AI Shopping in 2026: Getting Discovered in AI and Closing the Sale on Your Own Store

A shopper asks ChatGPT for trail runners under $150 that fit a wide foot. The assistant compares options, reads product pages, and returns a short list with reasons attached. The shopper taps one and lands on your product page, inside the ChatGPT app, already most of the way to a decision. Everything that happens from that tap forward is running on your store, on your theme, through your checkout.

That handoff is the shape of AI shopping for most merchants in 2026, and it got sharper in March. On March 24, 2026, OpenAI stepped back from running checkout in ChatGPT (opens in new tab), saying: “We’ve found that the initial version of Instant Checkout did not offer the level of flexibility that we aspire to provide, so we’re allowing merchants to use their own checkout experiences while we focus our efforts on product discovery.” OpenAI said that for merchants on Shopify, product data is already integrated into ChatGPT through Shopify Catalog, and that no additional work is required from individual merchants.

So the discovery half arrived switched on, and the closing half landed back in your hands. The protocol plumbing underneath all of it is covered in how agentic commerce actually works on Shopify.

OpenAI stepped back from running the transaction itself and pointed its effort at discovery. It didn’t remove Instant Checkout. For some eligible products and merchants, ChatGPT may also show an Instant Checkout option (opens in new tab) that lets the shopper complete checkout in ChatGPT instead of leaving for the merchant’s site. Narrowed and made optional is a different fact from removed, and the difference matters if you’re planning around it.

For Shopify merchants specifically, stores may appear to ChatGPT users when relevant (opens in new tab), and those users check out on the merchant’s online store. On the Shopify side, buyers are redirected to complete the purchase in your online store checkout.

There’s no ChatGPT checkout toggle in your admin, because ChatGPT sends Shopify buyers to your online store checkout, which we covered in the agentic commerce piece. What matters here is what runs instead: your own checkout, with every setting you already control.

The protocol underneath didn’t slow down. The Agentic Commerce Protocol published spec version 2026-04-17 (opens in new tab), three and a half weeks after the announcement, adding cart, feed, orders, delegated authentication and an MCP transport binding. Discovery got a spec expansion; the in-ChatGPT checkout use of it got deprioritized.

Buying on your own store isn’t the consolation prize. Here’s what a Shopify-powered direct checkout in Microsoft Copilot (opens in new tab) doesn’t support. Copilot isn’t the only channel with that toggle, since Shopify has direct checkout for Google AI Mode and Gemini, Microsoft Copilot and Meta, and Shopify lists the same unsupported product types, delivery methods and pixels for the Google and Meta checkouts.

Read that list against your own revenue mechanics. A subscription brand, a build-your-own-box brand, a made-to-order brand, a BOPIS retailer, and anyone running a points program are all looking at a checkout that can’t carry the thing their model runs on. Turning direct checkout off means customers still discover products and get redirected to the online store to complete the purchase.

That direct checkout does support Shopify Functions for shipping, discounts and cart and checkout validation, and any automatic discounts or discount codes you’ve set up work. The order still reaches your admin with attribution, on the server-side pixels described below.

So for a merchant selling a single simple SKU with no subscription, no bundle, no configurator, no loyalty tier and no BOPIS, every loss on that list is null. What’s left is client-side analytics and a subset of accelerated wallets, traded against a buyer who already has a wallet loaded on that surface. For that merchant, direct checkout probably wins, and the advice below to audit your toggles should end with you leaving it on.

The trade only tilts the other way once a retention program is involved, which is where the account sign-in restriction and the loyalty blocks start costing real money.

Google Analytics and custom pixels don’t fire in Microsoft Copilot’s direct checkout. The checkout fires only server-to-server pixels for checkout started and completed, so none of your standard or custom client-side pixels run. Every client-side measurement you built on that checkout, your GA4 view of the funnel, your remarketing audiences, your custom pixel events, has a hole in it shaped exactly like your newest channel, and the server-side signal that survives won’t rebuild an audience.

That’s a reason to look hard at any direct-checkout toggle you’ve left on by default, and a reason the merchant-hosted path has an advantage that doesn’t show up in any announcement: your own checkout fires your own tracking, the way it always has. The measurement question is the same one behind all of our checkout optimization work, and it doesn’t get easier when the transaction happens somewhere you can’t instrument.

Direct checkout in AI channels is active where supported, and the default Allow Shopify to manage for me setting (opens in new tab) activates direct checkout for the relevant channels and auto-enrolls you in future ones. In Microsoft Copilot specifically, direct checkout is activated by default for eligible stores (opens in new tab).

