ChatGPT product feed: what merchants can actually submit
OpenAI's product feed spec lists nine required fields. Here is what they are, what the spec leaves unsaid, and what to fix on the product page itself.
By Crawlito · Updated 9 min read
In short
- OpenAI's product feed spec lists nine required fields: item_id, title, description, url, brand, seller_name, image_url, availability and price.
- OpenAI's feed submission guide says to start in Business Hub, claim your business and pass identity verification before you upload products.
- OpenAI says feed eligibility does not guarantee display, and it does not describe how products are ranked.
- The spec asks for current price and availability and for public product and image URLs, so the feed, the page and any Product markup should say the same thing.
- OpenAI's crawler page says OAI-SearchBot is the agent for ChatGPT search; it does not say whether feed ingestion uses it.
On this page
- How can products reach ChatGPT?
- What are the nine required feed fields?
- What does the spec not say?
- Keep the feed, the product page and the schema consistent
- Make sure OAI-SearchBot can fetch product pages
- Get price, availability and variants right
- How do I check a feed row against the page a crawler sees?
- Frequently asked questions
- Check what a crawler sees on your product pages
OpenAI publishes a file format for product feeds, and its feed spec lists nine required fields for each row: item_id, title, description, url, brand, seller_name, image_url, availability and price. At the time of writing, OpenAI's submission guide says onboarding starts in Business Hub, with a business claim and an identity check. What the spec does not say is just as useful to know: it does not state how products are ranked. This guide sticks to what OpenAI documents and then covers the part you control without anyone's permission: making the feed, the product page and the structured data agree, and keeping the page readable by crawlers.
Everything about the feed below comes from OpenAI's product feed reference and its linked file upload guide, read on 11 November 2026, and its Submit product feeds guide, read on 15 November 2026. All three can change, so check them before you build anything.
How can products reach ChatGPT?
OpenAI's documentation points at two routes, and a third that I could not confirm.
| Route | What OpenAI documents | What it leaves unsaid |
|---|---|---|
| Product feed | A field reference, and a guide to onboarding through Business Hub with identity verification | Which businesses are eligible, how rows are ranked |
| Crawling for search | OAI-SearchBot surfaces sites in ChatGPT search; sites that opt out are not shown in answers | Whether it plays any part in shopping results; the crawler page does not mention shopping |
| Platform integrations | Nothing I found on OpenAI's pages | Which platforms generate or submit feeds for you |
Many third-party guides describe an application step, an invite process or an automatic Shopify integration. They disagree with each other and some are clearly out of date. OpenAI's submission guide describes only the Business Hub route, plus a personal claim link for businesses with an Ads account or an OpenAI account manager. If you are on a hosted platform, check that platform's own documentation and ask OpenAI or your platform vendor about access before you plan around it.
What are the nine required feed fields?
The spec's "Basic product data" section lists these fields as required for every row. The descriptions below are condensed from it.
| Field | What the spec asks for |
|---|---|
item_id | A stable, unique ID per item or variant, never reused for a different item |
title | The product name, including the selected variant when relevant; aim for 150 characters or fewer |
description | Factual plain text; aim for 5,000 characters or fewer |
url | The product detail page, with the variant selected when possible; keep it stable |
brand | The brand as shown on the product page |
seller_name | The name of the seller supplying the offer |
image_url | The main image for this variant, as a direct image URL such as JPEG or PNG |
availability | One of in_stock, out_of_stock, pre_order, backorder or unknown |
price | The regular price in major units with an ISO 4217 currency code, such as 79.99 USD |
A few details are easy to miss:
- A row with an omitted, empty or unrecognized
availabilityvalue is rejected. If you do not know the stock status, sendunknownexplicitly; the spec says it does not claim the item is in stock. - Price uses a decimal amount, a space and an uppercase three-letter currency code, with no thousands separators.
- Each purchasable variant gets its own row with its own
item_id, a sharedgroup_idandlisting_has_variationsset to true. Option values go in avariant_dict, for example{"color":"Black","size":"10"}. - Product and image URLs must be publicly accessible, and text must be UTF-8.
gtinandmpnare optional in this spec, as aresale_price, shipping fields, return fields and review fields. Ads and checkout have their own additional fields.
