How AI Shopping Agents Work, and Where Your Store Fits In

AI shopping agents get described as robots with credit cards. The reality is more ordinary, and more useful to understand. Most of the time, an assistant is doing research: reading product data, comparing options and handing the shopper a short list. Sometimes it helps finish the purchase. Either way, your store still makes the sale, ships the order and handles the return. Here’s how the journey works, one step at a time, and which parts of it you still control.

Three terms people mix up

• AI shopping assistant. A chat tool that answers product questions and recommends options. ChatGPT, Gemini, Copilot and Perplexity all do this.
• AI shopping agent. An assistant that takes steps for the shopper, such as checking stock, filling a cart or starting a checkout. The line between assistant and agent is blurry, and the same product can be both.
• Agentic checkout. Payment completed inside the AI tool instead of on your site, usually with a wallet the shopper already saved.

When store owners worry about AI shopping agents, they usually picture the third one. In practice, the first two are where most of the activity is.

One purchase, step by step

Here’s a simplified version of what happens when someone types “I need a carry-on that fits under a plane seat, under $150” into an assistant.

1. The ask. The shopper describes a need, often with constraints: budget, size, delivery date, materials.
2. Retrieval. The assistant pulls candidate products from what it can read: merchant feeds, listings it has indexed, pages it fetches live, and reviews.
3. The shortlist. It drops anything that breaks the constraints. Products with missing dimensions or unclear prices are easy to drop, because the assistant can’t confirm they fit.
4. The comparison. It weighs price, delivery time, return terms and what reviewers say, then explains its picks in plain language.
5. The handoff. The shopper clicks through to the store’s site, or, on some platforms, completes checkout inside the chat.
6. The order lands in your admin. Shopify or WooCommerce records it like any other order, and your emails, fulfillment and support take over from there.

Steps two to four are where stores win or lose, and they run on data you already own.

Where the product information comes from

An assistant doesn’t see your hero banner or your brand video. It reads structured facts. The main sources are:

• Product feeds. Google Merchant Center is the big one, and OpenAI now publishes its own product feed specification for merchants who share catalogs with ChatGPT.
• Platform connections. Shopify’s Agentic Storefronts channel can make eligible products discoverable in several AI tools from one place in the admin.
• Your own pages. Product structured data, policy pages and plain text descriptions, read by crawlers or fetched live when a shopper asks a question.
• Other people’s words. Reviews, forums and comparison articles that mention your products.

Who actually sells the product

This is the part that reassures most owners. In the programs I’ve looked at, the merchant stays the seller. Google’s documentation for its checkout in AI Mode and Gemini says merchants remain the seller of record. In practice that means:

• Your prices, taxes and shipping rules still apply.
• Your return policy governs returns, which is one more reason to make it clear.
• Customer service questions still come to you. The shopper may forget which assistant they used, but they’ll remember the store name on the box.
• Fraud screening and chargebacks still run through your payment provider.

Checkout inside the chat is still settling

It’s easy to assume every assistant now buys things directly. The record so far says otherwise. OpenAI launched Instant Checkout in ChatGPT in September 2025, then in March 2026 moved back to merchants’ own checkouts and put its effort into product discovery. Google offers checkout to a limited group of merchants. Shopify’s channel sends ChatGPT shoppers to the store’s own checkout page.

My read is that discovery inside assistants is real and growing, while payment inside the chat varies a lot by platform. Treat anything you read about agentic checkout as true at the time of writing, and check the platform’s own documentation before you plan around it.

What you control, in order of impact

1. Product data: complete titles, attributes and identifiers, with accurate price and stock.
2. Policies: shipping costs and times, return terms and warranties, written in plain text.
3. Feeds: a clean Merchant Center account, plus any platform AI channels you choose to join.
4. Access: robots.txt and firewall settings that let the crawlers you want reach your pages.
5. Measurement: purchase tracking you trust, so you can see whether assistant traffic converts. My conversion tracking setups include that test.

The short version: if a fact matters to a buyer, it should live in a proper field, in your feed and on the page.

What shows up in your analytics

Assistant traffic is real but often small, and it’s easy to miscount. Some visits arrive with a clear source, such as chatgpt.com or perplexity.ai, in GA4’s referral reports. Others arrive with no referrer at all and land in direct traffic, especially from mobile apps. Orders completed through a platform’s own checkout may appear in your store admin under a sales channel rather than as a website session. Three habits help:

• Build a GA4 channel group or exploration that collects the main AI referrers in one place.
• Check order sources in your store admin, not only in analytics.
• Ask buyers how they found you in a short post-purchase question.

Myths I hear from store owners

• “Agents will buy everything for people.” Routine, well-specified purchases are a natural fit. A sofa or a wedding outfit still sends most people to the store’s site for a closer look.
• “I need a special AI file to be found.” Google says there’s no special markup or AI text file required to appear in its AI features. Good product data and normal SEO do the job.
• “This only matters for big brands.” Assistants match needs to products. A small store with precise data can answer a specific question better than a big one with vague listings.
• “My website matters less now.” Plenty of shoppers still visit before they buy, especially for bigger purchases, and your site is where the facts in your feed get checked.

If you run WooCommerce, the catalog, feed and checkout testing behind all of this is part of my WooCommerce development work.

Worth reading next: agentic AI for small business, explained in plain English and the Shopify store setup checklist I run before every launch.

Not sure what an assistant sees when it reads your catalog? Request a free website review and I’ll tell you which gaps would matter most.