How AI shopping agents in ChatGPT, Gemini, Copilot and Perplexity find, compare and buy products, and how Shopify brands make their catalog agent-ready.
A shopper types "running shoes for flat feet, under $150, delivered by Friday" into ChatGPT. Four products come back. Yours isn't one of them, even though you sell exactly that shoe at $129 with two-day shipping.
Nobody outbid you. The agent never had enough information to pick you. It couldn't read your arch-support spec, couldn't confirm stock in a size 10, or couldn't find your return policy, so it went with stores whose data answered every part of the question.
This guide explains how AI shopping agents work as of November 2026, where they find products, how they decide what makes the shortlist, how (and whether) they actually buy, which agents your customers already use, and what a Shopify brand controls at each step.
TL;DR
- An AI shopping agent takes a shopping goal in plain language, searches across stores, compares options against the shopper's constraints, and either completes checkout or hands the shopper a ready-to-buy shortlist.
- Agents read data, not storefronts. They pull from product feeds and catalogs such as Shopify Catalog and Google's Shopping Graph, which holds more than 60 billion listings (Google, 2026).
- Most agents still stop before payment. Only 11% of US consumers would let AI make a purchase decision for them (Gartner, May 2026), so the shortlist is where you win or lose.
- The money is real. Morgan Stanley estimates agentic shoppers could account for $190 billion to $385 billion of US ecommerce spending by 2030 (Morgan Stanley, December 2025).
- For Shopify brands, the agent channel is already on. Agentic Storefronts lists eligible stores in ChatGPT, Copilot, Google AI Mode, Gemini and Meta by default. Your job is the data the agent reads and the relationship after the first order.
What is an AI shopping agent?
An AI shopping agent is software that shops on a person's behalf. The shopper states a goal ("a waterproof jacket for Scotland in April, under $200"), and the agent searches multiple stores, filters by the constraints, compares the survivors and returns a recommendation. Some agents then complete checkout. Others hand off to the merchant's store.
The word "agent" matters because of that last step. A tool that only answers questions is an assistant. A tool that carries a task through several steps, and can act on the result, is an agent. Zinc, a retailer purchasing API, draws the line sharply: if the software only returns links, it's an assistant (Zinc, 2026).
In practice, almost every major product in 2026 sits somewhere between the two. ChatGPT, Google AI Mode and Perplexity all research, compare and recommend. Whether they buy depends on the surface, the merchant and the category.
AI shopping agent vs AI shopping assistant vs chatbot
The three terms get used interchangeably, including by vendors selling them. They describe different jobs.
| Chatbot | AI shopping assistant | AI shopping agent | |
|---|---|---|---|
| What it does | Answers one question, then waits | Recommends products inside one store or app | Carries a shopping goal across stores, from search to checkout |
| Who runs it | The merchant | Usually the merchant | Usually a platform: OpenAI, Google, Microsoft, Perplexity, Amazon |
| Where it lives | Help widget on your site | Your website or mobile app | ChatGPT, Gemini, Copilot, Perplexity, Alexa |
| Can it buy? | No | Adds to cart; the shopper checks out | Sometimes, with the shopper's approval |
| Whose interest it serves | Your support queue | Your conversion rate | The shopper's brief |
That last row is the one you should keep a note of. An assistant on your site works for you. An agent works for the shopper and treats your store as one option among dozens. If you want the merchant-side version, an assistant inside your own store, our guide to AI shopping assistants covers it.
Two nearby tools get mistaken for agents. A price comparison site lists offers and stops there, while an agent applies the shopper's constraints and can act on the result. And the recommendation and search features on your own site rank products for people already in your store, which is a different job. Our guides to AI merchandising and personalized product recommendations cover that side.
This post is about the other kind: the agents your customers bring with them.
How do AI shopping agents find products?