The upload guide adds delivery details. At the time of writing it describes a full-snapshot feed pushed over SFTP, published at least daily, with Parquet preferred and JSONL, CSV and TSV (compressed) also supported. It says a product missing from a snapshot keeps its last processed record for up to 14 days, and that is_eligible_search set to false is how you make a product ineligible for search. The spec says is_eligible_search defaults to true when omitted.
What does the spec not say?
Be careful here, because this is where most third-party advice goes beyond the evidence.
- Eligibility. The feed reference describes the format, not who qualifies. The submission guide gives the onboarding steps: claim your business in Business Hub, verify your identity in the product section, choose the countries and currencies you sell in, then deliver full snapshots by manual upload, a hosted HTTPS URL or SFTP. Claiming a ChatGPT business profile covers the claim and the identity check. The feed reference also says setting the checkout flag does not complete checkout onboarding, and that checkout needs a separately enabled integration it does not describe. The upload guide mentions an additional Ads eligibility requirement for feeds uploaded through Ads Manager and points to a separate Ads guide, which I did not review.
- Ranking. Nothing on the page explains how rows are ordered or chosen. The spec says search eligibility does not guarantee display.
- Checking against your site. The spec does not describe whether or how OpenAI compares a row with the live product page. It only requires public URLs and current values.
- Crawler involvement. It does not say that OAI-SearchBot or any other crawler fetches your product pages in order to process a feed.
Any post that states those as facts, including claims about response times or approval criteria, is relying on something other than the spec. Treat it the same way.
Keep the feed, the product page and the schema consistent
The spec asks for the current price, current availability and stable URLs. It also notes that sale-window and expiry dates do not schedule price or availability changes, and says to update the feed when a sale starts or ends. That makes the product page and the feed two copies of the same facts, and a third copy usually exists: your Product markup.
OpenAI does not publish a mapping to schema.org. These are the closest equivalents, so you can see where three sources must agree:
| Feed field | On the page | Product JSON-LD |
|---|---|---|
title | Product heading and selected variant | name |
url | The variant's URL | offers.url |
brand | Brand shown to the shopper | brand |
image_url | Main image for this variant | image |
availability | Stock status text and buy button state | offers.availability |
price | Visible price | offers.price and offers.priceCurrency |
Google's Product structured data documentation separates merchant listings, for pages where customers can buy, from product snippets, for pages that only review or compare. Its merchant listing page requires a name, an image and an offer with a price greater than zero and a currency code. Our Product schema guide goes through those properties, and product variants and AI search covers keeping a selected variant's price, image and availability aligned, which the feed spec also asks for.
On Shopify, Product markup normally comes from the theme, so edits happen in theme code; Shopify's themes documentation is the starting point, and Shopify and AI crawlers covers the JSON-LD and app issues specific to stores.
Make sure OAI-SearchBot can fetch product pages
A feed does not make your pages readable. OpenAI's crawler page says OAI-SearchBot is for search and that sites which opt out are not shown in ChatGPT search answers, though they can still appear as navigational links. Its settings are independent of GPTBot, which is for training. If your robots.txt, CDN or firewall blocks OAI-SearchBot from product pages, that is worth fixing whether or not you submit a feed. GPTBot, OAI-SearchBot and ChatGPT-User explained has the details, including OpenAI's published IP ranges.
Check three things on product URLs specifically:
- robots.txt. No group that names OAI-SearchBot, or the
*group where none does, disallows/products/or whatever path your store uses. - Challenges. A bot rule that serves an interstitial to automated requests means the crawler reads the challenge, not the product. See WAF challenge pages and AI crawlers.
- Raw HTML. Google says dynamically generated markup can make Shopping crawls less frequent and less reliable and recommends Product structured data in the initial HTML. The same reasoning applies to price and availability text that only appear after a script runs.
The spec's own requirement is that product and image URLs be publicly accessible. A page that returns a login, an age gate or a challenge to automated requests is a poor fit for that, though OpenAI does not say how it treats such pages.
