Is ChatGPT merchant the right path for your store?

How ChatGPT works for a WooCommerce merchant

ChatGPT merchant refers to a channel where the store remains yours and products may appear when someone asks where to buy. It is not a third-party storefront where you upload photos and wait for orders, and it does not replace WooCommerce. The price, availability and product page remain those of your site. A product feed transmits those facts in the format the channel reads. The page text and file data work together. When they do not match, it becomes much harder to recommend that product with reliable pricing and availability.

In practice, the bottleneck is not the sales copy. It is the missing fact. In a small or medium WooCommerce catalog, the brand is often not stored in a field, the price in the file is a number without a currency, the identifier changes with every export, and the same photo covers several products. The person reading the feed sees rows. The person reading the page sees text. If the row says one thing and the page says another, there is no single fact to confirm. The useful work is closing those gaps before exporting, not rewriting the entire catalog in the same tone.

What being a merchant on ChatGPT means

Being a merchant on ChatGPT means you may be identified as a place to buy from, while the order, price and product page remain on your WooCommerce store.

The merchant program is not a second store that you have to stock manually. It does not publish a list of criteria an assistant uses to decide whom to cite: no generative engine does. What you can prepare is the part that depends on your catalog. An identifiable seller, an accessible product page, a price with currency, availability aligned with stock, an image of that product, a written brand and an identifier that does not change. Without these facts, it becomes much harder to suggest a precise purchase. With them, the citation is still organic, and no tool can guarantee it.

It is not a replacement for Google, nor an automatic publication channel for Amazon, eBay, Etsy or Vinted. Those channels remain separate tasks: you prepare the text according to each one’s rules and publish it yourself. The ChatGPT file follows OpenAI’s specification, not the title length that another channel used yesterday.

For people with little time for catalog routines

You do not need to rewrite every product page tonight. Start with the facts that empty a row: currency next to the price, real stock, an original image, the brand in a field, a stable SKU and the product-page URL. An all-caps title or supplier-copied description can be fixed later, and only if the text is genuinely weak. An illustrative example: a ceramics workshop with about forty products has no dedicated team. In one evening, it can check the products missing a brand or photo instead of rewriting the entire range.

For people who publish manually on other channels

Anyone who pastes product pages into several storefronts tends to reuse the same block of text. Reuse alone is not enough for this channel. The row must contain the file’s nine fields, and the store page must say the same things. A title written for another audience can remain there. The feed contains facts, not slogans. If two channels request different details, keep two texts instead of shortening the product name to an arbitrary figure that is not in the product page.

How a store product ends up in an answer

The product comes from what the page says and what the feed confirms, not from a secret list of stores that a merchant can read.

Assistants read the page text. The file and structured data do not replace that text and are not read “first” or “on their own.” According to Google Search Central, there are two main classes of structured data for products. WooCommerce already produces basic Product markup. Text and markup work together: markup helps confirm price and availability when it matches the page. A page without an additional data block is not unreadable. What holds a product back is missing or vague information, in the text and in the data.

The typical mismatch is concrete. The page says that the sand variant is out of stock while the file marks the parent product as available. Or the page shows 28 euros while the row contains 28 without a currency, or a different price copied from an old price list. In both cases, there is no single fact. Completing the file while leaving the page vague does not solve the problem. Updating the page while forgetting the file does not either.

The page text and feed data work together: the file confirms price and availability when they match what the customer reads on the product page.

Questions at the bottom of a product page help only when they come from real data, such as the capacity in milliliters that you entered. A rating or aggregateRating (only if it reflects real reviews) should be added only when it reflects genuine store reviews, never a number you do not have. Question markup should not be treated as a shortcut to a rich result in Google: those results have not been shown since May 7, 2026. The text remains useful when it states verifiable facts.

The nine feed fields in OpenAI’s specification

According to OpenAI’s specification, the required feed fields are nine: item_id, title, description, url, brand, seller_name, image_url, availability and price with currency.

A ChatGPT product feed is the file that transmits these facts in the format published by OpenAI. It is not the site’s product page and it is not the Google Shopping feed, even though some names are similar. The nine fields are required in the file, not a citation criterion published by ChatGPT. Without them, however, it becomes much harder to suggest a product with reliable pricing and availability. The field-by-field operational details are also available on the page about the WooCommerce product feed for ChatGPT.

• item_id: a stable identifier, usually the SKU. If it changes with every file, the same product appears to be new.
• title: a readable name, not all in uppercase, including the fact that distinguishes the product, such as material, size or color, when it is in your data.
• description: the facts from the product page. A generic sentence does not replace capacity, material or included items.
• url: the product-page address, not the homepage or a category page.
• brand: the brand written in a field and confirmed by you. For a workshop, it may be the workshop’s name, but it must be written rather than guessed from the photo’s style.
• seller_name: the store name as customers know it, identical across all rows.
• image_url: the address of an image of that product. It is not called image_link: that is a Google Merchant Center name.
• availability: whether the product can be purchased, aligned with the store’s stock. The value must say the same thing as the page.
• price: the merchant’s price with the currency. A bare number is not enough. No tool should invent or estimate that price for you.

GTIN and MPN are optional, and a GTIN must never be invented. Only in the Google-compatible profile, when OpenAI confirms it, is a GTIN or MPN required unless identifier_exists = no. The names link, image_link and identifier_exists belong to Google Merchant Center, not the ChatGPT feed. A product that legitimately has no manufacturer barcode should be declared as such. A code copied from a similar product or taken from the web to fill the gap should be removed.

The description can contain facts that the nine fields have no dedicated place to name: material, capacity, what is included in the package, and condition, new, refurbished or used, if you have stated it. These are not required fields in this file. They matter because the page text, which the assistant reads, should not be empty. The Google category, useful for Google Shopping, should not be presented as a tenth OpenAI field. Confirm it separately against the published taxonomy and apply it only if you agree.

What to check in WooCommerce before exporting

Before exporting, every product you want in the file should have a price with currency, stock, brand, original image, stable SKU and product-page address, all matching what the page displays.

In an up-to-date store, these are the actual menus. The product opens from Products. Currency and store location are in WooCommerce, Settings, General: currency belongs to the store, not to an individual product page, and an export that copies only the number loses it. Brand, material, size and color, when you do not have a dedicated brand field, are often in Products, Attributes. The featured image and gallery are on the product page. Stock status is in the inventory panel, and for variations you open the Variations tab rather than checking only the parent product.

1. Open the product and read the title, price, SKU, stock and featured image.
2. Note the currency in WooCommerce, Settings, General, and write it next to the price in the row.
3. Check that the brand is an attribute or field, not an inference from the color or style of the photo.
4. Open the product-page address and compare its price and availability with the file row.
5. If you sell variations, check the price, stock and image of the variation the customer is actually buying.
6. Leave GTIN blank if you do not have it. Do not copy it from a similar product or take it from the web.
7. Keep only rows in the file where these facts match the page.

According to the WooCommerce developer blog, starting with WooCommerce version 11.1, variation galleries will be enabled for every store. This helps prevent a variation in another color from remaining attached to the parent product’s image. According to the WooCommerce developer blog, WooCommerce 11.2 introduces GTIN matching in product CSV imports, which is useful for cleaning and importing product data. Matching helps avoid creating a duplicate when the code already exists. It does not authorize you to manufacture a code for products that do not have one.

For people starting from a supplier price list

A supplier PDF contains names, sometimes list prices and sometimes photos. It rarely contains the SKU you use to sell, and almost never the price you sell at. An illustrative example: the same ceramics workshop receives a PDF of glazes and clay mixes. The rows are reviewed and shown before confirmation. The price that enters the feed is the price you decided for your store, not an estimate and not a historical price from another channel. Extracted photos are matched to the right product and checked because a price-list image may be too small or shared across several codes.

An illustrative before-and-after product-page example, without sales results. Before: title “CERAMIC MUG,” price 28 without a currency in the file, empty SKU, no brand, the same photo as the plate, description “Beautiful handmade mug,” and the parent product available while the sand variation is out of stock. After, using only your own facts: title “Glazed ceramic mug, 300 ml, sand color,” price 28 with the store’s currency, stable SKU, brand equal to the workshop name you entered, a photo of that mug, a description with material, capacity and color, and the sand variation’s availability aligned with stock. No certification added and no barcode invented.

Common product-feed errors

A feed row is lost when a fact is missing, when it contradicts the product page or when it reuses data from another product.

You do not need to quote a message from memory if you have not read it. Observe the symptom: the product does not enter the file you keep, or it enters with a field that is empty or different from the page. The cause is almost always catalog data, not a wording formula.

The price is present but the currency is missing

This is what it looks like: the customer reads 28 euros in the store, while the file contains 28. It happens because WooCommerce stores the currency in the store settings and many exports copy only the number. Fix it by checking WooCommerce, Settings, General, and writing the currency in the price field of every row. If you changed currency and did not export again, old rows remain wrong even when the product page is correct.

A barcode was invented to fill the gap

This is what it looks like: the code fails its check digit, or two products carry the same GTIN. It happens when someone fills the field to “complete” the file by copying a neighboring code or searching for one on the web. Fix it by leaving the field blank, declaring that the product has no manufacturer barcode, and keeping only codes you actually have and that are already valid. In the Google-compatible profile, when OpenAI confirms it, the absence is declared: an identifier is not manufactured.

An identifier you do not have must not be invented: a false GTIN causes more harm than an empty field declared as absent.

The same photo is used for several products

This is what it looks like: several rows point to the same file, or the featured image is small or in a format that channels do not accept. It happens with a reused placeholder or a group shot used instead of a product image. Fix it with one main photo per product. A useful check looks for products without an image, a main photo that is too small, the rejected format and the same shot shared across products. This is not a threshold published by ChatGPT as a citation criterion. It is the point below which the image remains a weak fact even when the image_url field is filled.

The identifier changes with every export

This is what it looks like: the same mug appears as a new product each time you regenerate the file. It happens when item_id is based on the row position, title or a random number. Fix it with the product or variation SKU and keep it even if you rewrite the title. An incorrect address is a related error: the row points to the homepage or a category, and anyone opening the link cannot find the price promised by the file. Fix it with the permanent address of that product page.

Two other signs are less technical but just as common. An all-caps title is difficult to read and often comes from a management system or PDF. Dirty HTML in the description, pasted styles and invisible characters, makes the text longer without adding a fact. An undeclared condition leaves it unclear whether the item is new, refurbished or used. These are data corrections, not tone corrections: they can be seen in the preview, applied only after confirmation, and do not require rewriting the page from scratch.

What to observe after preparing the file

After preparing the file, observe whether shopping answers cite the store and whether price and availability remain aligned with the page, without treating citation as a guaranteed result.

The file that makes sense to submit, when the program you are admitted to accepts it, is the one containing only rows that have already been checked. Submission-screen steps may change and should not be memorized. What remains under your control is the quality of the row. A free product-feed validator helps you see which rows do not pass before considering the work complete. If a row is missing, return to the product instead of inventing the field.

According to Princeton GEO, Generative Engine Optimization can increase visibility by up to 40% in generative-engine answers. That is a result from the study, not a promise about your catalog and not an effect that a tool can guarantee. The narrower observation is this: a page with complete facts that are consistent with the file increases the chances that an assistant has something precise to use. Without those facts, it becomes much harder.

• The page’s price and availability match the row, even after a discount or stockout.
• Products kept out of the file because a field is missing: they are a work list, not a channel failure.
• Store citations in shopping answers, and who is cited instead of you, at the frequency you choose.
• Clicks, impressions and positions from Google Search Console, which measure Google rather than ChatGPT: keep the two sets of figures separate.

An AEO check for WooCommerce compares data readiness with the published specifications, field by field. It tells you whether the catalog is in order. It does not say that ChatGPT will cite the store. A visibility check, weekly, every two weeks or monthly, answers a different question: during this period, was the store cited, and who replaced it? Pause it if you do not want to continue. The two checks should not be confused.

How Katapic prepares the catalog for this channel

Katapic measures the WooCommerce catalog, flags missing fields for the ChatGPT feed and applies corrections only after your confirmation.

A scan of a public WooCommerce store is read-only and changes nothing. Without an account, the scan returns an SEO score and an AI search score. With a free account, you get further scores and a diagnosis across the entire catalog. The checks on data, images, text and duplicates do not use artificial intelligence. The dashboard opens on Start here: the findings with the greatest impact, how many products each affects and one click to view them.

Corrections that do not require artificial intelligence arrive with a preview and count: all-caps titles, dirty HTML, product condition, a declaration that there is no manufacturer barcode, grouped variations and a Google category to confirm. They are applied only after confirmation. Rewriting starts from the title, specifications and company profile, plus sources it can cite. An unsupported claim is removed. If the new version does not raise the score, it is not saved and costs nothing. The barcode is never taken from the web or generated. From the photo, Katapic recognizes the product, brand, category and attributes.

For six channels, Google Shopping, the Meta catalog, the ChatGPT product feed, the store, the PDF catalog and AI answers, you can see which products pass and which fields are missing, with the public rule behind each one. The ChatGPT and Google Shopping feeds can be downloaded with only the products that pass. The free Katapic for WooCommerce plugin, open source under the GPLv2 license on WordPress.org, connects the store with one click, without keys to copy, and reads and writes only products, never customers, orders or payments. Before overwriting anything, it checks whether you edited a title or description manually in the store. For Amazon, eBay, Etsy and Vinted, it prepares text to copy and paste: publication remains manual. The optimizer writes questions and answers only from real data, and the plugin turns them into structured data. This is not a citation guarantee. It is a way to have the data ready. The process starts with a free catalog scan on Katapic.

Frequently asked questions about ChatGPT and the WooCommerce catalog

Connecting WooCommerce to ChatGPT means preparing a feed that is consistent with the product pages, not switching on a store setting.

How do I connect WooCommerce to ChatGPT?

The store is not connected with a switch. WooCommerce remains the source of the facts: title, price with currency, availability, brand, product-page address and image. Those facts go into a feed in the format of OpenAI’s specification, and the page must repeat them in the same way. A plugin can import the catalog and send approved pages back to the store without touching customers, orders or payments. Sending the file to the ChatGPT merchant program depends on the access that program recognizes. Close the empty fields before submitting it.

How do I submit the WooCommerce product feed to ChatGPT?

Prepare a file with the nine fields in OpenAI’s specification and keep the rows where the price, availability, brand, image and product-page address are complete and consistent with the page. The image field is called image_url: image_link belongs to Google Merchant Center, not this feed. Submission depends on the program the store is admitted to, and the steps on that screen may change. If a row does not pass, open the product and check which fact is missing or contradicts the store.

What is the ChatGPT product feed for WooCommerce?

It is the file that transmits catalog facts in the format published by OpenAI. It does not replace the site’s product page and is not the Google Shopping feed, even though some names are similar. The required fields are nine: item_id, title, description, url, brand, seller_name, image_url, availability and price with currency. GTIN and MPN are optional, and a GTIN must not be invented. The page text remains necessary: the file confirms price and availability when they match what the customer reads.

I manage a WooCommerce store. What tool tells me whether ChatGPT recommends my products?

No generative engine publishes the criteria it uses to choose a product, and no tool can guarantee that ChatGPT will recommend the store. You can observe a concrete fact: a check performed weekly, every two weeks or monthly tells you whether AI shopping answers cite the store and who is cited instead of you. In parallel, a readiness check compares the fields with the feed’s published specifications. Citation remains organic and depends on the relevance of the data, not on a switch.

Which fields does the OpenAI product feed require?

OpenAI’s specification lists nine, all to be filled with genuine store facts: item_id, title, description, url, brand, seller_name, image_url, availability and price with currency. item_id is a stable identifier, usually the SKU. url is the product-page address, not the homepage. brand is the brand, while seller_name is the store name. image_url should not be confused with image_link, which is a Google Merchant Center name. GTIN and MPN remain optional. A code you do not have should not be entered.

What should I check in the catalog before selling on ChatGPT with WooCommerce?

Open the product and read the title, price, SKU, stock and featured image. Then open WooCommerce, Settings, General, and note the currency: in the feed, the price carries it with it. Check that the brand is written in a field rather than inferred from the photo’s style. If you sell variations, compare the variation’s price, stock and photo, not only those of the parent product. The address in the file must open that product page. Leave GTIN blank if you do not have it. Export only rows where these facts match the page.