The one place the default runs the other way is the Agentic plan, which is for businesses that don’t have a Shopify online store at all. Direct checkout is deactivated by default there (opens in new tab), and Shopify says you retain full ownership of the customer relationship and post-purchase experience for orders placed through AI channels.

So if you have an online store, the capability losses above are switched on right now unless somebody turned them off. Go and look.

On March 24, 2026, Shopify described shopping in ChatGPT (opens in new tab) in marketing terms: “Hundreds of millions of ChatGPT users can now shop from Shopify merchants without ever leaving their chat.” On desktop, ChatGPT links out directly to the merchant’s store in a separate browser tab. On mobile, the full online store checkout displays in a ChatGPT in-app browser.

So “without ever leaving their chat” describes a webview. It’s your checkout page, your payment methods and your checkout customizations, rendered inside the ChatGPT app frame on a phone, and a plain new tab on a desktop. Shopify says merchant customizations carry over, including brand experience, pricing logic, payment methods and checkout customizations, which is the same fact stated as a benefit.

Shopify’s VP of Product put the whole arrangement this way in OpenAI’s own announcement:

Open ChatGPT on a phone, find your own product through it, and put a real order through your checkout inside that in-app browser. Whatever your theme, your express wallets, your address autocomplete and your one-page checkout extensions do inside that frame is what a pre-qualified buyer experiences, and it’s the last surface between an assistant’s recommendation and your revenue.

ChatGPT’s shopping research is the clearest description of what an AI shopping assistant does before the buyer reaches you. Shopping research (opens in new tab) helps a shopper explore and compare products in ChatGPT through an interactive discovery experience, and OpenAI says it’s designed for decisions that involve comparisons, trade-offs, or multiple constraints. The output is a buyer’s guide with a small set of top picks, each with a rationale explaining why it matches the shopper’s needs, key strengths, tradeoffs, and links to merchants who currently sell it.

OpenAI says results are organic and based on publicly available retail sites, with the assistant reading product pages directly and citing sources. Your PDP isn’t only a conversion asset in this channel. It’s an input to the recommendation itself, read live, which is why the discovery half and the closing half of AI shopping run through the same page.

That changes what the page has to do. The buyer has already seen the comparison, the tradeoffs, and the reason your product was picked, so restating your value proposition answers a question they’ve stopped asking. What they still don’t have is fit detail, sizing specifics, the return window, the shipping date, and whatever the assistant hedged on. Put those above the fold and treat the pitch as already made.

ChatGPT may generate simplified product titles and descriptions (opens in new tab) based on information it receives from third-party providers. OpenAI’s stated reason is to make results easier to read, since merchants often use varying titles and descriptions for the same product. The merchandising copy you tuned isn’t necessarily the copy the shopper reads in the carousel.

OpenAI says of the order merchants appear in: “Merchants are ranked based on factors like availability, price, quality, and whether they are the maker or primary seller of that item.” It also says “Product results are selected independently by ChatGPT and are not ads, nor influenced by any OpenAI partnerships.” Being the maker or primary seller is a data question, which means your feed and your structured data carry weight the copy can’t, and that’s the ground covered in product feed and structured data for AI search.

When merchants update their pricing or shipping terms, there may be some delay before it’s reflected in the information the shopper sees. A shopper who was told $129 and lands on $149 bounces, and the assistant that recommended you takes the hit alongside you. Price and availability freshness is the maintenance job behind this channel.

OpenAI says that when a shopper has ChatGPT memory turned on, shopping research can use what it already knows about them in ChatGPT to better tailor suggestions. For example, if it knows the shopper is interested in gaming, it can factor that into laptop or accessory recommendations. The product carousel works the same way: ChatGPT considers the query and context such as Memory or Custom instructions.

You can’t see any of it. The personalization happens on OpenAI’s side, against a profile you don’t hold, and OpenAI states that chats are never shared with retailers. What you control is whether your product record is complete enough to match a constraint the shopper stated three messages ago.

The admin’s agentic home (opens in new tab) reports four things per AI channel: sales over a period, orders over a period, sessions from that channel to your online store, and conversion rate, calculated as sessions that completed checkout divided by sessions.

Those are first-party numbers from your own store, current, and specific to your catalog. Rank them above every published statistic in this space, including the vendor figures below. Shopify’s search preview tool scores listing quality across five dimensions: description completeness, image coverage, product reviews, variant data, and store policy completeness. Two caveats: this reporting isn’t available for headless stores yet, and the sales figure aggregates referral-based sales together with direct-checkout sales rather than separating them.