AI shopping agents find products in structured catalogs first and on the open web second. Google pulls from its Shopping Graph, ChatGPT and Copilot read Shopify Catalog through Agentic Storefronts, and Perplexity takes merchant feeds. Pages an agent can't parse, or products missing from these catalogs, rarely make the shortlist.
Each major agent has its own front door:
- Google AI Mode and Gemini draw shopping results from the Shopping Graph, which Google says holds more than 60 billion continuously refreshed listings (Google, October 2026). Shopify stores feed it through the Google & YouTube sales channel and Merchant Center. Our Google AI Mode shopping guide walks through the setup.
- ChatGPT, Microsoft Copilot and Meta read Shopify Catalog. Shopify's Agentic Storefronts, launched December 10, 2025, "is active by default for eligible stores" and lists products across these channels (Shopify Help Center, checked November 2026). The details are in how to sell on ChatGPT.
- Perplexity takes products from Shopify Catalog, its free Merchant Program and PayPal store sync. See Perplexity Shopping.
- Amazon's Alexa for Shopping (renamed from Rufus in May 2026) searches Amazon's own catalog first. Its Buy for Me feature reaches out to brands' own websites too, and some Shopify brands found their products listed without asking (Modern Retail, January 2026).
Two things follow from this. First, an agent sees your feed before it sees your product page, so a stale feed is a stale storefront. Second, the open web is a fallback, not a strategy. Agents that browse retailer sites directly run into blocks: Amazon blocked Meta's new Muse agent from Amazon.com on September 20, 2026, 12 days after launch, saying the agent didn't identify itself (TechRepublic, September 2026).
{{cta}}
How do AI shopping agents compare products?
AI shopping agents compare products by turning the shopper's request into hard constraints (price, size, compatibility, delivery date) and checking every candidate against them. Products with missing attributes drop out first. The survivors get ranked on fit, reviews, price and policies, and the shopper usually sees only two to five options.
That's a much harsher filter than a search results page. A person scrolling a category page forgives a vague title and clicks through to check. An agent doesn't click through on hope. If your listing doesn't state that the shoe has a wide toe box, the agent can't confirm it, and an unconfirmed product loses to one that says so.
Here's what agents check, roughly in the order they disqualify you:
- Hard constraints. Price, variant availability (size 10 in stock, not just "in stock"), shipping to the shopper's region, delivery estimate.
- Explicit attributes. Material, dimensions, ingredients, compatibility. These come from your product fields and metafields, not from a lifestyle photo.
- Policies. Return window, warranty, shipping cost. Shopify's Agentic Storefronts includes a knowledge base for exactly this non-product information.
- Social proof. Review count and rating, and what the reviews say about the attribute the shopper asked about.
- Consistency. If your feed says $129, your product page says $139 and Amazon says $119, the agent trusts none of them.
The writing on the product page still matters, just for a different reader. Our post on writing product descriptions for AI shopping assistants shows how to phrase attributes so a model can quote them.
How do AI shopping agents buy?
AI shopping agents buy in one of two ways: in-chat checkout, where the agent completes payment inside the assistant using the merchant's checkout, or a handoff, where the agent sends the shopper to the merchant's store to pay. Either way, the shopper approves the purchase, and the merchant stays the seller of record.
Here's where each major agent stood in November 2026:
| Agent | How it checks out | How a Shopify store gets in |
|---|---|---|
| ChatGPT | Handoff to the merchant's store. OpenAI wound down in-chat Instant Checkout in March 2026 | Agentic Storefronts (default for eligible stores) |
| Google AI Mode and Gemini | In-chat checkout through the Universal Commerce Protocol, or handoff | Agentic Storefronts; direct checkout on by default for eligible stores |
| Microsoft Copilot | In-chat Copilot Checkout, launched January 8, 2026 with PayPal, Shopify and Stripe | Agentic Storefronts |
| Perplexity | In-chat Instant Buy powered by PayPal, free for US users since November 2025 | Shopify Catalog, Merchant Program, PayPal store sync |
| Amazon Alexa for Shopping | Amazon checkout; Buy for Me buys on brands' own sites | Amazon listing, or Buy for Me picks you up |
| Meta Muse | Browses retailer sites | Agentic Storefronts lists products in Meta |
Sources: Shopify Help Center (November 2026), OpenAI (March 2026), Microsoft and PayPal (January 2026), Perplexity and PayPal (November 2025), Amazon (2026), TechRepublic (September 2026).
