A company can keep the same legal name and still become three different entities in AI search. The split usually starts in public profiles that describe role, category, and geography differently.
In a composite scenario drawn from service-company audits, a 42-person firm in North Rhine-Westphalia had a plain enough business when described by its managing director. It handled compliance documentation and supplier onboarding for mid-sized industrial firms. Not glamorous work. Useful work. On its own website it sounded like a specialist service provider. In one directory it looked like a local consulting office. On an English page it drifted toward software platform language. In a procurement portal it sat under administrative services.
The AI answers inherited the mess. ChatGPT called it a local compliance consultant in one run. Perplexity framed it as a supplier onboarding platform in another. A Google AI Overview-style summary placed it near general back-office outsourcing. The strange detail was that the company name was stable. The address was stable. Even the logo was stable. The identity was not. Public profiles had renamed the firm without changing its name.
Entity footprint is more than a name
People often think of entity work as making sure the company name, address, and website match. That is only the bottom layer. It matters, but it does not settle the question an answer engine has to answer. The engine also needs to know what the business is, what role it plays, which market it belongs to, and why it belongs there.
An entity footprint is the public pattern of names, roles, categories, locations, and supporting sources that makes a business recognizable across AI search, because machines infer identity from repeated evidence rather than from a company’s preference. That definition sounds heavier than “keep your listings updated,” because the problem is heavier.
A German SME may appear on its own site, Google Business Profile, industry directories, association pages, procurement portals, job boards, PDF catalogues, old press notices, Wikidata entries, and sometimes Wikipedia. Some of these profiles are created by the company. Some are scraped. Some are copied from older sources. Some are technically editable but politically difficult to change. Each source contributes a small vote about the business.
The trouble is that the votes are not all about the same thing. One source votes on category. Another votes on geography. Another votes on history. Another votes on products or services. If those votes point in different directions, AI systems may still name the company, but the summary becomes unstable.
That is how a firm gets found and misdescribed at the same time.
The three kinds of public-profile renaming
When I map German firm profiles for AI search, I usually separate three types of renaming. The first is literal renaming: old legal names, old English names, merged brands, spelling variants, abbreviations, or directory titles that never caught up. This is the easiest type to notice. It is not always the easiest to fix, but at least people can see it.
The second is category renaming. Here the company name remains correct, but the profile assigns the wrong business type. A service provider becomes software. A manufacturer becomes wholesale. A compliance documentation specialist becomes general consulting. A design office becomes a contractor. In AI answers, this type causes more damage than literal spelling variation because the company appears to be correctly identified while its role changes underneath.
The third is market renaming. The firm is placed in the wrong geography, buyer group, or use case. A company serving industrial clients across Germany becomes “local.” A regional service company with national contracts becomes tied to one city because the only strong citation is a map listing. An export-capable manufacturer becomes domestic-only because English profiles lack proof. This error is harder for internal teams to see because each source may be locally true. The head office is in one city. A category page mentions one buyer group. A press note covers one regional project. AI search stitches those fragments into a narrower business than the real one.
These three renamings often work together. The North Rhine-Westphalia compliance firm had a small literal issue: one older English profile used a former service name. The larger damage came from category renaming and market renaming. Its German site emphasized service expertise. Its English page used platform-like phrases. A procurement source placed it in administrative support. A local directory tied it tightly to one city. None of those sources was a total lie. Together they gave the answer engine too many masks to choose from.
Wikipedia and Wikidata are not magic identity machines
The search query “firma in wikidata eintragen” usually carries a quiet hope. If the company can get into Wikidata, maybe AI systems will finally understand it. Sometimes a structured public entity helps. Sometimes it does almost nothing. Sometimes it makes a weak record look more official.
Wikipedia and Wikidata have their own rules and community norms. Many ordinary SMEs will not belong in Wikipedia, and forcing the issue is usually a bad sign. Wikidata can represent entities more flexibly, but a sparse entry with a name, website, and broad industry label does not solve an evidence problem. It can even freeze a thin category into a public place that other systems repeat.
