A product feed is the one piece of AI shopping data a merchant sends on purpose. Reviews, editorial roundups and retailer listings are gathered by other people. The feed is yours, and a wrong price or a missing variant in it is yours to fix. This post lists the feed routes the platforms document today, who can use each, and a field checklist that keeps them consistent. Everything about a platform is dated as of October 2026.
Which product feed routes exist as of October 2026?
| Route | Who can use it | What the platform documents |
|---|---|---|
| Google Merchant Center | Merchants who meet Google's policies, no application | Free listings are on by default in most accounts. Eligible products may show in Search, Images, Lens, YouTube, Gemini and the Shopping tab. Showing is not guaranteed. |
| Shopify Catalog (agentic storefronts) | Eligible Shopify stores | Active by default. Feeds channels such as ChatGPT, Google AI Mode and Gemini, Microsoft Copilot and Meta. The Google route runs through the Google & YouTube channel or Merchant Center. |
| OpenAI direct product feed | Approved partners | Application required. Upload over SFTP. Full snapshot at least once a day. A self-serve platform is described as planned. |
| Perplexity | Not verified | A merchant program was announced in November 2024 for large retailers. We found no current feed specification or sign-up path. |
| Crawled product pages | Stores whose pages are crawled | OpenAI's merchant page says a feed is not required but gives more control over how products appear. Google says crawled products can show without markup or a feed. |
The rows differ in one way that matters: only Google's route has no application step and uses a feed you control. For OpenAI's direct feed, a merchant has to apply, and OpenAI's merchant page mentions a waitlist for applicants.
What does a Google Merchant Center feed do?
Google says free listings are turned on by default in most accounts, and that eligible products may show in Search, Images, Lens, YouTube, Gemini and the Shopping tab. It also says the setting is permission to show products, not a promise that they will be shown.
Google's guidance on AI features in Search lists keeping Merchant Center and Business Profile information up to date among its fundamentals. Its May 2025 announcement said AI Mode shopping runs on the Shopping Graph, which then held more than 50 billion product listings, with more than 2 billion refreshed every hour. None of the Google pages we read says how feed data reaches the Shopping Graph, so treat the link between the two as unstated. The related post on Google AI Mode and AI Overviews for product searches covers what Google says about both.
A feed is not the only path. Google says products crawled on your store can show up even when nothing was marked up or submitted. A feed gives you control; it is not a gate.
Google also lists six optional conversational attributes for AI-driven surfaces: questions and answers, document links, related products, an item group title, variant options and a popularity rank. Google says they complement the main specification and do not affect product approval, and that you need not repeat what the description, highlight or detail attributes already say. They suit products that need explaining. Merchant Center also has an AI performance insights report for AI Mode and AI Overviews, in a limited set of countries.
What does OpenAI's direct feed require?
OpenAI's developer documentation says onboarding product feeds is open to approved partners and points to an application form. The delivery model it describes is SFTP, with the same file or set of files overwritten by each new snapshot, and a recommended full upload at least once a day.
The product spec lists nine required fields for a discovery feed: an item ID, title, description, URL, brand, seller name, image URL, availability and price. Optional fields cover GTIN and MPN, shipping, returns and review aggregates, and variant fields are required when a product has variants. Three flags control where a row may be used, and they are separate on purpose:
- Search: on unless you set it to false. OpenAI adds that eligibility does not guarantee display.
- Ads: off unless the feed sets an Ads default, with its own feed guide and its own flag.
- Checkout: off by default and needs a separately enabled integration.
The standard upload targets the United States, and OpenAI's merchant page says shopping is live for ChatGPT users in the U.S. It also says the company is moving away from a standalone Instant Checkout, and that there are no fees on purchases that start in ChatGPT. The merchant page says it is building a self-serve platform; the page is undated, so we could not tell which year its timing refers to. A separate post explains how ChatGPT shopping works for brands.
OpenAI's help center describes the inputs to shopping results as structured metadata from first-party and third-party providers, plus other third-party content. It also says that when merchants change prices or shipping terms there can be a delay before the change shows. A feed that is current does not remove that lag, so do not promise a customer instant propagation.
What happens on Shopify without any feed work?
Shopify's help center says agentic storefronts are active by default for eligible stores, and that products reach AI channels through Shopify Catalog, or through the Google & YouTube channel or Merchant Center for Google AI Mode and Gemini. For ChatGPT, Shopify lists these requirements: the store sells to customers in the United States, products are eligible for Shopify Catalog, and Terms of service, Privacy policy and Return and refund policy are filled in. Shopify also says to put any legal disclosures within the first 6,000 characters of a product description.
Three practical points follow from Shopify's pages:
- Check the switch. Sales channels, then Agentic, shows which AI channels have your catalog. Turning off Catalog access can take up to 7 days to take effect.
- Keep Google syncing. For the Google route, Shopify asks you to leave automatic product and shipping sync on in the Google & YouTube channel, and to keep your policies matching what is in Merchant Center.
- Robots.txt does not control it. Shopify says Catalog syndication happens independently of robots.txt, so blocking a crawler affects only open-web discovery. The post on AI crawlers and robots.txt covers that split.
Which fields should match across feeds?
