ChatGPT Shopping: How to Get Products Recommended in 2026

By Minel Gunesoglu, founder of Is My Brand in AI. I build free tools and honest, source-backed guides on getting your brand cited by ChatGPT, Perplexity, Gemini and Google AI Overviews. Last updated June 10, 2026.

TL;DR: ChatGPT Shopping surfaces product recommendations organically in 2026, the Instant Checkout feature ended in March 2026, but the product-answer layer is still active and is not pay-to-play. ChatGPT reads product pages via OAI-SearchBot, so sellers must allow that bot in robots.txt and publish Product JSON-LD with at least aggregateRating, Offer, gtin and mpn. Shopify stores reach ChatGPT through the Google Merchant Center feed; other stacks need manual schema on every product page.

If you want your products to appear in ChatGPT Shopping, three things have to be true: OpenAI's shopping crawler can read your pages, your product data is machine-readable, and independent review signals support the recommendation. That product-answer layer is what people mean by ChatGPT Shopping, and it is the surface this guide covers. This is not a consumer "how to shop with ChatGPT" walkthrough, and it does not cover Google Shopping, Amazon advertising, or general ecommerce SEO. The scope is narrow on purpose: the mechanics of ChatGPT's shopping surface are different enough from those other channels that conflating them is the most common reason sellers waste effort.

The first question almost every founder asks is whether this is pay-to-play or reserved for big brands with media budgets. It is neither. As of June 2026, ChatGPT Shopping recommendations are organic, OpenAI has stated the results are not ads and are not sponsored. A small store with clean structured data and genuine reviews can appear next to a household name. What decides placement is whether ChatGPT can read your product page, whether your data is machine-readable, and whether independent reviews corroborate your claims. Budget does not buy a slot. The rest of this guide is the technical path to earning one.

What ChatGPT Shopping Actually Is Right Now (Honest State, June 2026)

ChatGPT Shopping is the feature inside ChatGPT that returns product recommendations, names, images, prices, ratings, and outbound merchant links, when a user asks a buying-intent question. OpenAI introduced the product-recommendation experience in April 2025, and it has shifted meaningfully since. The most important update for sellers: the Instant Checkout feature, which briefly let users buy directly inside ChatGPT, was discontinued in March 2026, after roughly five months and almost no merchant adoption (CNBC). Guides written in 2025 still describe in-chat purchasing as a live channel. It is not. If a resource you are reading leans heavily on Instant Checkout, treat the rest of it as stale.

What remains active, and what actually matters, is the recommendation layer. ChatGPT still surfaces products, still displays carousels of options, and still sends shoppers to merchant pages to complete the purchase on the brand's own site. The transaction happens on your store, not in the chat. For an ecommerce brand, that is arguably a better arrangement: you keep the customer relationship, the checkout data, and the upsell opportunity, while ChatGPT functions as a top-of-funnel discovery engine.

Scale is the reason to care. Fortune reported roughly 800 million weekly ChatGPT users as of April 2025, and that base has grown since. Even a small fraction asking product questions represents a discovery channel comparable to a major search engine. The behavior is also different from classic search: a ChatGPT shopper has often already described their constraints ("under $150," "wide toe box," "good for trail and road") before any product appears, so the recommendations they see are pre-qualified. A click from ChatGPT tends to arrive warmer than a cold search click. Earning a place in that shortlist is the entire game, and the chapters below are the levers that move it.

How ChatGPT Decides Which Products to Surface

ChatGPT's product selection runs on two distinct signal layers, and understanding the split is what separates a working strategy from guesswork. The first is the product-card / feed layer, structured commercial data (price, availability, GTIN, rating) that populates the visual carousels and product cards. Much of this data is sourced from the Google Shopping ecosystem rather than crawled fresh from each store. The second is the conversational-recommendation layer, the model's underlying judgment about which products genuinely fit the user's described need, shaped by editorial reviews, comparison articles, forum discussions, and the prose on your own product pages.

Most guides describe only the first layer and stop. That is why their advice ("submit a feed, add schema") produces inconsistent results: a brand can have a flawless feed and still be absent from the conversational answer because no independent source corroborates that its product is good. Both layers have to be satisfied. The feed layer makes you eligible to appear as a card; the conversational layer decides whether the model recommends you in prose.