For AI search, the more practical question is not “can we get a profile?” It is “what claim would this profile support?” If a Wikidata item says the firm is an Unternehmensberatung, while the website says compliance documentation and supplier onboarding, the structured profile may strengthen the wrong category. If a directory says software company because the English page used “platform” too freely, the mistake may travel into summaries even if the company never sold software as a product.
I do not treat public profiles as trophies. I treat them as witnesses. Some are reliable witnesses. Some are confused. Some remember an old version of the company. Some heard the story from another witness who was also confused.
That is why the first repair is usually a ledger, not an edit. I put the public sources in one place and mark name, role, category, geography, language, and claim support. The pattern appears faster when the sources are made boring.
Directories often speak louder because they are blunt
Company websites tend to explain. Directories tend to label. For a human reader, the explanation is richer. For an answer engine under query pressure, the label may be easier to use. This is uncomfortable but important.
A directory category like “compliance consulting” or “supplier management software” may carry more weight than a careful services paragraph if the services paragraph never states the business type directly. An association profile may have one sentence that looks clean enough to quote. A procurement portal may assign an official-sounding category. An old local listing may say the firm operates only in Düsseldorf because that is where the office sits, even though the service contracts reach far beyond the city.
The composite North Rhine-Westphalia firm had a directory entry that described it as “Beratung für Lieferantenmanagement.” That was close but too narrow. Another source used “supplier onboarding platform,” probably because the English service page talked about digital workflows. The AI answer sometimes blended them into a software-consulting hybrid. A human would ask a follow-up. The machine made a category.
This is one reason I dislike profile cleanup that only checks whether details are “correct.” Correct is not enough. The profile must support the claim the business wants answer engines to repeat. If the firm is a service company that uses tools, the profile should not make the tool sound like the product. If the firm serves industrial clients, the profile should not bury that behind office-location language. If the company is not a general outsourcing firm, a broad administrative category needs correction or counterweight from stronger sources.
A blunt source can be useful. It just has to be blunt in the right direction.
The company site has to out-explain its profiles
Some marketing teams want to repair every external source first. That can be necessary, especially when a high-reach directory carries an obvious error. But the company site must usually do a harder job: it has to out-explain the public profiles without turning into a defensive document.
A strong entity page for AI search does not need to shout. It needs a clean identity sentence, service or product boundaries, buyer context, geography, proof, and language that distinguishes adjacent categories. For the compliance firm, that would mean stating that the company provides compliance documentation and supplier onboarding services for mid-sized industrial firms, then explaining where software supports the service without making software the entity category. It would also mean naming the region honestly while not reducing the service area to the office city.
The German and English versions need this structure in their own language. The English page should not simply borrow broad SaaS language because it sounds familiar to export buyers. The German page should not assume that Fachbegriffe alone will protect the category. Both pages should make the same business recognizable.
External profiles can then be compared against that source of truth. Not because the company site is automatically more trusted by AI systems. It is not. But because without a clear owned reference, cleanup turns into scattered polishing. One directory gets fixed. Another remains broad. A procurement portal still carries an old category. An English profile keeps using a former service name. The answer engine sees the mixture and behaves accordingly.
The entity footprint becomes stable when the public record repeats the same core facts in several places.
Repair starts with the profile that changed the role
All public profiles are not equal. Some are low-reach and harmless. Some are visible but vague. Some carry the exact phrase that AI systems repeat. Those deserve attention first.
I prioritize profile repair by citation reach, then by meaning, then by ease of editing. A high-reach association page that misstates the company category matters more than a small directory with a weak description. A procurement portal that assigns the wrong supplier type may matter even if it is hard to change. A low-value listing that is easy to edit can wait if it is not shaping answers.
The work is slow in the way useful repairs are slow. It means reading the AI answer, finding the source that carried the claim, checking whether that source supports the claim, and then deciding whether to edit, counterbalance, or leave it. Some sources cannot be changed. In that case the company needs stronger owned pages and better-aligned profiles elsewhere. AI search does not require a perfect public record. It does punish a public record where the easiest source says the wrong thing.
The most revealing question is simple: which public profile renamed the company first?