Products appear in ChatGPT shopping answers through organic selection based on query relevance, product attributes and merchant metadata, with a structured product feed as the only direct input a merchant controls.
OpenAI states that product results are independently selected and are not ads. That leaves two levers: what your feed says, and what your product pages say. Across 2.79 million product mentions we tracked between June and August 2026, the answers name brand owned domains 95.3% of the time, which is the opposite of how ecommerce search usually works.
OpenAI's own documentation describes the selection inputs directly. ChatGPT considers the query and its context, including Memory and Custom Instructions, and can weigh price, availability, reviews, ease of use and other product attributes. Product results are described as independently selected, organic, and not influenced by OpenAI partnerships.
When several merchants offer the same product, ranking can consider availability, price, quality, and whether the merchant is the maker or the primary seller. That last clause explains most of what we see in the data.
There is also a strategic shift worth noting. OpenAI's merchant documentation says it is moving away from a standalone Instant Checkout experience and prioritising discovery plus merchant owned checkout: people find and evaluate products inside ChatGPT, then buy on the merchant's own site or app. Shopping is live in the United States, and purchases that begin in ChatGPT carry no fees.
This is the finding that should change how a retailer plans for this surface. Between 24 June and 31 August 2026 we recorded 2,792,419 product mentions across 6,064 prompts and 887 domains.
| Who gets named | Mentions | Share |
|---|---|---|
| Brand owned domain | 2,661,586 | 95.3% |
| Marketplace | 81,604 | 2.9% |
| Retailer | 45,280 | 1.6% |
| Unclassified | 3,949 | 0.1% |
Marketplaces and multi brand retailers together account for 4.5% of product mentions. In classic ecommerce search, they typically dominate. For an ecommerce team that has spent a decade optimising marketplace listings, that is a genuine change of asset. The "maker or primary seller" rule in OpenAI's ranking documentation appears to be doing real work here, and it means a brand's own product detail page is the asset that matters, not its listing on someone else's platform.
The concentration figures point the same way, and they are severe.
| Domains | Share of all product mentions |
|---|---|
| Top 10 | 70.8% |
| Top 50 | 84.3% |
| Top 100 | 89.9% |
Compare that with citations in ordinary AI answers, where the ten most cited domains hold just 6.2% of all citations and it takes 5,000 domains to reach 57.9%. Editorial AI search is a long tail. AI shopping is a short head. The two surfaces reward completely different strategies, and treating them as one project is the most common planning mistake we see. Which sources an answer leans on outside the product block is a separate question, answered by AI citation tracking.
One consolation: the answers themselves are generous. A shopping answer named 14.8 products on average, with a median of 16 and a maximum of 64. There is room on the shelf. The problem is that the same domains keep filling it.
Direct feed onboarding is available to approved partners, and merchants can apply. If you sell on Shopify or Etsy your catalogue is already integrated, so no separate application is needed. The same is true of the analytics side, where platform integrations decide how much joining up you have to do yourself.
OpenAI recommends a complete feed snapshot at least daily by file upload, with intraday changes sent through the API. Promotions can only be supplied through the API, not through the file.
| Setting | Requirement |
|---|---|
| Preferred format | Parquet, ideally ZSTD compressed |
| Also supported | JSONL.GZ, CSV.GZ, TSV.GZ, UTF-8 required |
| Sharding | Up to 500,000 items per shard, files under roughly 500 MB |
| Filenames | Overwrite a stable filename rather than replacing it |
| Retention | The last processed record is kept up to 14 days if a product is temporarily absent |
| Getting started | Begin with roughly 100 items before automating the full feed |
The required core fields on every row are id, title, description, link, image_link, availability, price and brand, with a header row in canonical lowercase underscore separated names, one product or variant per row, valid HTTP or HTTPS URLs, and a three letter currency code in the price.
The validation limits catch people out:
| Field | Limit |
|---|---|
| item_id | 100 characters |
| title | 150 characters |
| description | 5,000 characters |
| brand | 70 characters |
| GTIN | 8 to 14 digits when supplied |
| Product URL | Must resolve, HTTP 200 required |
A valid GTIN or MPN is required when identifier_exists is omitted or set to yes. And a detail that surprises marketplace sellers: OpenAI uses the registered merchant name as seller identity, and an uploaded seller_name cannot override it.
Three flags control where a product can appear: is_eligible_search for appearing in ChatGPT search, is_eligible_checkout for direct purchase, and is_ads_eligible for ads processing. That third flag is the seam between organic product results and the paid block, which AI ads tracking records separately. Checkout eligibility depends on search eligibility.