Where do ChatGPT orders actually land? Not under a channel the way a Shop or Copilot order does. Orders from AI channels display in your admin with channel or referrer attribution (opens in new tab), and Shopify said on March 24, 2026 that ChatGPT orders flow into the admin with ChatGPT referral attribution (opens in new tab).

That “or” is doing real work. ChatGPT hands the buyer to your store, so the sale is an online store order carrying a referrer, not an order belonging to a separate sales channel, while direct-checkout channels create channel-attributed orders. If you go looking for a ChatGPT channel filter and find nothing, the revenue isn’t missing, it’s attributed by referrer instead. Start from the agentic performance reporting in Sales channels and work outward.

When an order is placed in an AI channel offering direct checkout, that channel gets order details (opens in new tab) including the customer’s name, email address, phone number and physical address. Shopify says AI channels don’t get your full order history, your orders from other sales channels, or your general customer database.

When the sale completes on your own checkout, none of that disclosure question arises, because you get what you always get, on your own terms of service. Whether marketing consent transfers with a direct-checkout customer record is a question for your legal team, so don’t assume either answer before they look at it.

Adobe reported on April 16, 2026 (opens in new tab) that in March 2026, AI-referred traffic to US retail sites converted 42% better than non-AI traffic, against March 2025 when the same traffic converted 38% worse. The reversal is the finding, not the level. Adobe states the figure is based on direct online transactions covering more than a trillion visits to US retail sites, plus a companion survey of more than 5,000 US respondents.

Read the number precisely. 42% better is relative to non-AI traffic on the same sites, not 42 percentage points and not an absolute conversion rate. Adobe also reported Q1 2026 AI referral traffic to US retail up 393% year over year, and that retail homepages average a 75% machine-readability score against 66% for product pages. Adobe sells analytics tooling and an AI content visibility checker, and that last framing sells the last one directly, so weigh it accordingly.

Salesforce reported on January 8, 2026 (opens in new tab) that 2025 global online holiday sales came to $1.29 trillion and that AI and agents powered 20% of all retail sales, fueling $262 billion in revenue. Salesforce describes this as drawn from the activity of more than 1.5 billion global shoppers across more than 89 countries, using its own Agentforce 360, Agentforce Commerce, Agentforce Marketing and Agentforce Service data with factors applied to extrapolate to the broader industry.

That 20% is an influence-attribution model over Salesforce’s own customers, not the share of orders that arrived through an AI link, and treating it as referral share would overstate the channel by a wide margin.

The useful figure in that report for anyone weighing this channel is the comparison: Salesforce found shoppers arriving from AI search channels converted nine times more often than those arriving from social media referrals. Each of these numbers comes from a company that sells something worth more when the number is bigger. Your own admin doesn’t have that problem.

Does blocking AI crawlers remove your products from ChatGPT? On Shopify, no. Blocking AI crawlers at the robots.txt or network layer affects open-web discoverability only (opens in new tab), and doesn’t stop product data being sent by Shopify Catalog to the agentic storefronts you’ve activated. Robots.txt rules are directional and advisory, and not all crawlers are guaranteed to follow them.

There are two pipes, and robots.txt governs one of them. The catalog pipe has its own control in Sales channels, and its own lag, which we covered in the agentic commerce piece. If you blocked the bots last year and believe you’re out of ChatGPT, check the channel settings instead of the robots file.

The other half of the picture is what blocking costs you. Some retailers block automated access to their sites. OpenAI says: “When this happens, shopping research will either skip those sources or rely on other sites with similar products.”

A block doesn’t remove the category from the buyer’s guide. It removes you and puts a comparable product from a site that didn’t block in the slot you would have held.

So blocking is neither a clean opt-out nor a harmless one, which is the shape of most crawler decisions once you look at both sides. Which agents to leave alone and which controls actually gate visibility is the subject of our answer engine optimization work.

OpenAI’s move wasn’t an industry retreat. Here’s where the purchase completes on each major surface.

Google went the opposite way over the same nine months. On November 13, 2025, it described a buy-for-me flow (opens in new tab) that tracked price and, when it hit the shopper’s target, bought the item on the merchant’s site with Google Pay after the shopper confirmed. By May 19, 2026, Google described shoppers checking out with Google Pay in a few taps (opens in new tab), with Universal Cart rolling out across Search and the Gemini app in the US and YouTube and Gmail to follow, and the brand remaining merchant of record. Even there, Google offers an alternative path where shoppers transfer their items to the merchant’s site to complete the purchase.