Under the hood, three open protocols do the plumbing. OpenAI and Stripe's Agentic Commerce Protocol (September 2025) defines how an agent reads a catalog and places an order. Google and Shopify's Universal Commerce Protocol (January 2026) does the same for Google's surfaces and beyond. Google's Agent Payments Protocol adds signed proof that the shopper approved the payment. Shopify runs all of this for you through Agentic Storefronts. Our comparison of ACP, UCP and AP2 goes deeper if you want the mechanics.
What doesn't exist yet, at scale, is the fully autonomous agent that buys without asking. Gartner found only 11% of US consumers would let AI make a purchase decision, even in low-stakes categories. Visa's research points the same way: about half of shoppers fear "decisions made without me," and 85% want explicit control over the data an agent can access (Visa, February 2026).
So the agent builds the shortlist. The person still taps "buy."
What can go wrong when an agent buys?
The common failures are small, and they start with product data. An agent orders a size 10 because the feed said it was in stock and the warehouse had none. A shopper asks about free returns and the agent quotes a policy that changed last month. Under both ACP and UCP the merchant stays the seller of record, so the refund and any dispute land on you (our comparison of ACP, UCP and AP2 covers how each protocol handles it). The wider list of fraud, privacy and pricing-control risks is in our agentic commerce guide.
How big is agentic shopping right now?
Agentic shopping is small in share and growing fast. Morgan Stanley estimates agentic shoppers could reach $190 billion to $385 billion of US ecommerce spending by 2030, a 10% to 20% share of online retail, and found roughly 23% of Americans had made a purchase using AI in the past month (Morgan Stanley, December 2025).
AI is already part of how many US shoppers research. Locus's Q2 2026 survey of 1,000+ US online shoppers found that 45% now use AI tools like ChatGPT and Claude as a primary or secondary tool for discovery, comparison and purchase decisions. Among those AI-assisted shoppers, 39% say they are more likely to try new brands. That second number matters for smaller Shopify brands: an agent's shortlist can include a brand the shopper has never heard of, as long as its product data is clear enough to be picked.
The traffic is already measurable. Adobe found that visits from generative AI tools to US retail sites rose 693.4% in the 2025 holiday season, and those visitors converted 31% better than other traffic (Adobe Analytics, January 2026). Amazon says more than 400 million customers used its shopping assistant in the 12 months to Q3 2026 (Amazon, October 2026).
Gartner puts a longer-range number on it: 20% of digital commerce transactions will run through AI platforms by 2030 (Gartner, September 2026).
For a Shopify brand doing $5 million a year online, a 10% agentic share by 2030 is $500,000 of revenue where an agent picks the shortlist. That's a channel worth getting your data right for now, while competitors still treat it as a curiosity. Our deeper look at machine customers covers what that shift means for who owns the customer.
How do you make a Shopify catalog agent-readable?
You make a Shopify catalog agent-readable by giving every product complete, structured, consistent data: specific titles, filled-in attributes and metafields, accurate variant-level stock, plain-language policies, and the same price everywhere. Then you check that Agentic Storefronts and your Google channel are switched on and syncing.
A practical order of work:
- Confirm your channels. In the Shopify admin, open Sales channels > Agentic and check which AI channels are active. Then confirm the Google & YouTube channel syncs products and shipping automatically.
- Fix titles first. "The Everyday Tee" tells an agent nothing. "Men's organic cotton crew neck t-shirt, relaxed fit" tells it four things it can match against a request.