A practical way to think about it: the feed layer answers "does this product exist, at what price, with what rating?" while the conversational layer answers "is this the right product for this specific person?" You influence the first with structured data and a clean merchant feed. You influence the second with review quality, third-party coverage, and product-page copy that explicitly states who the product is for and what it is not for. Brands that win in ChatGPT Shopping invest in both, the machine-readable plumbing and the human-readable evidence that the product is worth recommending. The following sections take each lever in turn, starting with the most overlooked one: whether ChatGPT is even allowed to read your site.

Bot Access: OAI-SearchBot, GPTBot, and Your robots.txt

Bot access is the single most common point of self-inflicted failure, because the relevant crawler is not the one most sellers block or allow by reflex. OpenAI operates separate bots for separate jobs, and conflating them, which nearly every guide does, leads people to block the wrong one. The two that matter here are OAI-SearchBot and GPTBot, and per the OpenAI bot documentation they do different things.

OAI-SearchBot is the crawler that powers ChatGPT's search and shopping features. When ChatGPT pulls a live product page to surface it in a recommendation, OAI-SearchBot is the agent fetching it. If you disallow OAI-SearchBot, your pages can be excluded from the shopping surface entirely, this is the bot you must allow. GPTBot is a different agent used to gather data for training OpenAI's models. Allowing or blocking GPTBot is a content-licensing decision, and it has no bearing on whether your products appear in ChatGPT Shopping. Many brands, wary of "AI scraping," block GPTBot and assume they have made a privacy-conscious choice, which is fine, but some go further and block OAI-SearchBot too, quietly removing themselves from the shopping surface.

Here is a robots.txt that allows the shopping crawler, permits training-data collection (adjust to your licensing preference), and keeps both bots out of pages that should never be indexed:

# Allow OAI-SearchBot — powers ChatGPT search & shopping (your products appear here)
User-agent: OAI-SearchBot
Allow: /products/
Allow: /collections/
Disallow: /cart/
Disallow: /checkout/
Disallow: /account/

# GPTBot — used for model TRAINING, not shopping. Separate decision; does NOT affect ChatGPT Shopping visibility.
User-agent: GPTBot
Disallow: /cart/
Disallow: /checkout/
Disallow: /account/

# Reference: https://developers.openai.com/docs/bots

If you want to permit training-data use of your catalog, leave GPTBot with access to your product paths as shown. If you would rather opt out of training entirely, change GPTBot's block to Disallow: /, and note that doing so still leaves OAI-SearchBot free to surface your products in shopping, because they are governed independently. The mistake to avoid is a blanket User-agent: * Disallow: / on anything in your product tree, which silently catches OAI-SearchBot in the net. Before changing anything live, fetch your own robots.txt at yourdomain.com/robots.txt and confirm OAI-SearchBot is not blocked by any broader rule above it, since robots.txt rules are evaluated by specificity and order.

Product Schema That ChatGPT Actually Reads (JSON-LD Spec)

Product schema is the machine-readable description of your item, and the gap between "passes Google's test" and "ready for ChatGPT Shopping" is wider than most sellers assume. Google's Rich Results Test will return a green pass on a Product block that contains only a name, an image, and a price. That same block can be effectively invisible to ChatGPT's shopping layer, which leans on the commercial identifiers that uniquely pin down a product: gtin (or gtin13/gtin8), mpn, an Offer with price and availability, and an aggregateRating. A passing Rich Results Test is necessary but not sufficient, it confirms valid syntax, not commercial completeness.

The reason the identifiers matter is matching. ChatGPT (and the Google Shopping feed it often draws on) needs to reconcile your product with the same product as it appears across review sites, price comparison engines, and other merchants. A GTIN and MPN are how that reconciliation happens. Without them, your listing floats unanchored, and the model has a harder time trusting that your "Example Trail Runner 2.0" is the same shoe a reviewer praised on another site. The aggregateRating is what feeds the star rating in the carousel and contributes to the conversational layer's sense of whether the product is well-regarded.