Google limits eligibility for the checkout experience (opens in new tab) to products eligible in the United States, Canada and Australia, and to participating merchants and partners, requiring an interest form, a technical implementation, and the native_commerce(checkout_eligibility) product attribute. Form plus build plus attribute plus participation is an allowlist. For a mid-market brand it isn’t a switch you flip this quarter.

Microsoft shipped in-surface checkout and still runs it. Copilot Checkout (opens in new tab) runs on Copilot.com and the Copilot mobile app, limited to English-language merchants selling to US buyers in USD, with a product feed required to power both discovery and checkout. Microsoft lists onboarding through Microsoft Merchant Center, PayPal’s Store Sync, or Stripe’s Agentic Commerce Suite, and says of Shopify merchants that they’ve been automatically enrolled, with controls in the Shopify admin and no additional onboarding steps required.

If you’re on Shopify, you aren’t deciding whether to join Copilot Checkout. You’re deciding whether to leave it on. Microsoft says it can’t give a timeline for regional and language expansion.

On November 25, 2025, Perplexity described checkout completing inside Perplexity (opens in new tab) through a PayPal partnership, free to all US users, with retailers remaining merchant of record. That describes November 2025 and nothing later. Separately, Shopify named Perplexity in its Agentic Storefronts launch (opens in new tab) on December 10, 2025.

Anthropic took a third route. On September 2, 2026, it published a blueprint (opens in new tab) with harnesses, patterns, guardrails and reference implementations of a shopping agent and a merchant agent, aimed at businesses building agents into their own apps and sites. An AI shopping assistant running on your own property, against your own catalog, is a real 2026 direction rather than a hypothetical.

Some 2026 capabilities were announced in future tense. Shopify announced Shop Pay coming soon to Copilot on March 24, 2026. Google announced UCP checkout for the UK after Canada and Australia. Those are absences at a point in time, not withdrawals, and the discovery half shipped to everyone while the transaction half stayed gated.

Amazon’s robots.txt (opens in new tab) disallows essentially every known AI crawler and agent user agent with a blanket Disallow: /. The list includes agent-specific tokens distinct from ordinary crawlers: PerplexityBot, Perplexity-User, ChatGPT-User, OAI-SearchBot, ClaudeBot, Claude-User, GPTBot, Google-Extended, GoogleAgent-Mariner, GoogleAgent-Shopping, CopilotNative, CopilotSapphire and more. Those last several are worth noticing on their own terms: the platforms now ship named shopping agents that a site owner can address individually.

At the same time, Amazon runs an agent that buys on other retailers’ sites. On May 13, 2026, Amazon announced Alexa for Shopping (opens in new tab), folding Rufus, which Amazon says helped over 300 million customers in 2025, together with Alexa+ into a free US assistant requiring no Prime membership or Echo device. It includes a Shop Direct feature for shopping other retailers across the web where a Buy for Me capability handles the entire purchase on the shopper’s behalf on the other retailer’s site.

Your store is a destination for that agent, and you can’t reciprocate. Amazon’s robots.txt tells you what Amazon disallows rather than whether any agent complies, and the rules for all of this are unsettled enough that the parties have already been to lawyers. Amazon’s outside counsel sent Perplexity a cease-and-desist letter dated October 31, 2025 (opens in new tab) demanding it stop deploying Comet’s agents on Amazon and alleging violations of the Computer Fraud and Abuse Act, and Perplexity responded publicly on November 4, 2025 (opens in new tab) framing Comet as a user agent operating with the user’s own credentials. There’s no settled outcome.

None of this requires you to predict which protocol wins. Every surface, in every configuration above, rewards the same two assets: a product record complete and current enough to be recommended, and a page and checkout good enough to close a buyer who arrives mid-decision.

The competitor beating you in an AI shopping result usually isn’t better than you. Their product data was complete, their page answered the one question the assistant left open, and their checkout didn’t lose the buyer in an in-app browser. That’s three fixable things, and the assistant handed them the same buyer it would have handed you.

If you want to know which of the three is costing you, that’s what a free Growth Audit is for.

We look at what an AI shopping assistant can read from your catalog, what happens to the buyer it sends you, and where the revenue is going instead. Start with checkout optimization, or book the Growth Audit and we’ll show you the gap.