AI shopping agents are assistants that find, compare and sometimes buy products for you. As of November 2026, the ones that can complete a purchase are Google AI Mode and Gemini (US, and Canada since November 4), Perplexity’s Instant Buy, Microsoft Copilot Checkout and Amazon’s Buy for Me, all US. ChatGPT still recommends products but now sends you to the store’s own site or app to pay.
Last checked November 11, 2026.
What an AI shopping agent is, and what it isn’t
A shopping chatbot answers questions. An AI shopping agent takes actions: it tracks a price, fills a cart, enters your shipping details or completes a checkout with a stored payment method. The line between the two is blurry, and most products sit somewhere in the middle.
A useful way to sort them is by how far the agent can go without you:
- Research only. The agent compares products and sends you to the store’s site to buy. You do every step of checkout yourself.
- Checkout inside the chat. You pick the item, the agent pre-fills payment and shipping, and you tap to confirm. The store still sells you the product.
- Buying on a trigger. You set a rule in advance (a target price, a reorder schedule) and the agent buys when the condition is met.
- Buying on another site for you. The agent goes to a store you never visited and places the order in your name.
ChatGPT now sits in the first group. Google, Perplexity and Copilot do the second. Google and Amazon do the third, and only Amazon openly does the fourth at scale. If you want the background on the card networks and protocols underneath all of this, our agentic payments explainer covers it, and the protocol tracker shows who supports ACP, UCP, AP2 and the rest.
Which AI shopping agents can buy for you: the comparison
| Agent | Can it complete checkout? | Where it works | How you pay | What the store has to set up |
|---|---|---|---|---|
| Google AI Mode, Gemini and Universal Cart | Yes, on eligible listings with a Buy button, or by moving your cart to the store’s site | US; Canada since November 4, 2026; Google’s merchant help also lists Australia | Google Pay, with cards saved in Google Wallet | A Merchant Center feed, a UCP integration approved by Google and the native_commerce attribute on each eligible product |
| Perplexity Instant Buy | Yes, on products marked “Instant Buy” | US (not Enterprise Pro or Enterprise Max plans) | PayPal, with a card option at checkout | A store compatible with Perplexity’s agentic checkout, such as through PayPal’s store sync or a supported platform |
| Microsoft Copilot Checkout | Yes, on Copilot.com and the Copilot mobile app | US buyers, English-language merchants, USD | Card in the merchant’s checkout; PayPal also powered checkout at the January launch | Shopify stores are enrolled automatically; others onboard through Stripe or a Microsoft Merchant Center account with a UCP-ready feed |
| Amazon Alexa for Shopping and Buy for Me | Yes: buys from Amazon at a target price, and Buy for Me completes checkout on the brand’s own site | US | The payment method you confirm on Amazon’s checkout page | Nothing for Amazon’s own store; outside stores can connect a feed through the Shop Direct Merchant Workspace and can opt out |
| ChatGPT shopping | No in-chat checkout as the default: OpenAI says purchases are completed on merchant websites or apps | US | The store’s own checkout | Optional product feed under the Agentic Commerce Protocol; Shopify and Etsy catalogs are already integrated |
| Walmart Sparky | Inside Walmart’s own site and app, including automatic reorders | US | Your Walmart account | Not open to other stores |
The sections below go through each one, with the dates the companies announced the changes.
The Google AI shopping agent: AI Mode, Gemini and Universal Cart
Google’s shopping agent is not one product but a set of features spread across Search and the Gemini app. Our guide to Google AI Mode shopping covers orders and returns in detail.
In November 2025 Google announced agentic checkout in Search and AI Mode. You track an item, set the size, color and amount you want to spend, and get a notification when the price falls into range. If the store is eligible, Google can then complete the purchase with Google Pay after you confirm the purchase and shipping details. The same announcement introduced “Let Google Call,” which phones nearby stores about stock, prices and promotions and sends you the answers.