Here is a complete Product JSON-LD block with every field the shopping layer cares about. It uses a deliberately fictional product and placeholder values:

<script type="application/ld+json">
{
  "@context": "https://schema.org/",
  "@type": "Product",
  "name": "Example Trail Runner 2.0",
  "description": "Waterproof trail running shoe with a wide toe box, 8mm drop, and a Vibram outsole. Built for mixed trail and road over long distances.",
  "image": [
    "https://example.com/images/trail-runner-2-front.jpg",
    "https://example.com/images/trail-runner-2-side.jpg"
  ],
  "sku": "ETR2-BLK-42",
  "mpn": "ETR2-2026",
  "gtin13": "0000000000000",
  "brand": {
    "@type": "Brand",
    "name": "Example Outdoors"
  },
  "offers": {
    "@type": "Offer",
    "url": "https://example.com/products/trail-runner-2",
    "priceCurrency": "USD",
    "price": "139.00",
    "availability": "https://schema.org/InStock"
  },
  "aggregateRating": {
    "@type": "AggregateRating",
    "ratingValue": "4.6",
    "reviewCount": "212"
  },
  "review": {
    "@type": "Review",
    "reviewRating": {
      "@type": "Rating",
      "ratingValue": "5"
    },
    "author": {
      "@type": "Person",
      "name": "Sample Reviewer"
    },
    "reviewBody": "Held up over 300 miles of wet trail with no break-in period."
  }
}
</script>

Replace every placeholder with your real product data before deploying, the GTIN above (0000000000000) is a non-value placeholder, not a usable code, and aggregateRating must reflect genuine review counts. Fabricated ratings are both a schema violation and a trust risk: ChatGPT's conversational layer can cross-check your stated rating against third-party review platforms, and a mismatch undermines the listing it was meant to strengthen.

Shopify vs. Custom Stack: Two Paths to Structured Data

Your platform determines how much of this you build by hand, and the fork splits cleanly between Shopify and everything else. Shopify stores reach ChatGPT primarily through the Google Merchant Center pipeline rather than through page-level crawling, which means a correctly configured Shopify catalog can be syndicated into the shopping surface with relatively little custom schema work. Shopify already emits Product structured data on many themes, and, more importantly, it integrates natively with Google Merchant Center, which (as the next section explains) is the feed that ChatGPT's product carousels heavily rely on.

For Shopify sellers, the practical work is therefore: ensure your products sync to Google Merchant Center with complete attributes (GTIN, MPN, brand, price, availability, and a condition), confirm your theme outputs valid Product JSON-LD, and verify OAI-SearchBot is not blocked. Many of the page-level identifiers are populated from the same product fields that feed Merchant Center, so the two efforts reinforce each other. One Shopify-specific gotcha: the default Dawn theme does not emit gtin in its Product JSON-LD unless you map the barcode field, so confirm your GTIN is actually present in the page markup, not just in the admin. The Shopify path is closer to "configure correctly" than "build from scratch."

Custom stacks, headless storefronts, bespoke builds, WooCommerce, BigCommerce, Magento, or any platform without native Merchant Center syndication, take the manual fork. On these, you are responsible for emitting the full Product JSON-LD on every product page yourself (the block in the previous section is your template), keeping availability and price in sync with real inventory, and, if you want feed-layer eligibility, setting up Google Merchant Center independently and mapping your catalog to it. The work is greater, but the control is greater too: you decide exactly what each field contains rather than inheriting a theme's defaults. The non-negotiable for both paths is consistency: the price and availability in your JSON-LD, in your Merchant Center feed, and on the page a shopper actually lands on must all agree, because contradictions are a fast route to being dropped.

Google Merchant Center to OpenAI: The Feed Chain

Google Merchant Center is the connective tissue between your catalog and ChatGPT's product cards, and the dependency is larger than most sellers realize. A March 2026 analysis by Semrush and EvolveAMZ found that roughly 83% of ChatGPT shopping carousels pull from the Google Shopping feed, meaning the structured commercial data behind the visual product cards comes, more often than not, through the Google Merchant Center ecosystem rather than from a direct crawl of each store. If you ignore Merchant Center, you are opting out of the layer that populates the most visible part of the shopping experience.

The feed chain runs in a defined order, and each link has to hold:

1. Build a clean product catalog in your store with complete attributes, GTIN, MPN, brand, title, description, price, availability, and condition, for every item.
2. Connect the catalog to Google Merchant Center, either through your platform's native integration (Shopify, BigCommerce) or via a product feed file / Content API for custom stacks.
3. Resolve all Merchant Center disapprovals and warnings. A product flagged for a missing GTIN or a price mismatch will not propagate cleanly. Merchant Center's diagnostics tab is where you fix this.
4. Keep the feed fresh. Stale prices and out-of-stock items that still show as available erode trust in the feed and can suppress your products.
5. Mirror the same data in on-page Product JSON-LD, so that when OAI-SearchBot does crawl a live page, what it reads matches what the feed says.