This is the checklist we would use. It compares the two specs we could read in full, Google's product data specification and OpenAI's product file reference. Shopify's channels fill these fields from your store data, so for a Shopify store the check is whether the store data is right.
| Data point | Google Merchant Center | OpenAI feed | Check |
|---|---|---|---|
| Identifier | id, required, unique, same on every update | item_id, required, stable, never reused | One stable ID per variant, based on SKU |
| Title | title, required, plain text, up to 150 characters | title, required, aim for at most 150 | Same words as the page title, with the variant |
| Description | description, required, up to 5,000 characters | description, required, aim for at most 5,000 | Facts only, plain text, no promotion |
| Page URL | link, required | url, required | Product page with the variant selected where possible |
| Image | image_link, required, at least 500 x 500 pixels (enforced from January 31, 2027) | image_url, required | Direct image URL, publicly reachable |
| Availability | in_stock, out_of_stock, preorder, backorder | in_stock, out_of_stock, pre_order, backorder, unknown | Note the spelling difference. OpenAI rejects a row with a missing or unrecognized value |
| Price | price, currency code, such as 15.00 USD | price, amount then currency, such as 79.99 USD | Equal to the page and the checkout price |
| Brand | brand, required for most new products | brand, required, a real name | Brand as shown on the page |
| GTIN and MPN | gtin strongly recommended, mpn required only without a GTIN | gtin optional with a valid check digit, mpn optional | Validate GTINs, never invent values |
| Variants | item_group_id, required for free listings of variants | group_id, listing_has_variations, variant_dict, one row per variant | One row per size or color, with its own price and stock |
| Shipping and returns | shipping required in the US for free listings | shipping_price, accepts_returns, return_deadline_in_days, return_policy, all optional | Same terms as your policy pages |
| Review aggregates | Not covered here | review_count and star_rating, optional, product reviews only | Only reviews you collected for that product |
Google's page announces the 500 x 500 pixel image size as a new requirement, with enforcement starting on January 31, 2027. Several fields in OpenAI's spec need setup beyond a standard upload, including multi-country support, so ask during onboarding before relying on them.
How do you keep feeds consistent in practice?
- List where each fact lives. For every data point in the table, write down the system of record: your commerce platform, a spreadsheet, a supplier file. Two systems holding the same fact is where mismatches start.
- Fix identifiers first. Variants, offers and every later update attach to them, and both specs stress stable IDs that are never reused. Valid GTINs and clean variant groups come next.
- Write titles and descriptions once. Use the same plain-text facts on the page, in the feed and in your structured data. See structured data for AI shopping and product pages AI assistants can use.
- Compare four places for a handful of products. Take five products from different categories and compare price, availability, shipping and returns on the product page, in structured data, in the Google feed and in the OpenAI feed. Fix every difference at the source.
- Decide when each feed updates. OpenAI recommends a full snapshot at least once a day, and says sale dates in the file do not schedule price changes, so update the file when a sale starts or ends. For Google, use your platform's sync setting.
- Read the diagnostics. In Merchant Center, the Needs attention tab on the Products page is where Google points you to issues with the product details you add. For ChatGPT, look for your products in its shopping results and compare what shows with your product pages.
What do our own numbers say about feeds?
Nothing directly, and the reason is worth stating. Our October 2026 samples record answer text and source URLs from OpenAI's API with web search (gpt-6-luna) and from Perplexity's Agent API with the Sonar model (perplexity/sonar). They do not record product cards, and nothing in them shows whether a product came from a feed.
What they do show is how much of the evidence around an answer is ordinary web pages. In OpenAI's API answers, the pages linked held a retailer in 17.6% of 800 answers, and held the site of a brand the answer recommended in 46.6%. In Perplexity's retrieved source lists, a retailer was present in 76.7% of 787 answers and the site of a recommended brand in 62.0%. Retailer here means one of 49 retail domains we listed by hand, so it undercounts other stores. The own-site figure uses a strict rule (a domain whose first label equals a recommended brand's name, such as samsung.com for Samsung), so it misses country sites, parent companies and other store names. The figures cover all five question types and all three markets, United States, Mexico and Canada, in the October 2026 edition. These are presence counts, not influence. They are a reason to keep your pages and your feed identical, and they are not a measurement of what a feed does. Both engines are reported on their own and the full set is in the October 2026 edition.
What is still unknown?
- Whether a feed changes which brands an answer names. No primary source we read says so, and our data cannot test it.
- Perplexity's current feed route. We could read the 2024 announcement, not a current specification.
- Timing. OpenAI says there can be a delay between a change and what shoppers see, and gives no figure.
- Regions. OpenAI's merchant page describes the U.S. today. Google's pages on its Universal Commerce Protocol (UCP) checkout name the United States, Canada and Australia. Check your market.
Frequently asked questions
How often should a product feed update?
OpenAI recommends a full snapshot at least once a day. We did not find a Google-stated minimum in the pages we read. For Shopify stores, Shopify says to keep automatic product syncing on in the Google & YouTube channel so that data stays current.
Does a feed replace structured data on product pages?
No. Google asks that the price in the feed match the landing page, the structured data and checkout, and says crawled products can show even without markup or a feed. Keep all of them identical so no source contradicts another.
How we write: drafted with AI tools from Dopbot Index data and public sources; every number and outside fact is traced to its source in a claims record and checked before publishing. More about how we write; corrections to hello@dopbot.com.