ChatGPT Shopping in 2026: How Merchants Win Visibility After Instant Checkout

More than 1 billion people use ChatGPT every week. OpenAI has spent the past year turning that audience into a shopping channel, and in March 2026 it changed the rules: the company dropped Instant Checkout and refocused on product discovery.

That decision invalidated much of the older optimization advice. Merchants now send shoppers to their own checkout. Ecommerce brands that understand the current system can influence purchases at the evaluation stage — if they know what ChatGPT actually reads.

What is ChatGPT shopping?

ChatGPT shopping shows product recommendations directly in chat when a user asks a shopping query, such as "Give me recommendations for best cat water fountain." Results appear in product cards with images, prices and short descriptions. ChatGPT may add labels like "Budget-friendly" and summarize what reviewers like and dislike.

OpenAI launched the feature in 2025, saying ChatGPT would retrieve products from across the web and recommend them based on third-party structured metadata like pricing and reviews.

Months later, OpenAI introduced Instant Checkout, letting users buy without leaving ChatGPT. The system ran on OpenAI's Agentic Commerce Protocol (ACP), which sent payment information to the merchant. In March 2026, OpenAI stepped back from Instant Checkout. The official announcement cited a lack of flexibility and said the company would focus on improving product discovery. OpenAI expanded the ACP so companies could share products and promotions.

ChatGPT shopping is currently live only in the U.S., with plans to expand. Merchants can build their own ChatGPT app, as Walmart did, taking shoppers into a Walmart experience with loyalty benefits and payments.

One point OpenAI states plainly: you cannot pay to appear in these results. Shopping results are not ads, and OpenAI labels and separates ads from its answers.

How do products appear in ChatGPT?

Two paths exist. Approved merchants can submit a product feed directly to OpenAI, keeping pricing current. Shopify and Etsy merchants skip the application — their product data is already integrated.

The second path requires nothing. ChatGPT pulls product data from your website and third-party providers. As long as OpenAI's crawlers can access your site, your products can appear in recommendations.

What technical foundations matter?

Before optimizing, merchants need four things in place.

• Server-side rendering. A 2024 Vercel test found OpenAI's crawlers don't read JavaScript, and a 2026 experiment found ChatGPT summarized only the raw HTML of a shared URL. Price, stock and key specs must live in the HTML your server sends.
• Product schema markup. At minimum, include the properties Google requires: name, image URL and the product offer. If you submit a feed, schema must match it.
• Robots.txt. Don't block OAI-SearchBot, which surfaces webpages in responses. You can still block GPTBot to keep content out of ChatGPT training.
• CDN configuration. Some CDNs, like Cloudflare, let users block AI crawlers. Check firewall and bot rules allow OAI-SearchBot.

How does the merchant feed work?

Applicants provide personal and company information, primary product categories and feed size in SKUs, plus confirmation the feed meets OpenAI's specifications. Approval puts you on an onboarding waitlist.

Feeds move via ACP over SFTP, through third-party providers like Salesforce and Stripe, by file upload or through an API. OpenAI recommends one full daily upload via file, with the API handling intraday updates. For smaller feeds, the API can do both. Promotions data must go through the API.

The specification requires nine fields: item_id, title, description, url, brand, seller_name, image_url, availability and price. Titles cap at 150 characters, descriptions at 5,000. OpenAI asks merchants to use the brand name as it appears on the page and, when possible, a URL that opens the specific variant.

Why does Google Shopping ranking matter?

The products ChatGPT recommends overlap heavily with Google Shopping results. A Semrush experiment found the top product ChatGPT recommended appeared in Google Shopping's top three 75% of the time. A March 2026 Peec AI study found about 83% of products in ChatGPT carousel positions also appeared in Google's top 40 products, excluding paid ads.

Google advises merchants to confirm free listings are on, write accurate titles matching the product page, add GTIN and MPN identifiers, upload more images, fix Merchant Center feed errors in Diagnostics, and collect product reviews for star ratings. Google also suggests updating the feed and website simultaneously to avoid data discrepancies.

How should product pages be structured?

AI crawlers read text and code. Display price and stock status as text. Use HTML tables for specifications, not images. Keep a clear heading hierarchy: H1 for the product name, H2s for "Product Description" or "User Reviews." Don't present reviews only as images or video — Beauty By Earth's self-tanner page pairs a video review carousel with written reviews underneath that OpenAI's crawlers can read.

Deepen schema beyond the minimum. Google's recommended properties — reviews, material, color, description, shipping details, item condition, return policy and GTIN — feed both Google Shopping visibility and ChatGPT's picks. For product variants, use ProductGroup alongside Product.

Optional feed fields add context OpenAI hasn't confirmed improves visibility: condition, material, dimensions, age_group, additional_image_urls, review_count, star_rating, accepts_returns and return_policy. That data helps ChatGPT match products to specific queries like "new toddler shoes made from rubber."

What is query fan-out?

ChatGPT breaks one prompt into multiple sub-queries, then combines the results. When testers asked "what's the best cat water fountain?", one generated sub-query was "best cat water fountain stainless steel" — the material was never mentioned in the prompt.

Product copy should answer that fan-out directly. Conversational descriptions like "Made from stainless steel, this cat water fountain doesn't hold onto odors like plastic and comes apart easily for cleaning" give ChatGPT extractable facts. Cover use case, budget and comparison. Avoid vague copy like "elegant water fountain all cats will love" — it gives ChatGPT nothing to return to the user.

Do off-site signals count?

Yes. When a user asks "Compare this water fountain to the Petlibro fountain," ChatGPT visits review sites — in testing, it cited Petlibro's website and a FountainPaws review. Brands absent from those pages, or covered with outdated information, risk losing the sale to a competitor.

Trust-building tactics include detailed reviews on Google, Trustpilot and Amazon; responding to journalist requests on platforms like Featured; and pitching publications that haven't covered the product well.

How do you measure performance?

Track a fixed set of shopping prompts on a schedule — Semrush's AI Visibility Toolkit supports this with ChatGPT Search as the platform. Filter analytics for ChatGPT referral traffic, tagged "utm_source=chatgpt.com" or via HTTP referrer. Referrer data can be stripped by browsers and recorded as direct traffic, so watch product-page direct traffic for unusual growth. Post-purchase surveys offering ChatGPT as a discovery option close the gap.

ChatGPT shopping changed within a year of launch, and more shifts are coming: expansion beyond the U.S., Google's rival Universal Commerce Protocol, and an OpenAI self-serve merchant platform promised later in 2026. Feed specs and checkout options will keep changing. What lasts — and what to monitor next — is crawler access, product data that matches your site, and trust signals from third-party pages cited in AI responses.