The reason to do both feed and on-page schema, rather than treating Merchant Center as a shortcut that lets you skip JSON-LD, is that the two layers serve different parts of the system. The feed feeds the cards; the on-page schema feeds the live-crawl checks and the conversational layer's understanding of the product. Brands that rely on the feed alone can appear as a card but lose the prose recommendation; brands that rely on schema alone can be discussed but miss the carousel. The 83% figure tells you which layer is more visible today, but the conversational layer is where the actual buying decision is influenced, so neither is optional.

Where ChatGPT Gets Product Reviews (Sourcing Map)

Reviews are the corroboration that turns an eligible product into a recommended one, and almost every guide tells you to "get more reviews" without saying where those reviews actually need to live. ChatGPT's conversational layer does not invent an opinion of your product from nothing, it synthesizes from the sources it can read. Knowing which platforms carry weight lets you concentrate effort instead of spraying it. Here is the practical sourcing map for an ecommerce brand:

• Google Reviews / Google Business Profile, broad, trusted, and tightly coupled to the same Google ecosystem that feeds the shopping carousels. For most retail and DTC brands, this is the highest-leverage place to accumulate genuine reviews.
• Trustpilot, widely cited in AI shopping answers as a third-party trust signal, especially for DTC brands and online retailers. A healthy Trustpilot profile with volume and recent reviews is frequently referenced when the model assesses brand reputation.
• Reddit, for product categories with engaged communities (footwear, electronics, skincare, tools), Reddit threads are disproportionately influential because they read as unfiltered peer opinion. You cannot manufacture these, but you can earn them by making a genuinely good product and being present where your category is discussed.
• G2 and Capterra, the relevant platforms if you sell software or B2B tools rather than physical goods. For SaaS, these carry the weight that Trustpilot carries for DTC.
• On-site reviews surfaced in your aggregateRating, your own collected reviews still matter, but only when they are machine-readable through schema and corroborated by the off-site platforms above. On-site reviews alone, with no external echo, read as self-reported.

The pattern across all of these: ChatGPT trusts what it can cross-reference. A product with a 4.7 on your own site, a 4.5 on Trustpilot, strong Google Reviews, and a couple of positive Reddit threads presents a coherent, corroborated picture. A product with a glowing on-site rating and no external footprint presents an unverifiable one. Roughly 98% of buyers verify AI recommendations before purchasing, according to a figure reported by PrimeAvenue, treat that as a directional signal rather than a precise measurement, but the direction is clear: shoppers click through and check, so the corroborating reviews need to be real and findable. The goal is not a single high number; it is consistency across the independent sources the model and the shopper can both reach.

Tracking ChatGPT Shopping Traffic: GA4 + UTM Setup

Tracking is where most sellers give up, assuming they need a paid attribution tool to know whether ChatGPT is sending buyers, and they do not. You can measure ChatGPT Shopping traffic with Google Analytics 4 alone, for free, using two methods that work together: referrer-based identification and UTM tagging.

The referrer method works because ChatGPT sends outbound clicks with a recognizable referral source. In GA4, create an exploration or a segment filtered by Session source / medium containing chatgpt.com (and, for completeness, openai.com). This captures organic clicks where you have no control over the link, the ones ChatGPT generates itself when it surfaces your product. Build a comparison segment so you can watch ChatGPT-referred sessions against your other channels over time. Note that referral data can be imperfect when links pass through a redirect, which is exactly why the second method exists.

The UTM method covers any link you do control, for example a product URL you submit somewhere ChatGPT can read, or a campaign link you want attributed cleanly. Tag those URLs like this:

https://example.com/products/trail-runner-2?utm_source=chatgpt&utm_medium=ai_shopping&utm_campaign=product_discovery

Then in GA4, your reports will show chatgpt / ai_shopping as a distinct source/medium, letting you separate AI-driven shopping traffic from regular search and social. To avoid GA4 lumping chatgpt.com referrals into "Organic" or "Unassigned," add chatgpt.com and openai.com as a custom channel group, or at minimum keep them out of your referral-exclusion list. The combination, referrer segment for organic clicks you cannot tag, UTM for links you can, gives you a free, complete-enough view of whether the work in the previous sections is actually producing sessions and conversions. Measure before and after you make changes, so you can attribute lift to the specific levers you pulled.

