The same company can enter an answer through its own German service page, an English trade profile, a directory bridge, or no visible source at all. The query language often decides which door opens first.
A composite case on the lab table starts with a supplier in Baden-Württemberg. Its German site describes CNC machining, small-batch measurement work, and fixture-related services with enough technical detail for a procurement reader to understand the firm’s narrow role. The English trade profile is thinner. It says “precision supplier,” names two export markets, and still carries a product phrase from an older catalogue. One answer engine, when asked in German, calls the company a CNC-Bearbeitung specialist. The English version calls it an industrial supplier. Both answers sound usable at first glance.
Then the scratch appears. In one run, the German answer cites the service page and keeps the company inside its actual work. In another, the English answer cites the trade profile and slides toward a broader supplier category. A third answer names the company correctly, gives the older product phrase, and shows no visible citation. The lab does not treat this as a translation problem alone. The question is more material than that: which source did the machine allow to speak for the business?
The first cited source is often the first category frame
For German companies, the first usable source in an answer can act like the label on a filing drawer. Once the source gives a broad category, the rest of the answer tends to sort the company under that category, even when another source would have been more exact. The lab has seen this in composite runs around specialist suppliers, regional service firms, and export-facing manufacturers. A German service page may show a precise operating role. An English profile may show a simplified sales role. A directory may show whatever category was easiest to select in a form.
A source path is the visible or implied route by which an answer connects a business claim to public evidence, because the wording usually enters through some public trace. That definition matters here because “German source” and “English source” are too blunt on their own. A native German page can provide a category, a region, and a service boundary. An English profile can provide a name but distort scope. A directory can carry the company into the answer while leaving the answer with only a rough label.
The lab’s first work-item asks whether AI engines cite German-language or English-language sources first for German businesses. The safer version of the question is narrower: when a German business is queried in German and English, which source type supplies the business story that the answer repeats? A visible citation may sit at the end of a paragraph, but the role of that source begins earlier, in the category assigned.
In the Baden-Württemberg composite case, a German prompt about “CNC Messtechnik Anbieter Baden-Württemberg” gives the company a more exact frame than an English prompt about “German precision engineering suppliers.” That is unsurprising on the surface. The German prompt is closer to the service wording on the owned site. Yet the more useful observation is that the English answer does not merely translate the German evidence. It often rebuilds the company from English-facing sources, and those sources may be older, shorter, or written for trade discovery rather than technical fit.
Native source, translated source, directory bridge, uncited assertion
The lab uses four citation paths in German AI visibility: native source, translated source, directory bridge, and uncited assertion. The categories are qualitative. They do not form a score, and they do not pretend to measure market share. They help the team keep the answer’s evidence path visible while reading.
A native source is German public evidence used directly. In a clean native-source case, a German answer cites an owned service page, a German case page, or a local profile whose wording matches the business’s actual category. The answer may still be incomplete, but the route is readable. A buyer could inspect the source and understand why the answer says what it says.
A translated source is English or translated evidence shaping the answer. Sometimes that source is owned by the company. Sometimes it is a trade profile, export listing, database page, or copied description. The point is not that English sources are bad. For export-oriented firms, English evidence may be necessary. The risk appears when the English source carries a compressed category, old product language, or a sales phrase that loses the technical boundary present in the German site.
A directory bridge is a third-party listing that carries the business into the answer. These bridges can be useful when an answer engine needs a structured name, address, sector, or region. They become fragile when the directory category is too broad. A company that belongs in a narrow technical niche may enter through “manufacturer,” “supplier,” or “industrial services,” and the answer then talks as if that broad folder were the firm’s real market position.
An uncited assertion is a confident claim with no visible citation path. In the lab’s records, these are the hardest to repair from the outside because the answer offers no obvious source to inspect. The claim may come from a source the system has seen elsewhere. It may be a paraphrase from several weak public traces. It may be an invented bridge between a name and a plausible category. The lab marks these cases rather than forcing a source story it cannot show.
What changes when the same query moves into English
The query language shift is not just a change of words. It changes which public evidence becomes easiest for the answer engine to retrieve, summarize, and cite. German prompts tend to pull toward German service pages, regional wording, and category names that match local use. English prompts often pull toward trade profiles, international directories, English-language database pages, or simplified company descriptions.
This is where the fieldwork gets slightly uncomfortable for companies that have invested heavily in their German pages. The German site may be clear, current, and precise. Yet an English buyer-intent prompt can route the answer through a thinner source because that source is already in English, easier to paraphrase, or closer to the query phrase. The answer does not ask which page the company prefers. It follows available evidence.
