ChatGPT Shopping: How to Submit Your Product Feed (and Whether You Need To)

ChatGPT serves over 700 million weekly users. A meaningful percentage of those interactions are now product research — questions like “what’s the best running shoe for marathon training under $150?” or “I need a gift for someone who loves ceramics, budget around $40.”

When ChatGPT answers these questions, it surfaces products. The merchants whose products appear are the ones whose data is in OpenAI’s index — either through direct feed submission via the Merchant Program, or through partner integrations with Shopify and Etsy.

This guide covers everything you need to know about getting your products into ChatGPT Shopping: the feed specification, the application process, the Agentic Commerce Protocol, and the question Shopify and Etsy merchants are asking — do I actually need to submit a feed if I’m already auto-integrated?

The short answer is: it depends on how serious you are about ranking. The longer answer is below.

What ChatGPT Shopping is

ChatGPT Shopping launched in September 2025 with the introduction of Instant Checkout and the Agentic Commerce Protocol (ACP). The system works in three layers:

Discovery. When a user asks ChatGPT a shopping question, the system surfaces relevant products with images, descriptions, pricing, and direct links. Results are not ads — OpenAI has been explicit that product placement is based on relevance to the query, not paid placement.

Feed ingestion. Merchants share product data with OpenAI through structured feeds (CSV, TSV, XML, or JSON) that get ingested and indexed every 15 minutes. The feed is the source of truth — ChatGPT trusts the merchant feed over scraping the live site.

Checkout. Some products support Instant Checkout, where the user can complete the purchase inside ChatGPT. The transaction is powered by the Agentic Commerce Protocol with Stripe handling payments. Merchants stay the merchant of record. As of early 2026, OpenAI is moving away from a standalone Instant Checkout experience and prioritizing better discovery with merchant-owned checkout — so most merchants will see ChatGPT users discover their products in-chat and complete the purchase on the merchant’s site.

The launch retailers include Target, Sephora, Nordstrom, Lowe’s, Best Buy, The Home Depot, and Wayfair. Beyond those flagship integrations, OpenAI runs a structured Merchant Program for any business that wants to participate.

Eligibility

ChatGPT Shopping is currently US-only — both for shoppers and for merchants. OpenAI has stated plans to expand to additional regions over time, but as of early 2026, you need to sell and ship to US customers to participate.

There’s no minimum size requirement. The Merchant Program is open to ecommerce businesses across most categories, with the standard exclusions for prohibited content: adult products, restricted substances, prescription medications, weapons, gambling, and other categories OpenAI flags as not appropriate for a general audience.

Two paths into the program:

Auto-integration via Shopify or Etsy. If you sell through Shopify Catalog or Etsy, your products are already integrated. No application required. We’ll come back to whether this is enough.
Direct feed submission via the Merchant Program. For non-Shopify, non-Etsy merchants. Apply at chatgpt.com/merchants for direct feed access.

The application process for direct submission

If you’re not on Shopify or Etsy, the application flow is:

Submit business details at chatgpt.com/merchants for verification. Standard business information: company name, registration details, website, contact info, estimated order volume.
OpenAI reviews and approves merchants for feed ingestion. Review timelines vary; approved merchants are added to a queue for onboarding.
Receive SFTP credentials. Approved merchants get a unique, secure SFTP endpoint with authentication credentials for feed delivery.
Build and submit a sample feed. Start with approximately 100 representative products following OpenAI’s Product Feed Specification. Include all required fields. OpenAI validates the feed against the specification before approving for full ingestion.
Configure refresh cycles. OpenAI accepts feed updates as often as every 15 minutes. Configure your feed export to push updates on whatever cadence makes sense for your inventory velocity.

The full process from application to live feed ingestion typically takes 2-6 weeks, depending on application volume and how quickly you can complete the technical work.

The Product Feed Specification

This is the part that matters most for merchants serious about ranking. OpenAI publishes a detailed Product Feed Specification at developers.openai.com/commerce/specs/feed/. The feed is a structured file in CSV, TSV, XML, or JSON format with attributes grouped by category.