ChatGPT vs. Perplexity vs. Gemini vs. Google AIO: Shopping Comparison

The four major AI surfaces handle product recommendations differently, and a strategy tuned only for ChatGPT leaves the other three on the table. Each reads structured data, but they weight feeds, live crawls, and reviews differently, and the practical setup overlaps without being identical. The table below summarizes the state as of June 2026:

Surface	How products are sourced	Key data signal	In-chat checkout	Best lever for sellers
ChatGPT Shopping	Google Shopping feed (~83%) + OAI-SearchBot live crawl	Product JSON-LD + GTIN/MPN + aggregateRating; Merchant Center feed	Ended March 2026 (recommendations only)	Merchant Center feed + on-page schema + Trustpilot/Google reviews
Perplexity	Live web crawl + structured data, heavy citation of sources	Clean Product schema + authoritative review/comparison pages it can cite	No native checkout	Be the cited source: strong schema + earn editorial/review mentions
Gemini	Tied tightly to Google Shopping / Merchant Center ecosystem	Google Merchant Center feed + Product structured data	Integrated with Google Shopping flows	A well-maintained Merchant Center feed (overlaps heavily with ChatGPT)
Google AI Overviews	Google index + Shopping Graph	Product structured data + traditional ranking signals + reviews	Routes to Google Shopping	Standard product SEO + Merchant Center + Rich Results-valid schema

The encouraging takeaway is how much the work compounds. A clean Google Merchant Center feed serves ChatGPT, Gemini, and Google AI Overviews at once, because all three lean on the Google Shopping ecosystem. Strong third-party reviews and well-structured Product JSON-LD serve all four, including Perplexity, which rewards being a citable, well-documented source. You are not building four separate strategies, you are building one foundation (machine-readable product data plus corroborated reviews) that pays off across every surface, then tuning the edges. Perplexity's product behavior in particular is worth understanding on its own terms if a meaningful share of your audience uses it; the foundation is shared, but its citation-first style rewards editorial coverage more than the others.

Common Mistakes and What ChatGPT Shopping Is Not

The fastest way to improve is to stop doing the things that actively suppress you, so this final section is a list of failure modes and a clear statement of the boundaries. Each mistake below corresponds to a lever covered earlier; the boundaries exist because misunderstanding scope wastes effort.

• Blocking OAI-SearchBot by accident. A broad Disallow: / or an over-eager bot-blocking plugin catches the one crawler you need. This is the most common silent killer of shopping visibility.
• Treating a Rich Results Test pass as the finish line. A valid Product block without gtin, mpn, and a real aggregateRating is syntactically fine and commercially incomplete.
• Fabricating ratings. A 5.0 on your own site with no external corroboration is a mismatch waiting to be caught, and it can do more harm than no rating at all.
• Relying on the feed alone or schema alone. The feed populates cards; the schema and reviews drive the conversational recommendation. You need both.
• Letting price and availability drift between your feed, your JSON-LD, and your live page. Contradictions get products dropped.
• Building a strategy around Instant Checkout. It ended in March 2026. Any plan that assumes in-chat purchasing is built on a removed feature.

Now the boundaries, what ChatGPT Shopping is not. It is not pay-to-play: as of June 2026 the recommendations are organic, not ads, so there is no slot to buy. It is not the same as Google Shopping, Amazon advertising, or general ecommerce SEO, those overlap with it (especially via the Merchant Center feed) but optimizing them is a separate scope this guide does not cover. It is not an in-chat checkout channel anymore. And it is not a place where budget substitutes for product quality: the corroborated-review requirement means a genuinely good product with real external reviews can outrank a heavily marketed one with a thin trust footprint. That is the honest, slightly unglamorous reality, the work is structured data, clean feeds, and real reviews, done consistently. There is no shortcut, which is also why a small brand that does the work can win.

---

ChatGPT Shopping rewards the same fundamentals across every related surface, and a few neighboring guides go deeper where this one stays focused. If you want the broader picture beyond the shopping carousel, the guide to general ChatGPT ranking covers how the model surfaces brands in non-shopping answers, and the walkthrough on editorial brand citations, not shopping explains how to earn mentions in informational responses. To measure progress, see how to track appearances across models and the wider playbook for AI search visibility for ecommerce. For ongoing monitoring, there are tools that monitor AI product mentions, and for product sites specifically, llms.txt for product sites is worth implementing alongside your schema. If your buyers also use Perplexity, the guide to Perplexity's product surface covers its citation-first behavior in detail.

If you want to know whether ChatGPT, Perplexity, Gemini, or Google AI Overviews currently mention your brand, Is My Brand in AI is opening early access, join the waitlist to be notified when the checker is available. I built it to answer exactly that question, and I update this guide monthly as OpenAI's shopping surface keeps changing.