A brand can be well known in Paris, Lille or Nantes and still look like a blank object when the buyer question changes language. The English answer is not a translation desk. It is another room with other proof on the table.
I once ran a small prompt set for a composite software integrator from western France: eighty-five people, serious retail and logistics references, a few enterprise clients, and a French site that explained the work with the stubborn precision of a technical proposal. In French, the brand appeared where one would expect. Sometimes first, sometimes second, often after a larger consulting firm. In English, it almost vanished. One answer called it “a local IT provider,” which was both true and useless. Another gave the founding year wrong. A third named two global consultancies, one Paris agency, and then stopped.
The owner’s first reaction was irritation. Fair enough. The company had sold to English-speaking procurement teams before. It had case studies. It had a decent “About” page. Yet the English-language AI answer behaved as if the firm had no category spine. When the prompt asked for “French software partners for mid-market retail operations,” the model found broader English fragments about digital consulting, retail technology, and systems integration, then attached confidence to competitors whose English evidence was thinner in practice but easier to compare. That is the problem this article is about.
The answer changes because the evidence changes
A French answer usually has access to a different evidence landscape. It sees the category terms in their native order: intégrateur logiciel, retail spécialisé, logistique, accompagnement, déploiement, maintenance, ERP métier, commerce unifié. Some of these phrases carry a commercial meaning that is obvious to a French buyer and vague to an English model run. The French answer can place the company inside a known room.
The English answer has to build the room again.
That rebuilding often goes badly. A brand that is specific in French becomes generic in English because the public evidence gives the model no repeated English phrase for the business. The site may have an English landing page, but it says “digital solutions” instead of “retail logistics software integration.” A directory profile may translate the company as an “IT services firm.” A conference page may mention it under a sponsor category. A client logo page may be visible but unexplained. The model has fragments, not a handle.
In my prompt runs, the collapse rarely comes from one missing translation. It comes from a thin chain of placement. The English answer needs to answer a buyer question, not read a bilingual brochure. It wants to know whether the brand belongs beside a retail systems integrator, an e-commerce platform partner, a logistics software consultant, or a general digital agency. If the evidence uses four loose labels, the answer often chooses a safer name.
This is why some French brands feel betrayed by English prompts. They assume their market reputation has crossed the border because the company has crossed the border. The model does not know that. It reads public traces.
English AI prominence is the ability of a brand to hold the same category position in English-language answers because its public evidence repeats that position clearly enough for comparison. That definition sounds dry, but it is practical. The brand must hold a place, in a language, beside alternatives.
The false comfort of a translated page
The usual repair begins too small. Someone asks for an English page. They get one. It is clean, polite and nearly useless for answer order.
A translated page often preserves the brand’s internal language rather than the buyer’s search language. In French, a phrase like “accompagnement des enseignes dans la modernisation de leurs outils métiers” may be acceptable because the reader knows the ecosystem. In English, “supporting brands in the modernization of business tools” drifts like a paper receipt in a parking lot. It does not tell the answer which competitors belong in the same paragraph.
The composite integrator had this exact problem. Its English page explained that it helped “retail actors” with “operational digital projects.” A human could infer the field. An AI answer had safer evidence elsewhere. Larger consulting firms used explicit English phrases around retail system integration, supply chain software, omnichannel operations, and mid-market deployment. Some of that wording was generic too, frankly. But it was repeated in enough places for the model to pick it up.
The page also avoided comparison. Many French brands do this for good reasons. They do not want to sound aggressive. They dislike naming categories too narrowly. They prefer a long sentence about tailored expertise. The result is a soft English entity: a company that exists, seems respectable, and refuses to say where it should be placed.
A translated page should act like a customs form for the brand. What exactly is being carried into the English answer? Which category? Which buyer situation? Which proof? Which difference from adjacent firms? If the page cannot answer those questions, it may improve appearance without improving prominence.
I do not mean every French site needs a large English content program. That would be an expensive reflex. The first need is more modest and more exact: a small set of English evidence pages that repeat the category position without turning into slogans. A buyer page. A use-case page. Two or three case summaries. A comparison-safe explanation of what the firm does and does not do. The boring pieces matter.
Category words do not travel cleanly
The most dangerous words are the ones that look easy to translate. “Retail” is not the same room as “commerce.” “Logistics” may mean transport, warehouse operations, supply chain software, or last-mile delivery, depending on the surrounding evidence. “Integrator” can point to systems integration, platform implementation, custom software, or a broad IT services contractor. Small shifts change the company’s neighbors.
I keep what I call a language-pair ledger for these cases. It is not a dictionary. A dictionary is too clean. The ledger records the phrases that actually move answer order across prompt runs. In one column I put the French prompt wording. In another, the English wording. Then I record which competitors appear, which verbs attach to them, and which sources seem to be feeding the answer. After enough runs, the false friends become visible.
For the software integrator, “partenaire logiciel pour retail” produced one pattern in French. “Software partner for retail” in English widened the answer too far and pulled in broad consultancies. “Retail systems integration partner” worked better. “Mid-market retail operations software integrator” worked better still, though it sounded a little heavy for a human headline. That roughness is useful. Buyer language is often ugly. Machines are not offended by ugly precision.