In January 2026 Google launched the Universal Commerce Protocol (UCP), an open standard it co-developed with Shopify, Etsy, Wayfair, Target and Walmart, and said it would power a checkout button on eligible listings in AI Mode and the Gemini app. Google said PayPal would come “soon.” Its Merchant Center help now says the checkout uses card numbers stored in Google Wallet and is available for select merchants.
On October 15, 2026, Google said Universal Cart was rolling out in the US across Search and the Gemini app. The cart saves products from several stores, sends alerts on price drops and restocks, suggests which of your cards to use and shows promo codes. You can pay with Google Pay in the cart or move the items to the store’s own cart, and either way the merchant stays the seller of record. On November 4, 2026, Google brought UCP checkout to Canada, starting with lululemon, Shoppers Drug Mart, Simons and select Shopify merchants.
Amazon’s agent: Alexa for Shopping and Buy for Me
Amazon renamed its Rufus assistant Alexa for Shopping on May 13, 2026, combining it with Alexa+. Amazon says it can track prices, buy items at your target price, reorder essentials, manage your cart and shop from other online stores. It is available to US customers in the Amazon Shopping app, on the website and on Echo Show, with no Prime membership or Echo device required. Our Alexa+ shopping guide covers voice codes, auto-buy and returns.
Two features make it an agent rather than a search box:
- Auto-buy. You set a target price and, as Amazon described it in November 2025, Prime members can have the item bought automatically when it hits that price.
- Buy for Me. For eligible products Amazon doesn’t sell, you confirm the address, taxes, shipping and payment method on Amazon’s checkout page, and Amazon’s AI completes the purchase on the merchant’s website. On November 4, 2026, Amazon launched a Shop Direct Merchant Workspace where stores can submit feeds, choose which items appear and opt out. Our guide to Amazon Buy for Me walks through what it can spend and how to cancel an order.
Amazon’s approach is the most autonomous of the big ones: once you set the rule, the agent acts without another tap. That is convenient for toilet paper and risky for anything with sizes.
ChatGPT, Perplexity and Copilot: where checkout happens
ChatGPT. OpenAI’s merchant page now says it is “moving away from a standalone Instant Checkout experience” and that purchases “are completed on merchant-owned websites or apps.” ChatGPT still shows product cards with prices and links, and its help center says product results are not ads and that you should confirm the final price, size and return policy with the merchant before buying. Shopping is live for US users. For how in-chat buying worked and what changed, see our ChatGPT Instant Checkout guide.
Perplexity. Perplexity’s help center says Instant Buy is available to all US users except those on Enterprise Pro or Enterprise Max. PayPal processes the payment and the retailer stays the merchant of record. Eligible sellers include Wayfair, Abercrombie & Fitch and thousands of merchants on BigCommerce, Shopware and Wix. More in our Perplexity shopping guide.
Microsoft Copilot. Microsoft launched Copilot Checkout in the US on January 8, 2026, naming Urban Outfitters, Anthropologie, Ashley Furniture and Etsy sellers at launch. Microsoft’s merchant page says it runs on Copilot.com and the Copilot mobile app, that only English-language merchants selling to US buyers in USD are eligible, and that Copilot never becomes the merchant of record. On Shopify, Copilot’s direct checkout is on by default for eligible stores and shows only to US customers. See our Copilot Checkout guide.
Walmart Sparky. Walmart’s assistant covers recommendations, meal planning and replenishment on the website, the app and in stores. Walmart told investors in May 2026 that Sparky users have an average order value about 35% higher than non-users. That is Walmart’s own figure from an earnings call, not an independent measurement.
Why many furniture catalogs aren’t ready for AI shopping agents
Big home furnishings names are already in the launch lists: Wayfair for Google and Perplexity, Ashley Furniture for Perplexity and Copilot. So the category isn’t shut out. The trouble is that an agent buys from data, not from a showroom, and furniture data tends to live in places agents don’t read: a dimensions diagram in an image, a lead time in a banner, a delivery fee that appears only after you enter a ZIP code.
What agents read
Every agent above leans on three layers of product data.
