Why ChatGPT Shopping doesn't show your products

You sell online. Someone asks ChatGPT for the best of what you sell, and the product carousel shows three competitors and not you. Search for a fix and you get the same advice copied across twenty blogs: push your Google Shopping feed, rush to enable Instant Checkout, add an llms.txt. All three aged badly. On March 24, 2026, OpenAI published a course correction almost nobody folded back into their guides: it stepped away from Instant Checkout and put its weight behind product discovery instead. That reshuffles the list of things worth doing this week.

This guide separates what OpenAI actually documents from what gets repeated out of habit, and ends with the only part that matters if you are not Target: what you can move this week.

What changed in March 2026

For nearly a year the story was Instant Checkout — buy inside ChatGPT, never leave the chat. Merchants chased that integration. In its March 24 announcement, OpenAI acknowledged in writing that the initial version of Instant Checkout did not offer the level of flexibility it wanted, that merchants can now use their own checkout flows, and that the company is focusing its effort on product discovery.

Translated: the prize stopped being the buy button and became showing up in the comparison. The lever is the same one it always was — ChatGPT knowing what you sell, at what price, and whether it is in stock — except now that is explicitly OpenAI's priority rather than a stepping stone to checkout.

To get there, OpenAI extended the Agentic Commerce Protocol (ACP), the open standard that connects merchant catalogs to ChatGPT, into discovery. Retailers including Target, Sephora, Nordstrom, Lowe's, Best Buy, The Home Depot and Wayfair already use it for that, and there are integration paths through third parties such as Salesforce and Stripe for merchants who would rather work through systems they already run.

The three doors your products come through

Most of the confusion in this space comes from treating this as one channel. It is three, and they do not compete with each other.

1. Shopify Catalog: automatic

If your store runs on Shopify, your product data is already integrated into ChatGPT through Shopify Catalog. OpenAI states it plainly: no additional work is required from individual merchants. No form, no feed to build, no integration to buy. If someone is charging you to "get you listed in ChatGPT" and you sell on Shopify, you are paying for something you already have.

2. The direct feed: powerful and expensive

The second door is handing OpenAI a structured feed of your own. The schema is publicly documented inside ACP and it is serious: 79 fields, 19 of them required for a non-Ads feed. The required ones are the predictable set — item_id, title, description, url, brand, image_url, price with its ISO 4217 currency code, availability — plus two flags that govern everything else: is_eligible_search, which controls whether the product can surface in ChatGPT search results, and is_eligible_checkout, which enables direct purchase and requires the first one to be on.

The useful fine print: OpenAI's own best-practices guide asks for concise, factual copy, variant IDs that hang off a consistent parent, properly encoded URLs, and — if you want to measure anything — your own attribution parameters on the product URL, for example utm_medium=feed. This is a living feed, not a file you upload once.

3. The open web: the door that applies to everyone

The third one is the door nobody explains and the one that governs most stores. To decide which products to show, ChatGPT considers — per OpenAI's own Help Center — structured metadata from first-party and third-party providers, plus other third-party content. Meaning: your product pages, and also what other people publish about your products.

That is where the classic work still pays. If your product page cannot be crawled, if the price in your schema disagrees with the price on the page, if your product has no reviews anywhere, no feed will rescue you: you are asking the model to bet on a claim it cannot corroborate.

What drives the ranking (per OpenAI, not per LinkedIn)

This is the part almost no guide quotes, even though it is published. When someone taps a product and ChatGPT lists the merchants selling it, that list is ordered — OpenAI says — by factors like availability, price, quality, and whether you are the maker or primary seller of that item.

Read it twice, because the consequences are uncomfortable. If you resell a product the manufacturer also sells, you start that list at a structural disadvantage. If your edge is price, you have a direct lever. And if your stock data is stale, that is not a UX detail — it is a ranking factor.

One more nuance, and it is the kind that saves money: OpenAI notes that ChatGPT may generate simplified product titles and descriptions from the information it receives, because merchants use varying titles for the same product. Labels like "budget-friendly" or "most popular" are model-generated too, and are explicitly not verified statements. Stuffing keywords into a title is doing work so a model can rewrite it.

The detail that changes the math later

Today the product carousel is organic. OpenAI's Help Center says so without hedging: product results are selected independently by ChatGPT, are not ads, and are not influenced by any OpenAI partnerships; ads run in a separate lane.

But the same feed schema that documents those organic results includes an is_ads_eligible flag for marking which products may be processed for ChatGPT ads, with a dedicated Ads feed guide alongside it. Both futures live in one file. That is not an accusation — it is a read on where the board is heading, and one more reason to build the part you cannot rent: clean data and a brand the model can corroborate.

What to do this week if you are not a chain

Seven steps, in order of increasing effort, all verifiable by you:

1. Check whether you are already in. If you sell on Shopify, your products are already integrated. Start there before hiring anyone.
2. Make sure your product pages can be read. No crawl blocks, and the main content in the HTML rather than behind JavaScript. Covered in the technical AISO guide.
3. Audit who you let in via robots.txt. The bot that trains and the bot that surfaces results in ChatGPT are not the same one; blocking the wrong one erases you. Bot by bot, here.
4. Complete your Product schema — brand, identifier (GTIN or MPN), image, offer with price and currency, availability, ratings. Ready-to-paste examples in schema markup for AI search.
5. Sync price and stock across your site, your schema and any feed. A stale price does not just annoy the shopper; it costs you position in the merchant list.
6. Get real reviews. The review summaries ChatGPT shows are drawn from public sites. If nobody talks about your product, there is nothing to summarize.
7. Be the primary seller of your own brand. If your product is cheaper and better documented on a marketplace than on your own site, the ranking will reflect that. A feed does not fix it.

The uncomfortable conclusion

The direct feed is the visible part of the problem and the part everyone wants to sell you. But before the product carousel there is a conversation: someone describes what they need, and the model decides which categories, brands and merchants are relevant at all. If your business does not exist at that layer — if the model does not know what you sell, where you operate or why you are trustworthy — your beautifully structured SKUs are competing for a slot on a list they were never invited to.

Start by finding out what AI says about your store today. Then decide how much engineering the feed deserves.

Frequently asked questions