Here is the part to read twice. OpenAI's file upload documentation states that products accepted through file upload have search enabled and checkout disabled. Uploading a catalogue does not switch on in chat purchasing, and a merchant who assumes it does will spend a quarter waiting for a conversion path that was never active.
For direct checkout eligibility the feed schema additionally requires seller privacy policy and terms of service URLs. Marketplace feeds can identify a marketplace_seller.
For ordinary content, structured data is a much weaker citation lever than most checklists claim. For shopping, the situation is different in an unexpected way: OpenAI's documentation does not identify schema.org or JSON-LD as a required merchant integration at all. What it identifies is its own product feed schema, plus a Google compatible delimited feed path for approved registrations, where supported Google fields are mapped into OpenAI's normalised schema. That compatibility path does not cover every program, market or attribute.
The flat file specification is also explicit that structured title and description alternatives are not mapped, and that plain title and description values are required.
Your product pages still matter. OpenAI says selection considers structured metadata from first and third party providers plus other third party content, and that titles, descriptions, labels, review summaries and ratings may be generated or summarised from information supplied by third parties, and are not necessarily verified.
That last sentence is worth a policy. If a third party description of your product is wrong, it can reach a shopping answer without passing through you. Knowing what the answers currently say about your products is the only way to catch it, which is what tracking product visibility in ChatGPT shopping is for: it records which products and merchants get named on your commercial prompts, and how that changes.
OpenAI launched Instant Checkout on 29 September 2025 for US ChatGPT Free, Plus and Pro users buying from US Etsy sellers, with more than one million Shopify merchants named as coming soon, including Glossier, SKIMS, Spanx and Vuori. The initial system supported single item purchases.
Underneath it sits the Agentic Commerce Protocol, an open standard co developed by OpenAI and Stripe and published under the Apache licence. The design point matters for anyone worried about disintermediation: the merchant remains the merchant of record and keeps control of products, checkout, fulfilment, returns and the customer relationship, using existing payment infrastructure. The launch used Stripe's Shared Payment Token, scoped to one merchant and one cart total, so ChatGPT can initiate a payment without ever seeing card credentials.
Given the shift back toward merchant owned checkout described above, treat ACP as the plumbing for agentic purchasing rather than as the destination. The discovery step is where the volume is.
Our figures come from Finseo tracking data, aggregated across accounts, with no customer, project or private domain identifiable in any number. Product mentions were recorded between 24 June and 31 August 2026 across 6,064 distinct prompts. Seller type classification is ours: brand means the domain of the product's own maker, marketplace means a multi seller platform, retailer means a multi brand reseller.
Prompt sets are chosen by the customers who run them, so the corpus skews toward the categories those customers sell in and is weighted toward the United States, where ChatGPT shopping is live. It is a large sample, not a random one, and the seller type split in particular should be read as a strong signal in these categories rather than a universal law of AI shopping.
Every external figure comes from OpenAI's own developer and help centre documentation or from Stripe, with the retrieval date stated, because this specification is changing quickly.
Is there a ChatGPT for shopping? Shopping is built into ChatGPT search rather than being a separate product. OpenAI describes product results as independently selected and organic, not advertising.
How does ChatGPT choose which products to recommend? From the query and its context, weighing price, availability, reviews and other product attributes. Where several merchants sell the same item, it can consider availability, price, quality and whether the merchant is the maker or primary seller.
How do I get my products into ChatGPT? Through a product feed. Shopify and Etsy catalogues are already integrated. Other merchants apply for direct feed onboarding and then supply a daily snapshot with intraday updates through the API.
Does schema markup get my products into ChatGPT shopping? OpenAI's merchant documentation does not name schema.org or JSON-LD as a required integration. It specifies its own feed schema plus a Google compatible feed path. Your product pages still feed selection through other routes.
Does uploading a feed enable buying inside ChatGPT? No. Products accepted through file upload have search enabled and checkout disabled. Direct checkout requires separate eligibility, plus privacy policy and terms of service URLs in the feed.
Do marketplaces dominate AI shopping answers? Not in our data. Brand owned domains took 95.3% of 2.79 million product mentions, against 2.9% for marketplaces and 1.6% for multi brand retailers.
Ours
External
Disclosure: the tracking figures in this article come from Finseo, an AI visibility platform. External figures are attributed to their original source.