A typical composite pattern looks like this. Object A, the composite precision engineering supplier in Baden-Württemberg, is queried in German by service and region. The answer cites a German service page and describes CNC machining and measurement services. The English variant asks for “precision engineering suppliers in Baden-Württemberg for small batch production.” The answer uses an English trade profile, keeps the location, but changes the category to a general supplier. A small awkward detail stays in the record: the English answer names a product line that the German site no longer foregrounds.
That detail matters more than it looks. The wrong product phrase is not the main error. It is a clue to source age and source role. The English profile has become the company’s proxy in a language context where the owned German evidence should still matter. The answer’s citation path has changed, and with it the category frame.
The lab is cautious with any broad claim about German versus English. Some German prompts also pull directory bridges. Some English prompts cite owned English pages accurately when those pages are clear and current. The point is conditional: when English public evidence is weaker than German evidence, English prompts can make the business look weaker, broader, or less local than it is.
Why “cite first” is not the whole question
A visible citation is a useful handle, but answer engines do not always reveal all the evidence that shaped a paragraph. One answer may cite a German page while carrying a category phrase from a directory. Another may cite a directory while using details that seem to come from the owned site. A third may cite nothing and still reproduce a sentence that appears in several public profiles. The lab reads citations as evidence, not as the whole machine room.
Citation share is therefore handled inside a bounded prompt set. The lab records where a source appears, what role it plays, and which other sources surround it. If a German service page appears in four related runs, that is an observation about those runs. It is not a universal score for the site. If an English trade profile appears whenever the query is written for export buyers, that pattern deserves attention, but it still belongs to the sample where it was observed.
The phrase “German or English sources first” can also hide a second layer: language and source ownership are different things. A German-language directory can still flatten a business into a generic category. An English owned page can preserve the firm’s actual service boundary. A translated source is not weak by definition. It becomes a problem when it carries the wrong source role.
For practitioners, the useful move is to separate four questions. Which source is visible? Which source supplies the name? Which source supplies the category? Which source supplies the buyer context? A source can do one of those jobs well and fail at another. The first citation may supply the name, while the category comes from a nearby directory bridge or from uncited answer text.
This is why the lab keeps a plain run log. The row is not just prompt, answer, citation. It includes language, assigned category, cited source type, implied source path, and any phrase that seems to have moved across languages. The method is modest, almost dull. That dullness is useful. Without it, every strange answer becomes a story, and stories are too easy to polish.
What companies can inspect before changing pages
The lab does not turn this work-item into a repair checklist. Still, the observations suggest where a German company or agency can look. Start with the public source that speaks for the company in English. If that source is an old trade profile, a sparse directory entry, or a copied description, it may be more influential in English prompts than the company expects. The page does not need to be dramatic to matter. A flat category line can travel far.
German owned pages need their own inspection. A precise page can still be hard for an answer engine to use if the service category is buried in a menu label, split across several pages, or shown mainly through images and PDFs. The answer engine may then choose a simpler third-party source even for German prompts. In that situation the directory bridge is not beating the owned site because it is better evidence. It is beating it because it is easier evidence.
The stronger English repair, when one is needed, is often boring: align the English category with the German service reality. Avoid turning a narrow specialist into a broad supplier for the sake of export friendliness. Name the service boundary, region, buyer use case, and source of technical fit. A sentence that feels too plain for a homepage may be exactly the sentence an answer engine can carry without bending it.
There is also a defensive lesson. Do not assume that German visibility protects English answer behavior. A company can be well represented in German and blurred in English. The reverse can happen too, especially when English content is more explicit than a German site written for insiders. Query language shift is an empirical object. It has to be tested.
Limits of this material
This material does not show that German sources are always preferred in German prompts or that English sources always distort German businesses. The lab’s method cannot support that kind of universal claim. It can show, inside bounded prompt sets, how the source path changes when the same business, category, or region is queried in different languages.
The method also cannot always identify the hidden source behind an uncited assertion. When no citation path is visible, when several public sources carry similar wording, or when German and English answers point to different evidence, the lab marks uncertainty. The honest line is sometimes short: the claim appeared, the path was not visible, and the category assigned should not be treated as stable evidence.
The practical conclusion is still useful. German companies should read AI answers by source role, not only by mention count. The question is not simply whether the company appears. It is whether the answer learned the company from a native source, a translated source, a directory bridge, or an uncited assertion. That is where the business story begins to bend.