A best-in-city AI answer is rarely built from one beautiful service page. It is more often stitched from dull public traces: a city name, a category label, review wording, and one sentence that sounds safe to repeat.
A composite service firm in North Rhine-Westphalia once gave me a familiar puzzle. Forty-two people, serious work, good clients, not much theatre. They handled compliance documentation and supplier onboarding for mid-sized industrial firms. Their human buyers understood them well enough after one sales call. AI search did not. In one answer they were a local consultant. In another, a software platform. In a third, almost an office administration provider with a slightly technical flavour. The model even named the right city once, then placed their work in the wrong state in the next paragraph.
That is the kind of mess people bring to me when they ask about “best service providers in Berlin” or “beste Dienstleister in Berlin” style queries. They usually expect a ranking problem. Why are the wrong firms in the answer? Why does a weaker competitor show up? Why does a company with fewer reviews look more authoritative? But the answer is usually less dramatic. The engine is not judging the business as a buyer would. It is assembling a recommendation from public evidence that has survived its own strange little test: clear category, clear geography, clear proof, and enough third-party support to make the claim feel safe.
A best-in-city answer starts with a category, not a city
When someone searches for a local service provider, the city looks like the center of the query. Berlin, Hamburg, Köln, München. That is the visible hook. But in AI answers, the category often does more work than the place name. The engine must first decide what kind of provider belongs in the answer. Only after that does geography narrow the field.
For a German service firm, this is where the first distortion often appears. A company may describe itself as a “partner for compliance processes,” a “service provider for supplier documentation,” a “consulting and operations team,” and an “onboarding support office” across different pages. Each phrase may make sense in a human sales context. Together they look like a drawer full of unmatched keys.
A directory, by contrast, may use one blunt category: “Unternehmensberatung.” Another profile may call the same firm “Büroservice.” An English page may say “vendor management platform,” even if the company sells mostly managed service work, not software. When an AI engine builds a best-in-city answer, the blunt category can win because it is easier to sort.
I call this the city-category hinge. The city-category hinge is the public phrase that lets an AI engine connect a business to a local recommendation query, because it names both the service class and the market area in a repeatable way. If the hinge is wrong, the answer may still include the business. It just includes it for the wrong reason.
This is why a firm can be visible and misrepresented at the same time. The engine found the business. It found Berlin, or Düsseldorf, or Nordrhein-Westfalen. It found a service category. The failure is in the join.
Reviews give texture, but they rarely repair the role
Reviews matter in local recommendation answers, but they do not usually solve category confusion by themselves. In my observation, review text adds surface confidence. It gives the answer phrases like “responsive,” “professional,” “reliable,” “easy to work with,” or “fast.” Those are useful, but they do not explain what the business actually does.
A service firm may have dozens of positive reviews that praise the team for being quick and competent. Fine. But if the reviews say “helped us with documents” and the website says “digital compliance operations,” an engine has to decide whether the firm is a document service, a compliance consultant, a software provider, or an outsourcing partner. The review wording can even pull the answer toward the simplest role.
This is not a complaint about reviews. They are part of the evidence record. They show that the firm exists in the world and that clients have interacted with it. But reviews are often written after stress has passed. The client does not describe the service architecture. They write the sentence that felt true at 19:30 after a deadline: “They helped us get the paperwork done.”
In best-in-city answers, that kind of sentence travels easily.
The useful repair is not to stuff reviews with technical phrases. That would look false and, in many cases, be false. The repair belongs on the firm’s own pages and profiles. The service page has to carry the role clearly enough that review language cannot pull the answer into the wrong corner. If a company handles supplier onboarding for industrial firms, that phrase should appear before softer claims about support, efficiency, or partnership.
A machine can quote praise, but it classifies from nouns.
City pages can help, but weak city pages can also flatten the firm
Many German service companies create city pages because search agencies tell them to. Sometimes these pages are useful. Often they are thin: a city name in the heading, a generic paragraph, a list of services, and a contact block. The page says “Dienstleister in Berlin” or “Beratung in Hamburg,” but it does not show why the firm belongs in that city answer.
For AI search, that thinness is dangerous in a quiet way. A city page may make the company eligible for a location query while also stripping away the evidence that made the company distinctive. The engine sees a clean local service phrase and repeats it. The page has done its job badly: it got the firm into the room wearing someone else’s name badge.
