AI shopping agents: how they choose products

AI shopping agents now sit between your store and a growing share of your shoppers. They answer the question, narrow the field and, more and more often, place the order. This guide is for Shopify merchants: which AI shopping agents exist, how they choose products, how far shoppers trust them, and what you can change on your side.

What are AI shopping agents?

AI shopping agents are AI assistants that do shopping tasks for a person. The simplest ones answer questions and compare products. More capable ones track prices, fill carts and, with the shopper’s permission, complete the purchase. The shopper sets the limits, and the merchant still sells and ships.

It helps to think of a ladder rather than one kind of tool. At the bottom, an assistant answers “which soap travels well?” In the middle, it builds a cart or watches a price. At the top, it buys on its own within rules the shopper set.

The large platforms already run the top rung in limited form. Google’s agentic checkout, launched November 13, 2025, buys a tracked item when the price falls within the shopper’s budget, after the shopper confirms. Amazon says its assistant can “auto-buy items at a set price.”

Agentic commerce is the wider name for this shift. If you want the protocols and the payment side first, start with our guide to agentic commerce. This guide stays with the agents and how they choose.

Which platforms run AI shopping agents?

The largest platforms all run one: OpenAI’s ChatGPT, Google’s AI Mode and Gemini, Amazon’s Alexa for Shopping (called Rufus until May 2026), Microsoft Copilot and Meta AI. They differ in what the agent can do, who can use it, and where it gets your product data.

Here is what each company says about its own agent, as of this writing.

Platform	What the agent can do	Where your products come from
ChatGPT	Find products and send the shopper to the merchant’s own checkout, in an in-app browser or a new tab	Shopify Catalog for Shopify merchants, and OpenAI’s product feed
Google AI Mode and Gemini	Agentic checkout within a set budget in AI Mode, and checkout in AI Mode and the Gemini app, in the US, from eligible merchants including select Shopify merchants	The Shopping Graph: more than 50 billion product listings
Alexa for Shopping (Amazon)	Auto-buy at a set price, and Buy for Me on other brands’ sites	Amazon’s store, plus stores across the web through Shop Direct
Microsoft Copilot	Copilot Checkout in the US, inside the Copilot chat without a redirect	Merchants via PayPal, Shopify and Stripe
Meta AI	A shopping mode for ideas, styling and Marketplace listings	Inspiration and brand storytelling across Meta’s apps

A few details matter for a Shopify brand. Shopify said on March 24, 2026 that products “become discoverable in ChatGPT by default via Agentic Storefronts.” Microsoft said Shopify merchants “will automatically be enrolled in Copilot Checkout following an opt-out window.” OpenAI scaled back its Instant Checkout feature in March 2026, so ChatGPT purchases now finish on the merchant’s own checkout.

On May 13, 2026, Amazon renamed Rufus as Alexa for Shopping. Products available through Buy for Me grew “from 65K at launch to over half a million,” Amazon says. Meta’s shopping mode is the least transactional of these: it surfaces ideas and lets people search Facebook Marketplace listings near them.

How do AI shopping agents choose products?

We found no published ranking formula from any of them. The signals they disclose are familiar: the shopper’s request and limits, price, availability, reviews, delivery and returns, and shopping history. Agents read all of it from product data and pages, so they can only weigh what your listing states.

Amazon is the most specific. Its shopping experience uses “signals including reviews, price, availability, delivery speed, return rates, and browsing and shopping history,” and review highlights summarize sentiment “based solely on the content from customer reviews.” That is a useful checklist even off Amazon, because every agent faces the same choice between similar products.

Google is specific about freshness. In AI Mode, the Shopping Graph holds “more than 50 billion product listings, 2 billion of which are updated every hour.” An out-of-date price or stock level is a reason not to pick you.

The shopper’s limits come first. Google’s agentic checkout acts only inside the budget the shopper set. Shopify’s developer docs add a caution for the data side: in its Catalog, “some fields might be inferred by Shopify’s AI,” so they may be missing or less accurate when product data is thin.

Two things are worth knowing about what is not organic. OpenAI’s product feed has a separate flag for ads eligibility, and OpenAI documents publishing ads in ChatGPT. Shopify’s Catalog lets approved developers request paid placements. Neither changes the basics: an agent cannot recommend a strength your page never states.

So what should an AI shopping assistant ideally consider? The shopper’s need first, then what each product page and data feed says about fit, price, availability, reviews, delivery and returns. Where your page is silent, the agent guesses, or picks the product that answered.

Do shoppers trust AI agents to buy for them?

