Why ChatGPT won't cite your product page in another market (and how localized schema fixes it)

You spend weeks perfecting a product page, and ChatGPT starts naming your shop for buyer questions in your home market. Then a customer in Germany or Japan asks the same question, and your brand is nowhere in the answer. The page did not get worse — retrieval systems simply treat a foreign-language answer as a different candidate than your home-market version. Winning citations across borders requires localizing more than the copy: the structured data must match the language, locale, and currency of every market you sell into.

AI retrieval does not match a page on vibes; it matches on the query being asked. When a buyer asks in French about shipping a product to Canada, the pipeline looks for an answer that is both the right language and the right locale. Signals that are scored near-zero across borders:

• Language of the answer. A page full of English where the query is German usually loses to any German page, even a weaker one.
• Currency and prices. A price in the wrong currency, or a price node that never changes, reads as irrelevant to a market that pays in euros or yen.
• Local shipping, returns, and taxes. A return window or delivery node pointing only at your home country fails a buyer asking about their own region.
• Locale-agnostic structured data. The same generic Product schema everywhere, with no region-specific offers, reads as though the page does not actually serve that market.

None of these require you to build a whole new site. They require your structured data to reflect the market the buyer is asking from.

The fix is less about translating your whole catalog and more about giving the local question a local, machine-readable answer:

1. Keep the localized schema with the localized page. If you serve German buyers, publish a German-language product page (or a clean hreflang variant) with its own Product schema written in German — not the English schema machine-translated inline. 2. Put offers in the buyer's currency. Use an Offer node with the correct priceCurrency for each market. A page priced in USD rarely wins a query framed in euros. 3. Say exactly where it ships, in the local language. Shipping details in MerchantReturnPolicy or a delivery node that state "ships to Canada" in French beat a generic English line. 4. Mirror the visible copy to the schema, in the same language. If the page text is German but the FAQPage schema is English, the model sees a contradiction and may drop the page entirely. 5. Keep the localized answer direct and quotable. One crisp sentence in the local language that answers the exact local question is what a model can actually lift into an answer.

This is exactly the gap a routine surveillance loop catches if you run it per market. Pick your top two or three selling regions, re-ask a fixed set of priority questions in each local language, and record which of your URLs get named. When a page is cited at home but missing in a second market, compare your local page's schema to the winning one: it is almost always a missing language match, a wrong currency, or a region-less shipping node.

At Visora we see this constantly in cross-border audits: a merchant's UK page is quotable, the German page for the same SKU falls out because its schema never left English. A free scan at geovisora.com/audit on each localized URL will show which citation layer is present in that market and which is missing, so you fix the right gap instead of guessing.

*Do I need a translated page in every language I sell to?* For the markets that matter, yes. A few well-built localized pages outperform a partially-translated catalog, because retrieval wants a clean, language-matched answer, not coverage for coverage's sake.

*Will currency in the schema really change citations?* It changes relevance. A query framed in euros pulls offers whose priceCurrency matches; a USD-only offer is filtered out of that candidate pool early.

*Should every page have hreflang?* Use hreflang when you have genuinely separate localized pages. They tell both search engines and retrieval systems which language/locale each URL serves and prevent the wrong variant from being cited.

*How long before a localized page gets quoted?* Usually weeks, mirroring the home-market loop. Once the page is live with matching schema, re-ask the same local question weekly and watch your per-URL mention count.

Being cited in one market is a signal of a good page; being cited in every market you sell to is a signal of a well-localized one. The barrier is rarely talent or budget — it is schema that still speaks only one language. Localize the structured data to match each market, keep copy and schema consistent in that market's language, and re-measure per locale. Start with the two markets you care about most, and run a free scan at geovisora.com/audit on each localized URL to close the gap before a buyer in that market asks the very question that currently leaves your brand out of the answer.