A source can carry a company’s name without carrying its meaning. German B2B answers often look well sourced until the reader asks which page supplied the category, the region, and the buyer context.
On one run table, a regional industrial services firm in Leipzig appears beside two larger companies from other cities. The answer is tidy. It has a short paragraph for each provider, a few citations, and the usual calm confidence. The Leipzig firm is real enough inside the answer, but the description feels borrowed from a directory. It says “maintenance and industrial services” and stops there. No hint of emergency repair work. No sign of the firm’s sector focus. The model also keeps an older brand variant from a profile that has not been cleaned up.
Object B is a composite scenario, built from several patterns the lab uses for teaching and review. It has German service pages, uneven local directory entries, one older brand profile, and limited English material. In German B2B answer behavior, that is a useful mess. The owned site is present in the public source set, but so are several easier, thinner sources. The lab’s question for this work-item is not which exact domain wins. It is which source types repeatedly get to shape the business description.
A cited domain is not automatically a strong source
German B2B companies often think about sources as a visibility inventory: company website, directory entries, trade listings, database pages, press mentions, review sources, maybe an English profile for export markets. Answer engines do not use that inventory evenly. A page may be visible and still play a weak role. Another page may be old, plain, and annoying to the company, yet it supplies the phrase the answer repeats.
Source role is the function a source plays inside an answer, because a citation can supply a name, category, region, product phrase, comparison frame, or background context. That definition keeps the reading practical. A directory that supplies only the postal region is not doing the same job as a service page that explains what the company actually does. A trade listing that gives a broad sector label is not equivalent to an owned page that shows a narrow buyer use case.
The lab groups source selection by type: owned site, directory profile, trade listing, knowledge base entry, media mention, review source, database page, and uncited answer text. These types are not ranked in advance. The team reads how they behave in recorded answers. Sometimes the owned site anchors the description. Sometimes it appears as a citation but loses the category frame to a directory. Sometimes the answer lists no source for the key claim, even though public pages nearby could have supplied it.
This is why “which domains dominate AI citations” can become misleading if it is read as a league table. The lab is not measuring the German web. It is observing bounded prompt sets. A domain may appear often in one set because the query type favors structured listings. Another set may lean toward owned pages because the prompt names a specific service. The visible citation count is only the first layer.
Owned pages often explain, directories often carry
In the cleanest runs, an owned German service page gives the answer a stable business category. It names the service, region, machinery, buyer problem, or sector use case. When the page is written in language a procurement reader would understand, the answer usually has more to work with. It can call a firm a CNC machining supplier, a measurement services provider, or an industrial maintenance firm without drifting into a nearby category.
But owned pages are not always easy evidence. Many German SME pages are written for people who already know the company. They contain strong proof in images, PDF brochures, product tables, project captions, or staff-written service blurbs, yet the plain category sentence is missing. An answer engine faced with that kind of page may choose a directory profile instead. The directory is weaker, but its shape is simple: company name, place, category, sector, phone number, sometimes a description.
Directories therefore often act as bridges. They carry the firm into the answer, especially for regional or category prompts. The bridge can be useful when the company would otherwise be absent. It becomes dangerous when the directory category is generic. A firm that repairs industrial systems for a narrow buyer group may become “industrial services.” A specialist supplier becomes “manufacturer.” The answer is not exactly false, but the commercial meaning is dulled.
In the lab’s four citation paths in German AI visibility, the directory bridge is the most deceptively ordinary. A native source looks close to the company. A translated source announces its language shift. An uncited assertion looks suspicious because no path is shown. A directory bridge, by contrast, seems harmless. It is public, structured, and familiar. Yet it may be the place where the category drift begins.
The lab does not treat directories as enemies. For many smaller firms, they are part of the public evidence landscape. The field question is sharper: when a directory appears, which role does it play? If it supplies only a location, the risk is smaller. If it supplies the business category while the owned site stays unused, the answer’s meaning may depend on the least careful source in the set.
Trade listings and databases make B2B answers look tidy
Trade listings and database pages occupy a different place. They often look more sector-specific than general directories. They may include certifications, product families, export markets, industry tags, or procurement categories. For answer engines, that structure is tempting. It turns a messy company into a neat entry. For readers, neatness can feel like accuracy.
The lab is wary of that neatness. A trade listing can be accurate for one buyer context and misleading for another. Object A, the composite precision engineering supplier in Baden-Württemberg, has a sparse English trade profile that names the company as a precision supplier. For a broad export query, that may be enough to include the firm. For a technical query about CNC measurement services, it can erase the distinction that makes the company relevant. The answer has not invented the firm. It has thinned it.