Product feeds. OpenAI’s product feed spec for the Agentic Commerce Protocol requires nine fields per row: item_id, title, description, url, brand, seller_name, image_url, availability and price. Variants need a separate row each, sharing a group_id, with listing_has_variations set to true and the chosen options in variant_dict. Optional fields include material, color, a structured dimensions object and weight. The spec says dimensions must be the product’s, not the shipping box’s. A missing shipping_price “means unknown, not free shipping,” and the standard shipping format has no field for handling or transit days. Returns go in accepts_returns and return_deadline_in_days, with final-sale items sent as accepts_returns=false.
Merchant Center. Google’s product data specification has product_length, product_width, product_height and product_weight, a shipping_weight used for carrier-calculated rates, minimum and maximum handling and transit times, and a shipping_label for grouping items such as oversize goods. To offer Google’s in-chat checkout, a product also needs native_commerce with checkout_eligibility set to true, plus a return policy with cost, window and link and any required consumer notices, according to Google’s UCP Merchant Center guide. Microsoft’s Copilot Checkout asks for the same UCP attributes in Microsoft Merchant Center.
Markup on the product page. Agents that browse also read schema.org Product markup. For furniture, the useful pieces are a ProductGroup with variesBy and hasVariant for each fabric or finish, width, height, depth, weight and material, and an offer with shippingDetails that states handling and transit time. Google’s merchant listing documentation supports shipping details and hasMerchantReturnPolicy in this markup.
Why furniture is hard
- Variants multiply. One sofa in four fabrics, three colors and two configurations is 24 purchasable items, each needing its own ID, price, stock status and image. Many stores publish it as one page with a picker, which leaves an agent guessing.
- Dimensions decide the sale. A shopper asks for “a sofa under 84 inches that fits through a 30-inch door.” That only works if width, depth and height are structured fields, not text in a photo.
- Materials are vague. “Performance fabric” or “solid wood construction” doesn’t tell an agent whether the frame is oak or veneered plywood.
- Delivery is freight. Google’s UCP guide lists items “requiring non-standard delivery fees (e.g., freight, premium delivery)” as ineligible for checkout on Google, along with items that bundle installation or setup, pre-orders, personalized goods, final-sale items and merchant financing. Much of a furniture catalog falls into one of those groups. Even on a store’s own site, Crazy Egg found that multi-step delivery and postcode screens caused the most back-and-forth for an agent.
- Lead times are long and variable. Made-to-order pieces can take weeks. OpenAI’s standard feed can say
backorderorpre_orderbut cannot carry handling days, and Google’s checkout excludes pre-orders, so an agent can’t promise a delivery date. - Assembly and returns are add-ons. Assembly fees, white-glove delivery and restocking fees often sit in a policy page rather than on the item.
A checklist for a home furnishings site
- Publish one feed row per variant, with a stable
item_id, a shared group ID and the options that distinguish it. - Put product length, width, height and weight in structured fields, in one unit, and keep the package weight separate for shipping.
- Fill
materialandcolorwith what the photo shows, and use the same values in the page markup. - Use real availability values (
in_stock,backorder,pre_order). Never leave stock blank. - Set handling and transit times in Merchant Center, and label oversize or freight items with
shipping_label. - Leave
native_commerceoff for freight, made-to-order, final-sale and installation items until the delivery cost can be quoted at checkout. They can still be found and bought on your site. - Configure the return policy in Merchant Center with cost and window, and send
accepts_returnsandreturn_deadline_in_daysin an OpenAI feed. - Publish terms of service and a privacy policy at stable URLs; OpenAI’s checkout fields and Shopify’s agentic channels both ask for them.
- Mark up each product page with Product or ProductGroup, Offer,
shippingDetailsandhasMerchantReturnPolicy, and keep it in sync with the feed. - Check that your bot protection isn’t blocking the agents you want. When Crazy Egg scanned 1,100 ecommerce brands for agent readiness, 41% of the sites blocked its scanner outright with bot protection.
If you run Shopify, the switches live in Shopify agentic storefronts; for WooCommerce, see WooCommerce agentic checkout.
