. They interact with your site like a human would — except they're software.

ChatGPT Operator
Amazon Buy For Me
Perplexity Comet
Claude Computer Use
OpenClaw

This isn't future speculation — these agents are live today.

Every one of these 10 agents is publicly available and actively processing shopping queries right now. The question is whether your store is ready for them.

CHAPTER 02

How AI agents see your site

There are five layers between an AI shopping agent and your products. Each one can block, confuse, or welcome the agent. Most brands don't know these layers exist.

01

robots.txt

THE SIGN ON YOUR DOOR

A public text file on every website that tells bots what they're allowed to access. Think of it as a sign on your shop door: "Humans welcome. GPTBot — stay out." Every website can have one at yoursite.com/robots.txt. Most CMOs don't know it exists.

Go to yoursite.com/robots.txt right now. You might be surprised by what you find.

02

Server-Level Blocking

THE INVISIBLE BOUNCER

Even if your robots.txt says "welcome," your CDN — the network that delivers your site globally, companies like Akamai, Cloudflare — might block AI agents automatically. Your security team turned on "bot protection" and didn't realize it blocks AI shopping agents too.

This is the most common accidental blocker we see across 1,000+ brands.

03

Structured Data

THE PRODUCT TRANSLATOR

Invisible code on your product pages that tells AI agents: this is a product, it costs $89, it's in stock, here are the sizes. Without it, AI agents can see your page but can't understand what they're looking at. Uses a standard called Schema.org.

Only 25% of brands we scanned have complete structured product data.

04

Product Feeds

THE CATALOG FILE

XML or JSON files listing your entire product catalog. Google Merchant Center feeds, Shopify's products.json, RSS feeds. Feed-based AI agents need these to discover and recommend your products.

Feed-based agents like ChatGPT Shopping and Google AI Mode rely entirely on these.

05

The Visual Experience

WHAT AI EYES SEE

Computer use agents see your site the same way a human does — through screenshots. If your "Add to Cart" button is hard to find for a human, it's hard to find for an AI agent too.

Browser agents like Operator and Comet literally take screenshots and click on what they see.

CHAPTER 03
ORIGINAL RESEARCH

The Data: 1,000+ Brands Tracked

We sent AI agents to every major e-commerce brand we could find. Here's what we learned.

95% of e-commerce brands don't explicitly block AI agents in their robots.txt.

But many are accidentally invisible due to missing structured data, broken feeds, or server-level blocking. The door is open, but the lights are off.

37brands with structured data markup
38brands blocking GPTBot specifically
96%brands fully open to AI agents
1,000+total brands tracked

Three brand archetypes

After scanning 1,000+ brands, clear patterns emerge. Most brands fall into one of three categories.

The Open DoorGLOSSIER MODEL
73GRADE B

Welcomes all AI agents. Has structured data, product feeds, clean checkout. AI agents can find, understand, and attempt to buy Glossier products.

Robots.txt: OpenStructured Data: CompleteProduct Feeds: ActiveGuest Checkout: Yes
The Locked GateAMAZON MODEL
34GRADE D

Explicitly blocks 40+ AI bot user agents in robots.txt. A strategic business decision to protect their product data. Amazon blocks every AI shopping agent from reading their catalog — while sending their own Amazonbot to read everyone else's.

Robots.txt: 40+ BlocksStrategic DecisionProtects Product DataSends Own Bot
The Accidental BlockerNIKE MODEL
67GRADE C

Robots.txt doesn't block AI agents, but the Akamai CDN returns 403 Forbidden to every AI user agent. Nike's security team is inadvertently preventing AI-driven discovery of Nike products. They may not even know.

Robots.txt: OpenCDN: Blocking403 to AI AgentsAccidental
Feed-based agents

ChatGPT Shopping, Google AI Mode, and other feed agents are most affected by data gaps— missing structured data, broken product feeds, and incomplete catalog information.

Browser-based agents

Operator, Comet, and other browser agents are most affected by checkout friction— login walls, ambiguous buttons, complex forms, and poor visual hierarchy.

CHAPTER 04

What to do about it

Practical, prioritized recommendations. Start with the quick wins — most take less than a day and have an outsized impact on agent access and product understanding.

Quick Wins1 DAY OR LESS
Check your robots.txt

Go to yoursite.com/robots.txt right now. See who you're blocking. Remove blocks on GPTBot, PerplexityBot, ClaudeBot unless you have a strategic reason to block them.

Add Schema.org Product markup

Add structured data to every product page. At minimum: name, price, availability, image, description. Most e-commerce platforms have plugins for this.

Ensure guest checkout works

Browser agents cannot create accounts. If checkout requires login, every browser-based AI agent will fail at the last step. Enable guest checkout.

Medium EffortABOUT 1 WEEK
Audit your CDN/WAF bot rules

Whitelist GPTBot, PerplexityBot, ClaudeBot, and other AI agent user agents in your Cloudflare, Akamai, or Fastly configuration. Your security team may have blocked them without realizing the commerce impact.

Set up Google Merchant Center feeds

If you don't have product feeds, create them. Feed-based agents like ChatGPT Shopping rely on these to discover and recommend your products. Shopify stores: your products.json feed already exists.

Add llms.txt to your site

A new standard that helps AI agents understand your site structure, policies, and capabilities. Think of it as a README for AI agents. Place it at yoursite.com/llms.txt.

Longer TermONGOING INVESTMENT
Implement Agentic Commerce Protocol (ACP) endpoints

As the ACP standard matures, adding dedicated API endpoints for AI agent interactions will give you an advantage. This is the emerging standard for agent-to-store communication.

Optimize checkout for autonomous buyers

Clear CTAs, simple forms, no ambiguous steps. Every unnecessary click is a point where an AI agent might get confused. Design for clarity, not cleverness.

Monitor your agent-access signals over time

As AI agents and platforms evolve, your site's posture will shift. Track robots.txt rules, structured data coverage, and WAF behavior continuously to catch regressions before they cost you sales you can't see in analytics.

CHAPTER 05

Where this is going

Agentic commerce is early, but the trajectory is clear. Here's what the next 12–24 months look like.

Blocking becomes irrelevant

Computer use agents browse your site exactly like a human — through screenshots and clicks. You can't block them with robots.txt because they don't read it. They open Chrome. The era of controlling AI access through text files is ending.

The question shifts

It's no longer "can agents access your site?" It's "will they succeed when they try?" The brands that win will be the ones where AI agents can actually complete a purchase, not just read a product page.

Agent conversion rate becomes a key metric

Alongside human conversion rate, e-commerce teams will track agent conversion rate — what percentage of AI agent visits result in a successful purchase? This metric doesn't exist in your analytics today.

First-mover advantage is real

Brands that optimize now will capture AI-driven traffic while competitors are still figuring out what agentic commerce means. When a consumer asks ChatGPT "buy me running shoes," the agent will recommend brands it can actually buy from.

The shift isn't coming. It's here.

84 million shopping queries per week on ChatGPT alone. 10 live AI shopping agents. 1,000+ brands tracked. The data is clear: brands that are ready for AI agents will capture a growing share of commerce. The ones that aren't won't even know what they're missing.

Get the weekly agentic commerce brief

The most important agent-access changes, new machine-readable signals, and infrastructure shifts from the ARC index. No credit card required.