- Fill the attributes agents filter on. Material, size range, dimensions, compatibility, care instructions. Use Shopify's category metafields so the values are structured, not buried in a paragraph.
- Keep variant stock live. An agent asked for a size 10 checks the size 10, not the product.
- Publish policies in plain words. Return window, shipping times, warranty. Load them into the Agentic Storefronts knowledge base so agents answer from your words.
- Match prices across channels. Feed, product page and marketplaces should agree, including sale prices.
- Don't block legitimate agents. Block abusive scrapers. An agent that can't read your product pages can't recommend them.
If you also sell on Amazon, the same rules apply with Amazon's own wrinkles; see Amazon Rufus optimization.
Once the catalog is fixed, check it against what the agents actually say. Once a month, ask ChatGPT, Gemini, Copilot and Perplexity a few of the questions your customers ask, then compare the price, stock and return policy each one quotes with your product page. Every mismatch tells you which field to fix first. Our guide to generative engine optimization covers how to track the results over time.
What does an AI shopping agent mean for repeat customers?
An AI shopping agent wins you a first order and then keeps the relationship for itself. The next time that shopper needs running shoes, they ask the agent again, and the agent runs the comparison again. You compete from zero every time, and the agent owns the moment of intent.
The agent channel is a great acquisition channel and a poor retention one. Shoppers who come back on their own come back through channels no agent sits between: your app, your push notifications, your loyalty program, your email list.
The brands that will do well treat agents as the top of the funnel and their own channels as the rest of it. They capture the customer on the first order (account creation, app install, opt-in) and give them a reason to come straight back. Inside your own app, you can run the agent experience on your terms: Appbrew's AI Concierge gives shoppers a personal-shopper style assistant inside your Shopify mobile app that understands intent and recommends from your catalog only. Our guide to the AI personal shopper explains how that works for a Shopify store.
{{cta2}}
Frequently asked questions
What is an AI shopping agent?
An AI shopping agent is software that shops on a person's behalf. It takes a goal in plain language, searches across stores, compares options against the shopper's constraints such as price, size and delivery date, and either completes checkout with the shopper's approval or hands them a ready-to-buy shortlist.
What is the difference between an AI shopping agent and an AI shopping assistant?
An AI shopping assistant usually lives inside one store or app and recommends that store's products. An AI shopping agent works for the shopper across many stores, runs the comparison itself and can act on the result, including checkout. ChatGPT, Gemini, Copilot and Perplexity work as agents; an in-store chat widget is an assistant.
How is an AI shopping agent different from a price comparison site?
A price comparison site lists the same product across sellers and leaves the choice to the shopper. An AI shopping agent starts from a goal in plain language, filters by constraints such as size, budget and delivery date, and returns a recommendation. On some surfaces, it can complete checkout with the shopper's approval.
Can AI shopping agents actually buy things?
Yes, on some surfaces. Microsoft Copilot, Google AI Mode and Gemini, and Perplexity can complete checkout inside the assistant for eligible merchants. ChatGPT sends shoppers to the merchant's store to pay since OpenAI wound down Instant Checkout in March 2026. In all cases the shopper approves the purchase.
How do I get my Shopify products into AI shopping agents?
Start in the Shopify admin under Sales channels > Agentic, where Agentic Storefronts lists eligible stores in ChatGPT, Copilot, Google AI Mode, Gemini and Meta by default. Connect the Google & YouTube channel for Google's Shopping Graph, and apply to Perplexity's Merchant Program. Then make product data complete and consistent.
Are AI shopping agents safe for shoppers?
The major agents keep the shopper in control: they ask for approval before payment and pay through tokenized cards, PayPal or Google Pay rather than sharing card numbers with every store. Shoppers still worry, with about half fearing decisions made without them (Visa, 2026), which is why most agents stop at a recommendation.
See how Appbrew helps Shopify brands build AI-powered native apps that improve product discovery, conversion, and customer retention.







.webp)