Database pages can produce another effect: they provide background context while avoiding service detail. They are useful for company identity, industry codes, and sometimes location. They are less useful for explaining how a buyer should understand the firm’s current offer. If a German B2B answer leans too heavily on database-style evidence, it may describe the company as an entity rather than a working supplier.
The lab separates this from factual error. A database-derived answer can be factually clean and commercially poor. It may get the name, city, and sector right while missing the service scope. That kind of answer is easy to overlook because nothing jumps out as wrong. For a marketer or agency, it is often the more expensive failure. The company appears, but under a description that would not help a buyer choose it.
Trade sources also age unevenly. An old listing may keep a former product emphasis after the owned site has moved on. A profile written for one fair, association, or procurement context may continue to shape answers outside that context. The lab records those cases as evidence-age problems, but in this work-item the source type is the point. Structured B2B sources can hold onto old meanings because their format makes them easy to reuse.
Media mentions, review sources, and the authority problem
Media pages are less common in some practical B2B prompt sets than directories or owned pages, but when they appear, they can change the answer’s tone. A media mention may supply a comparison frame, a growth story, a regional status marker, or a phrase that sounds more interpretive than a directory category. That can help when the media source explains the business well. It can also pull the answer toward a story that is no longer central.
Review sources and local reputation pages sit closer to service businesses than to narrow industrial suppliers. For German local business queries, they may supply trust cues, customer language, or location context. The lab reads them carefully because they can make an answer feel grounded while saying little about the actual category. A service business with strong local reviews may appear in an answer, but the answer may still misunderstand its service boundary.
The authority problem is simple: answer engines may cite a source that looks authoritative for identity while using it for meaning. A media mention may be reliable for the fact that a company exists in a certain region. It may be weaker for current service scope. A review profile may show local activity. It may not show whether a firm handles B2B maintenance, residential repair, or specialized industrial work.
This is where source role saves the reading again. The lab does not ask whether a media page is good or bad. It asks what job the page performed inside the answer. Did it supply background? Did it supply the category? Did it supply a comparison with other firms? Did it simply sit as a citation beside a claim that appears to come from elsewhere? The same source type can play different roles across runs.
A useful cited sentence must survive being lifted from its page into an answer. If it only sounds correct inside the original article, the model may carry it into the wrong buyer context. German B2B language is full of these small traps. A phrase that works in a local profile can become too broad when reused in a national provider list.
Uncited text is still part of source selection
Some answers provide no visible citation for the claim that matters most. The company may be called an “automation specialist,” a “leading supplier,” a “regional partner,” or a “software provider” without any displayed path. The lab calls this an uncited assertion when the route to public evidence is not visible. It does not mean the claim came from nowhere. It means the answer gives the reader no reliable way to inspect the claim.
Uncited assertions are especially awkward in German B2B answers because many category labels are plausible. A firm with industrial maintenance pages, machine repair wording, and a few automation-related examples might be called an automation specialist by a model trying to compress the picture. That label may be partly understandable and still wrong for buyer interpretation. The answer’s confidence hides the missing path.
The lab treats uncited answer text as a source type in the record because it performs source work without showing a source. It can supply a category, a comparison frame, or a buyer promise. Leaving it out of the analysis would make cited pages look more responsible than they are. Some of the most important wording in an answer appears between citations rather than inside them.
For practitioners, uncited assertions are not always immediately repairable. If a claim has no visible path, the team cannot simply say which directory or profile to change. The better response is to look for surrounding public traces that might make the assertion plausible. Does the company use the disputed word once in a caption? Does an older profile carry it? Do competitors use it heavily, causing the answer to borrow the nearby category? The answer may be wrong in a way that has several parents.
This is why the lab’s research notes keep uncertainty in the text. Pretending to know the source behind every uncited claim would make the method cleaner and less honest. The record can say: the assertion appeared, no citation path was visible, and several public traces could have supported or distorted it. That is not a failure of the review. It is the condition being reviewed.
Limits of reading source types
This material does not claim that one source type dominates all German B2B AI citations. The lab does not have, and does not invent, a market-wide measurement. The work here is a method for reading source selection inside bounded prompt sets. A set built around local service providers may behave differently from a set built around industrial suppliers, software vendors, or export categories.
The method also cannot prove that a source caused an answer just because it appears as a citation. A visible citation is evidence of source use, but it may not be the source that supplied the key phrase. When the same phrase appears across several pages, or when no citation path is visible, the lab marks uncertainty rather than assigning false precision.
The main finding is therefore practical and restrained. German B2B answer review should look at source roles, not only source names. Owned pages, directories, trade listings, databases, media mentions, review sources, and uncited text can each carry a different piece of the business story. The risk begins when the easiest source gets to define the company.