Many trust agents to help, far fewer to pay. In Accenture’s survey of 25,590 people in 16 countries, 74% would delegate routine tasks to an agent acting on instruction, 32% would let it decide what to buy if they pay, and 9% would allow fully autonomous purchases.

The chart below shows how support falls as the agent gets more control.

How much shoppers would let an AI agent do

Source: Accenture, Talk to my AI agent (2026), n=25,590

Show data

How much shoppers would let an AI agent do

What the agent may do	Value (%)
Routine tasks, on instruction	74
Choose what to buy, shopper pays	32
Buy fully on its own	9

Taste matters more than complexity. Accenture found that consumers are “delegating the parts of shopping that feel like work” while keeping the choices that carry personal meaning. The same person who lets an agent book a whole trip might still want to pick the hotel room alone.

Two more numbers from the same survey matter to a brand. 56% would tell their agent which brands to consider. Yet 37% of loyal consumers would let an agent switch brands for a better fit. Loyalty earns you a place on the agent’s list, and the product page has to hold it there.

These are stated intentions, not observed purchases, and the survey spans 16 countries. Read them as direction, not as a forecast for your store.

How should merchants prepare product data for AI shopping agents?

Make every product complete in the feeds agents read, state on the page what the product is, who it is for and what it is not, answer the questions shoppers ask, keep structured data matching the visible page, and let the AI crawlers in. Then measure what the assistants say.

Google’s guidance is blunt: “There are no additional requirements to appear in AI Overviews or AI Mode,” and “there’s also no special schema.org structured data that you need to add.” What it does ask is that your structured data match the visible text on the page and that your Merchant Center information be up to date. There is no trick here, only the basics done completely.

A short list for a Shopify store:

1. Fill the required fields. OpenAI’s feed specification names nine fields “required for a useful discovery feed”: item_id, title, description, url, brand, seller_name, image_url, availability and price. Use a real brand name, not a placeholder.
2. Say what the product is and who it is for. A merchant description often skips both. An agent matching “travel soap for a gym bag” needs the page to say who the soap is for, in words.
3. Answer the questions shoppers ask. Size, use, comparisons, returns. An answered question is something an agent can repeat.
4. Keep structured data honest. Markup that says more than the page shows works against you with Google.
5. Let the crawlers in. OpenAI says sites that block OAI-SearchBot “will not be shown in ChatGPT search answers,” and robots.txt changes take about 24 hours to apply. Check your store:

AI crawler access checker

Check whether 15 AI crawlers, from GPTBot to PerplexityBot, may read your site's pages.

Your website's address

For example yoursite.com.

Check my site

A quick check confirms you are a person, not a bot.

Descriptions checked 2026-10-13. We do not save the address or the result.

Open the full tool

The last step is measuring, and it is the one most stores skip. Assistants do not give the same answer twice, so one question asked once tells you little. We ask ChatGPT, Gemini and Perplexity the questions your shoppers ask, many times over, and report how often you appear, with the range across runs.

That is the loop we built SourceVision for. We propose the missing page content, what the product is, who it is for, what it is not and a short FAQ, from facts you confirm. Nothing is written without your approval. Every write is read back, every applied change is recorded with what it replaced so support can revert it on request, and a fresh measurement shows whether the assistants picked it up. See how it works.

SourceVision is in an invite-only beta.

Request an invite

Which protocols do AI shopping agents use?

Agents use open protocols at three layers. MCP connects an agent to tools and data, such as a store’s catalog. A2A connects one agent to another. Checkout and payment protocols, such as ACP, UCP and AP2, let an agent buy with the shopper’s approval and without exposing card details.

The Model Context Protocol calls itself “an open-source standard for connecting AI applications to external systems,” like “a USB-C port for AI applications.” On December 9, 2025, Anthropic donated MCP to the Agentic AI Foundation, a directed fund under the Linux Foundation co-founded by Anthropic, Block and OpenAI.

The Agent2Agent protocol was developed by Google and donated to the Linux Foundation. Its site puts the split plainly: MCP standardizes how an agent talks to tools, and A2A standardizes how agents talk to each other. The two are “highly complementary.”

Shopify already speaks to agents at the tool layer. Its agent documentation lists a Global Catalog that “searches across all Shopify merchants” and a Storefront Catalog “scoped to a single merchant’s store,” with cart and order tools beside them. The checkout and payment protocols are covered in our agentic commerce guide.

You do not need to implement any of these yourself on Shopify. You do need the data they carry to be right.

On Shopify, the same data flows to several agents at once through Agentic Storefronts. Read what you control there in our guide to Shopify Agentic Storefronts.