If you’ve worked with Google Merchant Center feeds, much will feel familiar. Many core fields overlap. But the OpenAI feed has unique attributes you won’t find in Google’s spec, and getting them right materially affects ranking.

Required fields

Basic data:

Product ID (unique identifier)
SKU
Product title
Description (accepts plain text, HTML, and markdown simultaneously on the same product — a unique feature compared to other feeds)
Product URL on the merchant’s site
GTIN (UPC/EAN)
Brand
Category

Variants:

group_id — links related SKUs (e.g., color variants of the same product). The group_id should represent the canonical parent listing on the merchant’s site. If you submit variant rows by color or size, every variant must share the same group_id. Submitting individual variant SKUs without a group ID is a common rejection reason.

Pricing & availability:

Price (current selling price)
Sale price (if different from regular price)
Availability (in stock, out of stock, preorder, backorder)
geo_price and geo_availability for region-specific overrides

Media:

Primary image (required)
Additional images
Video URLs
3D model URLs (for merchants with augmented reality assets)

Merchant details:

Brand name
Seller URL
Store policies (returns, shipping, etc.)

Shipping:

Shipping methods, costs, and estimated delivery times
For 3rd party sellers and marketplaces: include marketplace_seller field

Reviews:

Aggregate rating
Review count
Individual review content (optional but recommended)

Control flags:

enable_search — controls whether the product is discoverable inside ChatGPT
enable_checkout — controls whether Instant Checkout is enabled for this product (requires Agentic Commerce Protocol implementation)

What makes the OpenAI feed different from Google’s

A few differences worth flagging:

Description format flexibility. OpenAI accepts plain text, HTML, and markdown in the description field simultaneously. This lets the AI parse structured formatting, bullet points, and rich content for better conversational matching. Google Merchant Center is more restrictive about description formatting.

The 15-minute refresh cycle. Google’s standard is daily fetches. OpenAI accepts updates every 15 minutes. Merchants who don’t update frequently — particularly on price changes and stock updates — get penalized for staleness. This isn’t optional; it’s the competitive baseline.

Variant grouping. OpenAI is stricter about how variants are submitted. Every color/size variant of the same product must share a group_id representing the parent listing. Google tolerates more flexibility here.

Discovery and checkout flags. The enable_search and enable_checkout flags give merchants explicit control over which products are surfaced and which support in-chat purchase. No equivalent in Google’s spec.

The 15-minute refresh cycle, in practice

This is the operational reality most merchants underestimate. OpenAI treats your feed as the source of truth. If your prices changed at 9 AM and your feed update doesn’t push until 11 PM, ChatGPT shows the old price all day. If a product sold out and you don’t push a stock update for hours, ChatGPT recommends an out-of-stock product.

For merchants used to Google Merchant Center’s daily fetch cadence, this is a workflow shift. The feeds need to be event-driven (push on price change, push on stock change) rather than scheduled (push once a day at 3 AM). Most feed management tools support both modes, but the configuration matters.

The penalty for stale data is severe. ChatGPT flags products with discrepancies between feed data and live site data. Persistent discrepancies can drop products from results entirely until the data is reconciled.

Practically: build your feed pipeline to push updates whenever inventory or pricing changes. If you can’t do event-driven updates, run scheduled updates every 15-30 minutes minimum. Daily updates are not competitive on ChatGPT.

The Agentic Commerce Protocol

The Agentic Commerce Protocol (ACP) is an open standard for managing communication between buyers, AI agents, and merchants throughout a purchase. It covers checkout flow, payment handling, fraud signals, and fulfillment notifications. The first live use case is Instant Checkout in ChatGPT.

ACP has three associated specs:

Agentic Checkout Spec — defines the checkout flow
Delegated Payment Spec — handles secure payment sharing with payment service providers like Stripe
Product Feed Spec — covers the feed format we’ve been discussing

For merchants who want to enable Instant Checkout (so customers can buy inside ChatGPT without leaving), ACP implementation is required. If you use Stripe as your payment processor, ACP support can be very light-lift — Stripe has built tooling that lets merchants enable Agentic Commerce with minimal code changes. Other processors can connect via Stripe’s shared token API or ACP delegated payments.

