ChatGPT Shopping Product Feed: 9 Fields, Setup and Fixes (2027)

Why most ChatGPT feed advice is already out of date

Most of what's been written about ChatGPT product feeds dates from late 2025, and a lot of it has quietly gone stale. Fifteen-minute refresh cycles, in-chat checkout for every store, "no ads, ever": none of that describes ChatGPT shopping today.

Here's what does. In March 2026 OpenAI refocused ChatGPT shopping on discovery. Shoppers browse, compare and narrow things down inside the conversation, then usually buy on the merchant's own site. Your feed is the data behind all of it.

If that data is thin or wrong, ChatGPT either skips your products or describes them badly. The second problem is worse, because nobody tells you it's happening.

This guide sticks to what OpenAI's documentation says right now. It covers the fields that are actually required, the flags that control where products appear, how to submit, and what to check when a product shows the wrong price or doesn't show at all. We update it when the spec changes.

Three ways to get your products into ChatGPT

There isn't one ChatGPT feed. There are three routes, and the right one depends on your platform and whether you want paid placement.

Organic and paid are separate systems. A product can sit in both, and you don't need ad spend to appear organically.

How to submit your products to ChatGPT, step by step

If you're on the direct-feed route, this is the whole path from application to live products. Shopify merchants can skip to step 8.

1. Apply first. OpenAI says feed onboarding is available to approved partners. Fill in the form at chatgpt.com/merchants. OpenAI doesn't publish a review timeline, so apply now and build the feed while you wait.
2. Choose file upload, API, or both. OpenAI's own recommendation is a full file once a day, plus API updates during the day for price and stock. A small catalog can run entirely on the API. Promotions can only be sent through the API.
3. Map your catalog to the nine required fields. Keep item_id stable forever. OpenAI matches products by that ID, not by file name, so regenerating IDs looks like a brand-new catalog.
4. Write descriptions as facts. Materials, dimensions, compatibility, what's in the box. ChatGPT will summarize your copy anyway, so give it facts to work with rather than adjectives.
5. Pick a format. Parquet is preferred. Gzipped JSONL, CSV or TSV also work. Use UTF-8 and one stable file name (or a stable set of shards) that you overwrite on every run.
6. Validate a sample of about 100 products. Every sample row needs all required fields. OpenAI lists three failures that sink most first uploads: missing required fields, outdated field names, and malformed values.
7. Upload the full snapshot and keep it fresh. Push the complete catalog over SFTP, at least daily. A product left out of every shard stays live for up to 14 days, then expires. To pull one faster, set is_eligible_search to false.
8. Check how you actually appear. Ask ChatGPT the questions your customers ask, in the categories you sell. Note whether you show up, at what price, and with which description. Most teams skip this step, and it's where the expensive mistakes hide.

The product feed fields that actually matter

OpenAI requires nine fields on every row. Everything else is optional, but optional doesn't mean unimportant: the optional fields are what let ChatGPT answer a shopper's follow-up questions. All of this comes from OpenAI's Products spec.

The nine required fields

Don't fill gaps with placeholder text. OpenAI's spec says to omit an unknown optional value rather than write null or n/a, and unknown is only valid for availability.

Eligibility flags

Older names still work: enable_search, enable_checkout, is_eligible_ads, and id or sku for item_id. Send only one name per value, because the spec picks a winner when both appear.

Digital goods: what is_digital does and doesn't do

The spec has one digital-goods field: is_digital, where true means no physical shipment. It sits in a group of fields that OpenAI has to configure for your integration. Adding the column to a standard upload does nothing on its own.

If you sell e-books, courses, templates or software, raise digital support during onboarding. Leaving the field blank doesn't count as "digital". And the nine required fields still apply to every digital item.

Optional fields worth sending

Formats, delivery and refresh cadence

OpenAI wants a full snapshot of your catalog at least once a day, pushed over SFTP. Each snapshot replaces the last one as the source of truth. The rules below come from OpenAI's file upload overview.

You don't need a "done" marker. OpenAI processes the whole folder once uploads stop changing for a short while, so upload all shards in one go.

Can you use your Google Merchant Center feed?

Yes, with conditions. There's no button that connects Google Merchant Center to ChatGPT. What OpenAI does offer is a Google-compatible upload path: you export your Google-format feed and upload it, without renaming columns. OpenAI only accepts it after confirming the format for your registered feed.