Real use cases that work today
Going by what the companies have shipped and what independent testers have reported, these are the jobs where an AI shopping agent saves time without much downside:
- Waiting for a price. Google’s price tracking and Amazon’s auto-buy both let you name a price and walk away. The risk is low because you picked the exact product in advance.
- Reordering the same thing. Amazon’s scheduled purchases and Walmart’s replenishment suit detergent, pet food and coffee, where there is no size or color to get wrong.
- Narrowing a long list. Asking for “a carry-on under $150 that fits a 22-inch sizer” is where agents are strongest. In the Crazy Egg review described below, the agents handled attribute matching well and said so when a constraint couldn’t be met.
- Keeping a cart across stores. Universal Cart holds items from several retailers and watches their prices.
- Buying from a brand’s own site. Amazon’s Buy for Me reaches stores outside Amazon, with the caveats in the next section.
Jobs that still go wrong more often: anything with sizes, furniture that must fit a room, first-time purchases from unfamiliar stores, gifts that must arrive by a date, and anything expensive enough that a wrong choice would hurt.
What people who used it found
In July 2026 Despina Gavoyannis at Crazy Egg ran more than 120 shopping prompts across 16 product categories through ChatGPT and Google AI Mode. She found the agents good at collapsing multi-tab research into one answer and at matching products to attributes such as ingredients, skin type and price. The weak spot was assumptions: in one flow ChatGPT made about seven purchasing decisions without asking, including color, size and delivery method, noting that the choices “seem to match the user’s interest” or that “the user didn’t specify.”
The store side has its own reports. In January 2026 Modern Retail reported that small brands found their products listed through Amazon’s Buy for Me without their consent, including orders for out-of-stock and discontinued items. Amazon said brands can opt out and are removed promptly. For a shopper, that means the product page an agent relies on can be wrong, and an order can be placed for something the store can’t ship.
What to watch out for
- Assumptions you didn’t make. Before you confirm, check size, color, quantity, delivery speed and the store name. The Crazy Egg review shows agents fill gaps silently.
- Standing rules. Auto-buy and scheduled orders keep working after you forget about them. Review them the way you’d review subscriptions, and set caps where you can; our guide to spending limits for AI agents shows where those settings live.
- Who the seller is. With Google, Perplexity, Copilot and Buy for Me the retailer remains the seller, so returns and warranty go through the retailer. Read the order confirmation to see who that is.
- Fake agents. Scammers sell “AI shopping assistants” and browser extensions that harvest card numbers. The real agents live inside apps you already use. Our guide to fake AI shopping agents lists the warning signs.
- Websites that talk back. Agents that browse can be steered by hidden text on a page. See AI browser agents and prompt injection.
- Mistakes and refunds. If an agent buys the wrong thing, your rights depend on the payment method and where you live. An AI agent bought the wrong thing: who pays? explains the options; for a large disputed charge, talk to your card issuer or a consumer adviser.
How we checked this
We read the companies’ own announcements, help centers and developer documentation (Google, Amazon, OpenAI, Perplexity, PayPal, Microsoft and Shopify), the schema.org vocabulary, a published hands-on review by Crazy Egg, and reports from Modern Retail and Digital Commerce 360. Features and availability change often, so check the app itself before relying on one. Facts checked on November 11, 2026.
Go deeper
- ChatGPT Instant Checkout
- Google AI Mode shopping
- Perplexity shopping
- Alexa+ shopping
- Copilot Checkout
- Shopify agentic storefronts
- WooCommerce agentic checkout
- Agentic commerce protocols tracker
- Fake AI shopping agents: how to tell a real one from a scam
- More in Agentic payments
Changelog
- November 11, 2026: Rewrote the opening around which agents can complete a purchase. Added a comparison table, a section on product data and furniture catalogs with a checklist for stores, and new questions. Updated ChatGPT (merchant-owned checkout), Google (Universal Cart US launch, Canada checkout), Amazon (Shop Direct Merchant Workspace) and Copilot (eligibility and onboarding).
- September 25, 2026: Facts checked.