Get price, availability and variants right
Stale values are the most common way a feed and a page drift apart, and the spec gives you rules to check against:
- Price. The submitted price should be what the page shows. A sale price must be greater than zero, strictly lower than
priceand in the same currency, or it is not used. The spec says not to round a price into a different amount a customer would pay. - Availability. Send the real stock state with each row, and update it when an item sells out or returns. For pages that stay up while an item is unavailable, see out-of-stock products and AI search.
- Variants. Top-level attributes such as color and size must match the same options in
variant_dict; the spec says it does not reconcile conflicts for you. Do not use one row for a product whose price differs by size. - Offer IDs. If you use
offer_id, the spec says never to include the price in it.
How do I check a feed row against the page a crawler sees?
Take a few rows, including a sale item, an out-of-stock item and a variant, and compare each with the raw HTML of its URL. First check the status and that nothing is challenging requests:
curl -s -o /dev/null -w "%{http_code}\n" -A "OAI-SearchBot/1.4" "https://example.com/products/blue-mug"Then compare the price in the feed with what the server sends. This script reads a JSONL feed, fetches each row's url without running JavaScript, and reports whether the feed price appears in the page's JSON-LD and in the raw HTML:
import json, re, sys, urllib.request
def offers(html):
pattern = r'<script[^>]*application/ld\+json[^>]*>(.*?)</script>'
def walk(node):
if isinstance(node, dict):
if "price" in node:
yield node
for value in node.values():
yield from walk(value)
elif isinstance(node, list):
for value in node:
yield from walk(value)
for block in re.findall(pattern, html, re.S | re.I):
try:
yield from walk(json.loads(block))
except ValueError:
pass
for line in open(sys.argv[1]):
row = json.loads(line)
amount, currency = row["price"].split()
req = urllib.request.Request(row["url"], headers={"User-Agent": "OAI-SearchBot/1.4"})
html = urllib.request.urlopen(req, timeout=20).read().decode("utf-8", "replace")
ld = [(o.get("price"), o.get("priceCurrency")) for o in offers(html)]
match = any(float(p) == float(amount) and c == currency for p, c in ld)
print(row["item_id"], "feed:", row["price"], "json-ld:", ld, "match:", match, "in-html:", amount in html)A mismatch is a finding even when the page looks right in a browser: the cause is often a script that rewrites the price, a cached template, or a feed built from a different price list. The script sends the crawler's user agent from your own IP, not OpenAI's, so it cannot prove what OpenAI's crawler receives; your CDN's security log is the better evidence for that. Run it on a sample, then fix the source of the difference, not the single row.
Frequently asked questions
Do I need a product feed for products to appear in ChatGPT?
OpenAI's pages do not say. The feed reference describes how to submit products and the crawler page describes OAI-SearchBot for search, but neither states whether one is required or sufficient for shopping results.
Who is allowed to submit a feed?
At the time of writing, OpenAI's guide says to start in Business Hub, claim your business and pass identity verification before uploading products. It does not list eligibility criteria, and Ads Manager uploads have an extra requirement.
Does a valid feed guarantee my products are shown?
No. The spec says eligibility does not guarantee display, and it does not describe ranking. It also says its format guidance is not a guarantee that every invalid value is rejected at upload, so a clean upload is not proof that your rows are correct.
How often should I update the feed?
The upload guide says to publish full snapshots on a predictable cadence, at least daily. The feed reference adds that you should update it when a sale starts or ends and when availability changes.
Does blocking GPTBot keep my products out of ChatGPT search?
OpenAI says its GPTBot and OAI-SearchBot controls are independent. GPTBot relates to training, and OAI-SearchBot to ChatGPT search, so blocking GPTBot alone does not opt you out of search. If you submit a feed, OpenAI's Businesses in ChatGPT Terms also say you will let its crawlers and agents access your site.
Check what a crawler sees on your product pages
Run the free check on one product URL to see the robots.txt rules for named AI crawlers, the robots meta tag and headers, and whether the price and name are in the raw HTML. For a whole catalogue, an audit covers up to 500 pages with evidence and prioritized fixes. Neither tells you whether OpenAI will show your products; they show whether the technical barriers on your pages are gone.
Filed under Strategy and audits. Topics: ai-search, structured-data, ecommerce, oai-searchbot