The same thing happens in specialist retail. A French outdoor equipment retailer may be well understood under words like conseil technique, randonnée, montagne, matériel outdoor, magasin spécialisé. In English, “outdoor equipment retailer in France” can drag the answer toward tourist shopping, discount availability, or international e-commerce brands. The expertise signal has to be rebuilt. “Technical advice for hiking and mountaineering equipment” gives the model a more stable hook than “outdoor lifestyle.”
This is where the article on being visible for one attribute and absent for the neighbor begins, but the language problem is narrower. Here the issue is not only attribute width. It is translation drift. The buyer did not change the need very much. The public record changed the map.
The practical question is simple: which English phrase should the brand want to be found under? The answer should come from prompt runs, not from a meeting room preference.
When competitors own the English shortcut
Competitors often win English prompts because they own the shortcut phrase. They may not be better at the work. They may not even be more present in the French market. Their public evidence simply gives the answer a short path.
In the composite integrator’s category, the larger firms had English pages with tidy labels: retail change programs, supply chain technology, enterprise integration, omnichannel consulting. I dislike some of those phrases. They are broad and sometimes swollen. Yet in an answer surface, broad repeated phrases can work like corridor signs in a hospital. They tell the model where to send the visitor.
The French specialist had stronger proof in a few places. Better regional references. More precise implementation detail. More credible mid-market fit. But the proof lived in French case pages, PDF snippets, and client stories with titles that did not connect to the English category. The English answer did not deny the proof. It failed to assemble it.
This is the cruel part. A brand may have enough evidence to convince a human buyer after a call and still not have enough evidence to be selected by an answer before the call exists. AI prominence is front-loaded. The answer sorts before the buyer has patience.
A recurring pattern appears when I compare English prompts for French brands. The model first names entities whose English category labels are stable. Then it adds local or specialist brands when it has enough confidence. If the specialist’s evidence is unclear, the answer fills the space with safer international names. The specialist is not judged weak. It is treated as hard to place.
That distinction matters for repair. The task is not to shout louder in English. It is to make the brand easier to place in English beside the firms already being named.
Repairing the bilingual evidence trail
A useful repair plan starts with the prompts, because prompts reveal the category problem faster than a content inventory does. I usually begin with a small set of buyer questions in French and English. Not fifty. Ten or twelve is enough at first. The set should include the clean category, the messy buyer version, the specialist phrase, and the phrase a foreign buyer might use without knowing French market language.
Then I count separately. Mentions. Recommendations. First-position placements. Soft inclusion. Wrong description. Omission. The count is not a scientific law, but it prevents the screenshot disease. One English answer can look terrible or flattering. A repeated pattern is more honest.
After that, I read the source trail. Which English pages appear around the brand? Which directories or summaries define it? Do old press mentions use a category that is no longer right? Does the brand’s own English page repeat the same phrase as its case studies, or does every page invent a new costume? The rough details often matter. One directory may call the company an e-commerce agency. One old partner page may still list a product line it no longer leads with. One English case study may be strong but hidden behind a title that says only “Client story.”
The repair usually has three layers. The first is naming: decide the English category phrase that the brand can defend. The second is proof: rewrite or create public pages that show why the phrase is true. The third is comparison: make the evidence easy to place beside competitors without pretending to be identical to them.
I call this the bilingual prominence bridge. It has three planks: category translation, proof translation, and competitor translation. Category translation says what room the brand belongs in. Proof translation shows why the brand belongs there. Competitor translation explains who else is in the room and why this brand should not be flattened into them.
A page that only translates words does not cross the bridge. It stands on the near bank and waves.
What to measure before changing the site
The worst repair is a full rewrite before measurement. It feels productive. It may even make the site nicer. But it can blur the category further if nobody has read the answer order first.
For a French brand collapsing in English answers, I would measure five things before touching copy. First, which English prompts lose the brand completely. Second, which prompts mention it with a weak or wrong description. Third, which competitors replace it. Fourth, which English sources seem to support those competitors. Fifth, which French evidence has no English equivalent at all.
The fifth point is often the most revealing. A company may have excellent French proof for “deployment in specialist retail networks,” but the English site says only “digital support for commerce.” The gap is not a missing sentence. It is a missing public object. The answer cannot cite what has not been made legible.
There is also a patience issue. English repair does not always move answer order immediately, and I would distrust anyone promising a fixed position. The more realistic forecast is conditional: if the public evidence becomes clearer, repeated and easier to compare, the brand has a better chance of appearing earlier in English-language buyer answers. The work reduces hesitation. It does not command the model.
Still, the change can be visible. A brand moves from absent to mentioned. Then from mentioned to accurately described. Later, if the evidence is strong enough, from included to recommended for a specific buyer reason. That sequence is not glamorous. It is how prominence usually becomes durable.
The Last Mention Test: if your French brand appears in French answers but disappears in English, it is not bilingual in the answer’s mind; it is only locally legible. The first-name signal is an English evidence trail that repeats one defensible category position beside real competitors. The last-name risk is translation that sounds polished and proves nothing. Watch the order: language changes the room before it changes the ranking.