The dangerous translation is not the one that sounds clumsy. It is the one that sounds perfectly natural while quietly moving the business into another category, country, or customer promise.
The first line looked harmless. A Spanish answer had taken the name of a small French Catalan food and gift shop in Perpignan and made it smoother for a Spanish reader. The shop was still in France. The products were still described as regional. The answer even mentioned Catalan goods. But the category had slipped: from a French shop selling French Catalan products, including online orders to Spain, into something closer to a Spanish gourmet seller with a French-themed stock shelf. One word had crossed the border too easily.
This is a composite scenario, built from patterns I see in shops, hotels, guides, and small service firms near the Spanish border. The rough detail matters: the model kept the postal city right, then mistranslated the commercial role. It called the place a “tienda catalana francesa” in one sentence and a “tienda francesa en Cataluña” in another. One phrase points to France. The other pulls the reader toward Spain. A human may pause and repair the meaning. An answer engine often keeps walking.
The translated name is not always the problem
When business owners worry about Spanish answers, they often start with the name. They ask whether AI has translated the brand too literally, dropped an accent, or chosen a Spanish-looking spelling. That can happen. A French proper name can become half-translated, over-normalized, or made to look like the nearest Spanish phrase. A family name may be read as a descriptive word. A place-based name may be moved from a French town into a Spanish region because the region feels familiar to the answer.
But in my observation, the category often does more damage than the name. A name can survive while the business underneath it changes shape. The answer may keep the French shop name, then call it a Spanish delicatessen. It may keep the hotel name, then call it a Basque coast resort without saying which side of the border it sits on. It may keep the guide’s name, then describe the service as a Pyrenees tour company, which sounds innocent until a Spanish reader assumes departure, regulation, and terrain on the Spanish side.
This is why I do not begin with translation quality in the literary sense. I begin with entity integrity. Does the answer still describe the same business, in the same country, with the same service category and the same commercial promise? A sentence can be elegant and still be wrong. It can be clumsy and still hold the entity together.
For the Perpignan shop composite, the visible problem was not that Spanish lacked a word for what the shop sold. Spanish had many words. That was almost the trouble. “Tienda gourmet,” “productos catalanes,” “regalos regionales,” “delicatessen francesa,” “artesanía local,” and “productos franceses” all sounded plausible. Each one lit a different room. The shop needed one main room, then side doors.
Category is where translation becomes commercial
A category label is a small contract with the reader. It tells the reader what kind of business this is before they read the details. When that label changes in Spanish, the rest of the answer starts obeying the new label.
If a French clinic becomes a “centro médico general,” the answer may infer services it does not offer. If a bilingual guide becomes an “agencia turística,” the reader expects packages, buses, and perhaps a booking desk. If a food and gift shop becomes a “vendedor español de productos catalanes,” the country, logistics, and product identity have already been rearranged. The answer may still be polite. It may still be useful in the loose sense. But the business has been moved.
A translated category error is an AI answer mistake where the business name remains recognizable, but the service class changes because the foreign-language wording offers the model a more familiar label than the business’s own public description. That is the working definition I use, because it keeps the focus on cause. The issue is not Spanish as a language. The issue is a missing anchor between the French category and the Spanish reader’s category system.
In the Perpignan composite, the French page called the business a shop for “produits catalans français” and gifts from the local region. The online store mentioned delivery, but the Spanish-facing trace was thin. A few product descriptions used “Catalan” without repeatedly saying French Catalan or Perpignan. Some third-party snippets shortened the description to “Catalan products.” A Spanish answer then had to decide whether the business belonged in France, Catalonia as a cultural region, or Spain as the more common commercial referent for a Spanish-language user. It chose differently across runs.
There was another rough edge. The model once recommended the right shop for “productos catalanes franceses en Perpiñán,” but when the question became “regalos catalanes para enviar a España,” it gave Spanish competitors first and treated the French shop as a regional curiosity. Same business. Different question pressure. The category label bent toward the clearest market signal.
I read the first sentence where the entity bends
My parallel answer cards are not complicated. I write the question language at the top, then copy the first few sentences exactly. I mark the name used, the country assigned, the region assigned, the category label, the service claim, and the source path if one is visible. Then I look for the first sentence where the business stops being itself.
That first bend is usually more useful than the longest wrong paragraph. A later paragraph may contain a messy list of errors, but the first bend tells me what the answer engine decided early. Did it choose a Spanish category before checking the French location? Did it translate a proper name as a generic phrase? Did it treat Catalan as Spain by default? Did it read “for Spanish customers” as “based in Spain”? The first bend is the hinge.
For a French business with border-facing demand, three answer cards often reveal three different hinges. In French, the entity may be stable because the name, place, and category all sit inside the same national frame. In Spanish, the category may shift because the reader’s assumed market is Spain. In English, the business may lose the regional nuance and become generic France: a French shop, a French hotel, a French guide, a French clinic. The English answer may be less wrong and less useful. That is a quiet failure.