If you’re not implementing Instant Checkout — that is, customers discover your products in ChatGPT and click through to your site to complete purchase — you don’t need to implement ACP. You just need a clean feed.

OpenAI has been explicit that enabling Instant Checkout doesn’t artificially boost your products in initial recommendation results. But when a user has decided on a specific product and multiple sellers carry it, having checkout enabled gives a ranking advantage in the merchant selection step.

What Shopify and Etsy merchants need to know

Through Shopify Catalog and Etsy’s direct partnership with OpenAI, your products are already eligible to appear in ChatGPT Shopping without any additional submission work. Major brands like Glossier, SKIMS, Spanx, and Vuori are integrated through Shopify.

This is genuinely useful — but it’s also where most merchants form a misleading impression about what’s required.

The Shopify/Etsy integration is a passive sync. Whatever data exists in your store flows through to ChatGPT. If your titles are SKU-style, if your descriptions are sparse, if your products are missing GTINs, ChatGPT sees that data exactly as it sits in your store. Auto-integration isn’t a free win. It’s a free exposure of whatever data quality you already have.

This explains a pattern we see often: a Shopify merchant assumes they’re “covered” with AI shopping, doesn’t optimize their product data, and then notices that smaller competitors with better-optimized feeds are ranking above them in ChatGPT results. The competitor isn’t gaming any algorithm. They’re feeding cleaner data.

The honest assessment for Shopify and Etsy merchants:

Your products are eligible to appear via auto-sync
Whether they actually rank depends entirely on the quality of your store’s product data
The Merchant Program is closed to you if you’re on Shopify or Etsy — your only path is improving the data Shopify/Etsy is syncing
Optimizing your underlying product data lifts your performance across all AI shopping platforms, not just ChatGPT

The work to do, in order:

Audit your Shopify or Etsy product titles, descriptions, GTINs, variant attributes, and images
Add Schema.org markup to product pages (most Shopify themes don’t include the full schema set by default)
Configure robots.txt to allow GPTBot
If you have flexibility around product copy, rewrite descriptions for conversational matching (more on this below)

There’s no way to bypass the data quality work. Auto-integration without optimization produces auto-mediocrity.

What non-Shopify, non-Etsy merchants need to know

If you’re on WooCommerce, BigCommerce, Magento, custom hosting, or any other platform, you have a more direct path: apply to the Merchant Program through chatgpt.com/merchants and submit your feed via SFTP.

The advantage of direct submission is control. You’re submitting your own feed with your own attributes, your own variant grouping, your own discovery and checkout flags. You’re not at the mercy of how Shopify Catalog decides to map your data.

The disadvantage is the work. You need to:

Generate a feed in OpenAI’s Product Feed Specification format
Set up SFTP push to OpenAI’s endpoint
Configure refresh cycles (at minimum every hour, ideally every 15 minutes)
Implement Schema.org markup on product pages
Configure robots.txt for GPTBot
Optionally implement ACP if you want Instant Checkout

Most ecommerce platforms have feed export tools or extensions that can produce Google Shopping format feeds — these can be adapted to OpenAI’s format with relatively minor field mapping. Feed management tools like Feedonomics, Channable, and DataFeedWatch all support multi-platform feed generation.

Configuring robots.txt for GPTBot

GPTBot is OpenAI’s crawler. It reads product pages and validates feed data. Many default robots.txt files block it.

To explicitly allow GPTBot:

User-agent: GPTBot
Allow: /

While you’re configuring this, allow the other major AI crawlers too — your AI shopping work isn’t just about ChatGPT, and the same robots.txt configuration unblocks Perplexity, Claude, and Google AI Mode discovery:

User-agent: GPTBot
Allow: /

User-agent: ClaudeBot
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: Google-Extended
Allow: /

Changes propagate through crawler caches in a few hours. Make this change once, then move on.

Writing for conversational matching

ChatGPT ranks products based on how well they match natural language queries. This is different from traditional Google Shopping, which matches based on keywords and product taxonomy.

A user asking “what’s a good gift for my dad who likes woodworking but isn’t a serious hobbyist?” needs ChatGPT to find products that match three concepts: gift-appropriate, woodworking-themed, beginner-to-casual level. A product description that just lists features (“12-piece chisel set, hardened steel, walnut handles”) doesn’t surface this product as well as one that says (“A starter chisel set perfect for casual woodworkers and gift-giving — easy to maintain, comfortable handles for occasional use”).

The optimization tactics:

Write descriptions in natural language, not feature lists
Anticipate the questions customers would actually ask, then answer them
Include use case language — who this product is for, what occasions it suits
Add comparison context — what makes this product different from alternatives
Length matters — 200+ words of substantive content typically outperforms 50-word descriptions

This is more work than a typical Google Shopping description. It also pays off in conversion rates on your live site, so the work isn’t single-purpose.

Tracking ChatGPT-driven traffic

Track AI-driven conversions using:

UTM parameters on the URLs in your feed (preserved on click-throughs from ChatGPT)
Stripe transaction IDs (if using ACP and Instant Checkout)
Perplexity-style merchant reporting (OpenAI’s merchant dashboard, when available)

OpenAI is building a self-serve merchant portal for direct feed integration that should improve performance reporting. Until that’s broadly available, UTM tagging is the most reliable way to attribute ChatGPT-driven traffic.

Common rejection and ranking issues

If your feed is rejected or your products aren’t surfacing, the diagnostic checklist:

Variant rows submitted without group_id — every color/size variant must reference the parent listing
Missing GTINs — required for product matching
Stale pricing or inventory (older than 24 hours) — penalized fast
Description shorter than ~30 words — too thin for conversational matching
Generic titles (“Style #18383”) — no signal for AI to match against queries
Images missing or broken — at least one valid image_link is required
enable_search=false — accidentally disabling discovery
robots.txt blocking GPTBot — explicitly or via generic Disallow rules
Out-of-stock products marked as in-stock — fastest way to get penalized

Run through this list before assuming there’s a deeper algorithmic issue.

What this looks like with UCP Radar in the loop

UCP Radar is built specifically for the data quality layer that AI shopping platforms reward. We score your existing feed across 100 quality points (GTIN coverage, title quality, description depth, schema completeness, variant grouping, pricing accuracy, image quality) and use AI to enrich the gaps. The output is:

A supplemental feed that maps to OpenAI’s Product Feed Specification fields
A Schema.org loader script for your product pages
A UCP JSON Feed designed for AI agent consumption
An llms.txt file for AI crawler discovery

For non-Shopify merchants applying to the Merchant Program, this gives you a feed that’s already optimized for ChatGPT’s ranking signals before you submit. For Shopify and Etsy merchants relying on auto-sync, UCP Radar improves the data Shopify or Etsy is syncing — same pipe, but now what’s flowing through it ranks.

A note on the broader AI shopping picture

The optimization work for ChatGPT — clean feeds, GTIN coverage, structured Schema.org markup, conversational descriptions, robots.txt configuration — applies almost identically to Perplexity Shopping and Google AI Mode. The platforms differ in their submission processes and checkout integrations, but they converge on what they reward at the data level.

If you’re prioritizing where to focus first, ChatGPT has the largest user base, the most published technical specs (which makes optimization measurable), and the clearest merchant program structure. Perplexity has the most engaged audience by demographic. Google AI Mode draws from your existing Google Merchant Center work.

For most merchants, the right approach is to optimize the data layer once, then submit to all three platforms. The marginal effort to expand from one platform to all three is small once the data layer is right.

We’ve covered ChatGPT in depth here. Whether you’re optimizing for ChatGPT specifically or AI shopping broadly, the work converges on the same place: clean feeds, complete schema, real-time accuracy, and the structural elements that make product data legible to AI agents.

Start there. Everything else follows.