The differences that trip people up:

• File type. Tab-delimited .txt/.tsv or comma-delimited .csv, gzip allowed. XML, RSS, JSON and spreadsheets aren't accepted on this path, and g:-prefixed column names won't work.
• Required columns. id, title, description, link, image_link, availability, price and brand on every row.
• Availability spelling. Use preorder, not OpenAI's pre_order. unknown isn't accepted, and preorder or backorder rows need an availability_date.
• Identifiers. Send a valid gtin or mpn unless you set identifier_exists to no.
• Seller name. OpenAI uses the merchant name you registered. An uploaded seller_name column is ignored.
• No per-product opt-out. Every accepted product is search-eligible. Setting is_eligible_search=false does nothing on this path, so upload only the products you want in ChatGPT.

If you need per-product control or checkout, the native OpenAI format is the better long-term choice. Details are in the Google-compatible section of the Products spec.

How to optimize your feed for ChatGPT recommendations

OpenAI doesn't publish how it ranks products, so treat any "ranking formula" you read with suspicion. What OpenAI does say is that eligibility doesn't guarantee display, and that optional attributes can improve the quality of ChatGPT's answers. The practical job is to make each product easy to match against the questions shoppers actually ask.

• Write titles the way people ask. "Waxed canvas tote, 15-inch laptop sleeve, water-resistant" matches far more questions than "The Everyday Tote".
• Put the specs in the description. Dimensions, materials, compatibility, care instructions. OpenAI accepts plain text or bullet-style text.
• Model variants properly. One row per variant, a shared group_id, and a variant_dict. Otherwise ChatGPT sees a pile of near-duplicate products.
• Keep the feed and the page in agreement. Price, stock and title should match your product page. ChatGPT reads the web as well as your feed, and conflicting data costs you trust.
• Add GTINs to branded goods. They help ChatGPT recognize that your listing is the same item a shopper is comparing.
• Send reviews and return terms. Review counts, ratings and a clear return window answer the "is it worth the risk?" question inside the chat.
• Hold back messy data. OpenAI's best practices suggest leaving out an optional field until its data is stable. A wrong attribute does more harm than a missing one.

Why your products aren't showing up in ChatGPT (and how to fix it)

Most "ChatGPT can't find my products" problems come down to one of five causes. Work through them in this order.

You're not in the feed at all. On Shopify, check that the Agentic channel is switched on and your store policies are filled in. Everyone else: your application may still be pending, or your first upload failed validation. Look for missing required fields and old field names first.

ChatGPT can't read your site. ChatGPT's search uses OpenAI's OAI-SearchBot crawler. If robots.txt or your bot protection blocks it, your product pages can drop out of answers even with a perfect feed. Check your CDN's bot rules too, not just robots.txt.

You're in the feed but never recommended. Eligible isn't the same as chosen. ChatGPT picks products that match the shopper's question, and vague titles give it nothing to match. Rewrite the top 50 titles and descriptions around real customer questions, then re-test.

The price is wrong. Usually the feed and the product page disagree, or a sale ended and the snapshot didn't catch it. Upload at least daily and push price changes through the API. If resellers carry your products, check which seller ChatGPT shows first. The wrong price may be theirs.

ChatGPT says your product is discontinued. Check three places. Your feed: an item stuck on out_of_stock or dropped from the snapshot. Your site: an old product URL returning a 404. The wider web: an old forum thread, retailer page or review saying it was discontinued. The feed and site fixes are yours to make. For the third, publish a clear, current statement on the product page and correct the record where the wrong claim is spreading.

What changed in 2026: checkout, ads and Shopify

ChatGPT shopping has changed direction twice in a year. If you're reading an older guide, these are the updates it probably misses.

OpenAI explained the March shift plainly: the first version of Instant Checkout didn't give it the flexibility it wanted. In-chat purchase hasn't vanished completely. The is_eligible_checkout flag still exists, but it only works with a separately enabled checkout integration.

How ChatGPT compares with Google AI Mode and Copilot

Each AI assistant pulls product data from a different place. One clean catalog, exported in each platform's format, covers all three.

The common thread is data quality. Titles, specs, GTINs and accurate stock help you on every one of these surfaces, whatever format they arrive in.

How to track whether your products appear in ChatGPT

There's no organic ChatGPT dashboard that shows your product impressions. So you measure it from three angles.

1. Run a fixed set of prompts on a schedule. Pick 20 to 30 questions your customers really ask, like "best waterproof trail shoes under $100". Run them weekly and log whether you appear, the price shown and the description used. Keep the wording identical so the trend means something.
2. Watch referral traffic. ChatGPT adds utm_source=chatgpt.com to many outbound links. In GA4, build a channel group for chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com.
3. Check feed health. OpenAI's upload history reports rejected rows and missing columns. A product that never made it into the feed can't be recommended.

The first step is the tedious one, which is why most teams stop doing it after a month. Alhena's AI Visibility tracking runs those prompts on a schedule across ChatGPT and other AI assistants, so you can see where your brand shows up and where it's missing.

Frequently asked questions