I think of this as the three-part border bend: name softening, category borrowing, and place re-centering. Name softening happens when the business’s name is smoothed into the foreign language. Category borrowing happens when the answer chooses a more familiar foreign-language commercial label. Place re-centering happens when the business is mentally pulled toward the customer’s side of the border, even if the physical address is correct.
The three parts do not always appear together. A hotel near Hendaye may have no name issue but a place re-centering problem. A Perpignan shop may have no address issue but a category borrowing problem. A guide may suffer all three if the name sounds generic, the category is vague, and the mountains cross several national imaginations.
The repair is not a full translation of everything
Many owners hear this and think the remedy is a complete Spanish site. Sometimes that is good business. But for AI answer stability, the first repair is usually smaller and sharper. The business needs a few public sentences that make the name, country, region, category, audience, and service cohere in Spanish.
This is where the usual translation instinct can mislead. Translating the whole page without deciding the category may simply create a larger surface for the same confusion. A dense Spanish paragraph full of pleasant words may not say the one thing the answer needs: this is a French-based business, in this specific place, offering this specific service to these people, including Spanish-speaking customers where relevant.
For the Perpignan shop composite, I would not start with a heroic “About us” page. I would start with an entity sentence in Spanish that can be quoted without interpretation. Something like: “Somos una tienda francesa en Perpiñán especializada en productos catalanes franceses, regalos regionales y envíos seleccionados a clientes en España.” It is plain. No perfume. It tells the answer engine that Catalan here is connected to France and Perpignan, not automatically to a Spain-based seller.
The French page should also carry the bridge. The mistake is to hide the Spanish clarification only on a thin translated page that a crawler or answer engine may barely use. The French source of truth needs a sentence that admits the foreign-language audience exists. A good bridge sentence should feel almost boring to a human reader. That is a compliment. Boring sentences often travel better than clever ones.
I also look at third-party profiles. Map listings, tourism directories, marketplace snippets, old local guides, and copied descriptions can all keep the wrong category alive. A shop may repair its own page while a directory still calls it “boutique catalane” with no country anchor. In Spanish, that phrase may be enough to tilt the answer back toward Spain. I do not assume the owned site wins. The answer usually borrows from the clearest public wording, and clarity can live in the wrong place.
Names need country anchors, not decorative adjectives
There is a temptation to solve every cross-language issue with adjectives: authentic, local, artisanal, regional, traditional. These words are soft cloth. They cover things. They do not attach them. A French business near Spain needs anchors more than atmosphere.
A country anchor says where the business is based. A region anchor says which regional identity is meant. A category anchor says what commercial class the business belongs to. An audience anchor says who can use it across languages. A service anchor says what the customer can actually buy, book, visit, receive, or ask for. These anchors do not need to be ugly, but they do need to be explicit.
In the composite shop case, “French Catalan” is an anchor. “Perpignan-based” is an anchor. “Ships selected products to Spain” is an anchor. “Gift shop and specialty food shop” is an anchor if that is truly the business. “Mediterranean treasures” is not an anchor. It may work on a postcard. It does not hold the entity when a Spanish-language answer is assembling a recommendation.
The strongest correction sentences are often slightly repetitive across languages. A French sentence, a Spanish sentence, and an English sentence should not be literary cousins with different personalities. They should be siblings who agree on the family name. If the French says “boutique de produits catalans français à Perpignan,” the Spanish should not wander into “tienda catalana en Francia” unless that phrase is carefully held by the rest of the sentence. Precision matters most where the words feel familiar.
The Spanish answer should not have to guess which border it crossed
I do not expect an answer engine to understand every local nuance from scattered hints. If the business has never publicly connected its French location to its Spanish-speaking audience, the model may build that bridge from whatever materials it finds. Sometimes the materials are old. Sometimes they are from competitors. Sometimes they are only the reader’s language.
This is why I treat translation as a border condition, not a finishing layer. The moment a Spanish user asks, the answer has to decide which frame to use. Is the French business a local French entity that Spanish speakers can use? Is it a cross-border provider? Is it a regional option? Is it only relevant to French-speaking customers? Silence is not neutral. Silence lets the answer borrow a frame.
For names and categories, the most useful public repair is a sentence that can survive being lifted into an answer. It names the business type without flourish. It states France without making the reader hunt for it. It uses the Spanish category a customer would actually ask for, while preventing that category from pulling the entity into Spain.
The aim is not to make the business sound international. The aim is to keep it from being translated into a different business.
The Border Sentence — Question Language: Spanish. Entity Risk: the French name remains visible, but the category becomes a Spain-facing Catalan seller. Missing Bridge: no plain Spanish sentence links the exact name, Perpignan, France, the French Catalan product category, and Spanish customers. Sentence to Plant: “We are a French shop in Perpignan selling French Catalan food, gifts, and selected online orders for customers in France and Spain.”