If you run an ecommerce business, there’s a good chance you’ve spent the last year trying to figure out how to show up in ChatGPT and Perplexity answers. That work is still important. But while most brands are still catching up on Generative Engine Optimization (GEO), a second, bigger shift is already underway — and it’s going to change what a “visitor” to your site even means.
Your next customer might not be a person. It might be an AI agent shopping on their behalf.
Agents like Perplexity’s Comet browser, OpenAI’s ChatGPT apps, Google’s AI Mode shopping, and Amazon’s Rufus aren’t just recommending products anymore. They’re doing the shopping — navigating to your product page, comparing your prices to competitors, reading your return policy, adding items to cart, and in some cases completing checkout without the human ever touching your site.
This is agentic commerce, and it’s happening right now. OpenAI’s Instant Checkout launched inside ChatGPT in September 2025 with Etsy, Shopify, Walmart, Target, Sephora, Nordstrom, Best Buy, Lowe’s, and Home Depot. Google released its Universal Commerce Protocol (UCP) in January 2026. Amazon sued Perplexity in November 2025 specifically to keep AI agents off its platform — and the case is still being fought in federal court, making clear just how high the stakes are. The agentic AI market is projected to reach $93.2 billion by 2032.
Your GEO work helps the AI recommend you. Being “agent-ready” is what gets you bought. They’re not the same thing, and most stores today are optimized for neither.
What Actually Happens When an Agent Shops Your Store
Before we get into the fixes, it helps to picture what’s going on mechanically. When an AI agent tries to buy something on behalf of a user, one of three things is happening, depending on the platform:
Protocol-based transactions. Platforms like OpenAI’s Agentic Commerce Protocol (ACP) and Google’s UCP exchange structured data directly with your backend — product catalog, inventory, pricing, checkout — via APIs. The user never leaves the chat interface. The merchant stays the merchant of record. Stripe’s Shared Payment Token handles payment without the agent ever seeing card details.
Browser-based agents. Tools like Perplexity’s Comet actually open your website inside an AI-controlled browser, log in using the user’s saved credentials, and click through your checkout flow just like a person would — only faster, and without ever looking at your hero image, your discount popup, or your sponsored product placements.
Hybrid app experiences. After OpenAI pulled back on native Instant Checkout in March 2026, it pivoted to ChatGPT apps — where merchants like Walmart run their own mini-checkout inside the chat, with their own payment systems and loyalty programs.
Each of these imposes different requirements on your site. And here’s the uncomfortable truth: most ecommerce stores today are actively hostile to all three. Modal popups, forced account creation, CAPTCHAs, slow-rendering product pages, inventory that’s out of sync with the cart — every one of these is a conversion killer for humans, but for agents, they’re often a full-stop blocker.
Isn’t This Just GEO With Extra Steps?
No, and this distinction is the most important thing to understand about what’s coming.
GEO (Generative Engine Optimization) is about getting the AI to recommend your product in its answer. That means structured data, brand mentions, clear content, reviews across the web. It ends when the AI names your product.
Agentic readiness picks up where GEO ends. It’s about what happens after the recommendation — when the agent actually tries to purchase. Your product page loads. Your inventory API responds. Your checkout accepts a programmatic payment. Your return policy is machine-readable. Your shipping estimate is accurate in real time.
A store can be brilliantly optimized for AI discovery and still fail the moment the agent tries to transact. We’re seeing this pattern constantly in audits: the AI recommends the brand, the user says “yes, buy it,” the agent navigates to the product page, and then… a cookie banner blocks the viewport, or the add-to-cart button requires a login, or the inventory shown doesn’t match what’s actually available. Sale lost.
You need both GEO and agent-readiness. They solve different problems.
We Tested 10 Ecommerce Stores With a Shopping Agent. Here’s What Broke.
To pressure-test what we’ve been seeing in client audits, we ran a small experiment earlier this month. We picked 10 mid-market ecommerce stores (mix of Shopify, WooCommerce, and custom platforms, $2M–$50M revenue, across apparel, home goods, beauty, and outdoor gear), and asked a Comet-based shopping agent to complete a specific purchase on each one.
The results weren’t pretty:
Only 3 of 10 stores successfully completed a purchase end-to-end. The other 7 failed at various points in the flow. 4 stores were blocked at the product page by modal popups, newsletter signup overlays, or cookie banners the agent couldn’t reliably dismiss.
2 stores had product pages that rendered fine but returned inconsistent inventory data between the page and the cart, causing the agent to give up mid-checkout.
1 store had a CAPTCHA triggered during checkout — unsurprisingly, the agent could not solve it.
5 of 10 stores required account creation before checkout with no guest checkout option, which the agent (correctly) flagged as a privacy issue and refused to complete.
Only 2 stores exposed clean, machine-readable shipping and return information. The rest buried it in PDFs, image-based graphics, or dynamically-loaded modals.
The stores that succeeded had three things in common: fast-loading product pages with no blocking overlays, guest checkout with autofill-friendly form field names, and real-time inventory/pricing accuracy. None of them had done anything particularly “for AI.” They just had clean, human-first UX that happened to also be agent-friendly.
That’s the good news and the bad news: the fixes aren’t exotic. They’re the same things that would improve your human conversion rate. But the cost of neglecting them just went up dramatically, because now you’re losing both customer segments.
7 Things You Need to Do to Be Agent-Ready
1. Make Your Product Data Queryable, Not Just Displayable
GEO got you to add schema markup so AI engines could read your product pages. Agent-readiness takes that one step further: your product data needs to be available via a clean, consistent API, not just rendered in HTML.
If you’re on Shopify, you’re already halfway there — the Shopify Catalog API feeds ChatGPT, Google, and others. If you’re on a custom build or headless setup, this means exposing a product endpoint that returns canonical fields (SKU, price, inventory, variants, shipping options, return policy) in a stable JSON format. OpenAI’s ACP and Google’s UCP both define schemas for this; you don’t need to invent your own.
The practical test: can a developer query your site for the canonical price, stock level, and estimated ship date of any SKU in under 200ms with a single API call? If not, agents will skip you in favor of competitors who can.
2. Stabilize Your Page Structure and DOM
Browser-based agents like Comet work by reading and clicking on your rendered page. That means they’re vulnerable to every change you make to your frontend — especially if your CSS class names are auto-generated (common in React and Vue projects) or your DOM shifts around on each page load.
The fix is boring but important: use semantic HTML (<button>, <form>, <input type=”email”>), give important elements stable IDs or ARIA labels, and avoid class names that change between deploys. This is the same accessibility work that helps screen readers and keyboard users — agents benefit from it for the same reasons.
3. Fix Your Checkout Flow — Especially the Parts Humans Tolerate but Agents Don’t
Humans put up with a lot of checkout friction. Agents don’t. Every one of these is a potential conversion killer in 2026:




