Why does AI say our Korean brand isn't sold in the US?

A where-to-buy answer looks like one fact. It is several, and they are held by different parties. Whether a product is on a shelf in Ohio is recorded by the retailer that owns the shelf. Who sells it is the name written into each product listing. Where it ships from is recorded on a product page, if anywhere.

The brand holds fewer of those records than it assumes. That is why an assistant can tell a shopper that a product stocked by a national retailer “is not sold in the US”, and why the answer can stay wrong after the brand has checked its own site and found nothing incorrect on it.

Sources checked on 23 September 2026.

Where does an AI answer get “where to buy” from?

From four records, and the brand writes only some of them. The table maps each fact in a where-to-buy answer to the record it usually comes from and to what the brand can change.

One of the four is a record the brand never writes: the retailer’s store data. Two more are written by whoever publishes the listing, which is the brand only on the brand’s own site. So fixing a where-to-buy answer starts by finding which record the answer read, not by editing the brand’s site.

Why does “not sold in the US” appear when the product is on shelves?

Because the statement that the product is on shelves belongs to the retailer, and the brand has usually not made its own. Which stores hold a product is recorded in the retailer’s local inventory data: a store_code that must match the retailer’s own Business Profile, and an in-store availability such as in_stock or limited_availability. Google says it uses that data “to know which stores have each of your products” (Google). The brand appears only in the retailer’s separate product data, as an attribute; that specification tells merchants not to enter their store name as the brand unless they manufacture the product (Google).

That record only helps the brand if it can be joined to the brand’s product. The chain has two links: the inventory rows use the same id as the retailer’s product data (Google), and the product data carries the GTIN, which the product data specification requires for new products that have one (Google). A US listing carrying a different GTIN from the Korean packaging, or none, is a product an assistant cannot confidently match to the brand it was asked about.

Meanwhile the brand’s own English page often says nothing about US retail at all. The most retrievable statement about availability is then the Korean site, a launch article from before the retail deal, or a listicle that predates it. None of them is wrong about the past. All of them are wrong about now.

Why does an old distributor keep appearing?

Because its listings are still published. OpenAI’s product feed specification makes seller_name a required field on every offer, and for Google-compatible feeds the registered merchant’s display name becomes the seller identity (OpenAI). On a marketplace the specification separates the two roles: seller_name is the seller supplying the offer and marketplace_seller is the marketplace where checkout happens, so a former distributor’s marketplace offer still names the distributor. A listing a distributor set up stays live until the channel carrying it removes or re-assigns it. Submitting that feed to OpenAI directly is currently open to approved partners only (OpenAI); a Shopify store that sells to US customers reaches ChatGPT through Shopify Catalog by default, whether or not the store is based in the US (Shopify).

Ending a distribution contract does not retract a single page. Each marketplace and retailer listing has to be closed or transferred separately, and the brand’s own site needs a dated statement of who sells the product now. Without that statement, the former distributor’s listing is the most specific seller record an assistant can find.

How does “ships from Korea in three weeks” happen?

Usually from the only shipping page the brand has published in English: the Korean store’s international shipping terms. When a US warehouse exists but no US page says so, the international terms are the answer.

Schema.org has the fields for the correct version. OfferShippingDetails carries shippingOrigin, described as where the shipment comes from, and deliveryTime, the full delay from order to delivery (Schema.org). Google lists shippingDetails as a recommended property of Offer in merchant listing markup, and those listings can show “price, availability, and shipping and return information” in Search (Google).

Markup is not enough on its own. OpenAI’s product feed has no ship-from field and no delivery estimate, and states that a shipping amount “does not quote a checkout total or guarantee a delivery date” (OpenAI). Where the feed has no slot for the fact, the page text is the only carrier. Write the origin and the delivery window in a sentence a shopper can read.

What should the brand’s own US page state?

Every fact in the table that the brand can publish, in plain text, dated, on the English product page:

• Where it is sold in the US, naming each channel with a link: the brand’s own store, each retailer, each marketplace, TikTok Shop if it is one.
• Who is authorised to sell it, as a dated list. When a distributor relationship ends, update the list the same day.
• Where US orders ship from and the usual delivery window.
• The GTIN that matches the retailer’s listing, so the retailer’s record can be joined to this page.
• The same facts in markup. availability on the offer, and shippingDetails where they differ from the site-wide policy. Structured data describes what is on the page; it does not replace it.

A launch date belongs on the page too. A shopper asking in the week of a retail launch meets an answer built before the launch, and a dated line is the one thing that tells an assistant the older sources are older.

What can the brand not fix from its own site?

The retailer’s store inventory, the marketplace’s listing and the engine’s cache. Each is corrected where it lives: the retailer updates its local data, the marketplace closes or transfers the listing, and the engine picks up the change when it next retrieves the page.

Which of those an answer is actually reading has to be found before any correction is written. The steps for that are in who fixes wrong US AI claims, and the reporting checklist for a provider is in correcting an AI brand misattribution. The underlying pattern — a retailer page holding the product record instead of the brand — is the subject of Amazon listings vs your own site.