A better city page does not have to be long. It needs to connect the location to a real operating reason. Does the firm serve clients in that city because of branch structure, sector concentration, regional regulation, procurement networks, field work, or language coverage? Does it have clients there, staff there, delivery capacity there, or just a willingness to take calls from there?
In a composite pattern I see often, a regional service firm writes one page for every large German city. The Berlin page, Munich page, and Hamburg page differ only by place name. A model can read those pages, but it learns almost nothing from them except that the firm wants to be considered everywhere. The imperfect detail is usually small: one page still mentions the wrong city in the middle paragraph because the template was copied too quickly. Machines notice less than humans in some places and more in others.
For best-in-city answers, a city page should answer one grounded question: why would this provider be a plausible answer for this city, rather than merely a provider that accepts work there?
Directories are dull, and that is why they carry weight
People dislike directories because many of them are ugly, old, and commercially noisy. AI engines do not dislike them in the same way. A directory profile has a category, address, phone number, URL, service label, sometimes reviews, and sometimes a short description. It is boring in a structured way. That makes it easy to use.
This is where German service businesses often lose control of their own description. One directory says “Unternehmensberatung.” Another says “Software.” A procurement portal says “administrative services.” An association profile uses an old brand name. A local press article describes the founder’s background and forgets the current service line. None of these sources is malicious. They are just small public facts left to harden.
When an answer engine builds a list of service firms, it may prefer the public source that gives it the least trouble. A directory category can be easier to repeat than a careful but abstract homepage. The result is a best-in-city answer that sounds tidy and wrong.
I usually inspect directories with less contempt than clients expect. Not because directories are wise. Because they are often the first public place where entity, category, and location are tied together. If that tie is inaccurate, the model may inherit the error with a straight face.
The repair is practical. Align the company name. Remove dead former names where possible, or explain the name change where removal is impossible. Make the category narrower where the platform allows it. Rewrite the short description so the first sentence names the real role. On German profiles, use German category language that matches the service page. On English profiles, avoid turning a managed service into a platform just because “platform” sounds larger.
One wrong directory will not ruin a firm. Five slightly wrong profiles may teach an engine a stable misunderstanding.
Third-party mentions need a sentence that can survive being lifted
Local press, association pages, partner references, awards listings, procurement mentions, and event pages often appear in the evidence trail behind local AI answers. They are valuable because they are not the company speaking about itself. They are also risky because they may describe the business from the outsider’s narrow angle.
A local article may say a compliance documentation firm “helps companies with paperwork.” That is not false. It may be too small. An industry association may call the same firm a “digitalization partner.” That may also be true in a loose way. It may be too broad. AI answers are not good at recovering the missing middle from scattered hints.
When I review third-party mentions, I look for liftable sentences. A liftable sentence is a sentence that can be quoted or paraphrased outside its original page without changing the company’s role. “The company supports mid-sized industrial firms with supplier onboarding and compliance documentation” is liftable. “The company helps clients simplify their processes” is mush once removed from context.
This matters because AI search often treats third-party wording as a stabilizer. If several outside sources use the same category, the model becomes more comfortable repeating it. If outside sources diverge, the answer may choose the loudest or simplest one.
German firms sometimes invest heavily in their own site while leaving partner descriptions untouched for years. That is understandable. Nobody wants to chase every profile. But for best-in-city answers, the most visible third-party descriptions should not contradict the company’s own role. A small correction on a partner page can be more useful than another polished paragraph on the homepage.
The city answer is a public chorus. One singer out of tune is tolerable. When the wrong section leads, the whole tune bends.
Reporting the repair without pretending it is a ranking formula
The difficult part with best-in-city AI answers is that they look like rankings, but the mechanism is not a clean ranking formula. The answer may be influenced by source availability, local wording, directory structure, review language, entity confidence, and the prompt’s phrasing. A company can improve its evidence and still not appear every time. Another firm can appear for a bad reason. This is irritating, but it is better to say it plainly.
When I report this work, I do not promise that a firm will become “number one” in an AI answer. That is the wrong frame. I show which public sources currently make the firm eligible, which sources distort the category, which city claims are supported, and where the evidence is too thin. Then I watch repeated queries across engines and languages.
For a best-in-city query, the useful measures are not only presence or absence. I want to know whether the business is named with the correct role, whether the geography is right, whether the cited source supports the claim, and whether competitors are being selected because they have clearer evidence rather than better fit.
The composite North Rhine-Westphalia firm did not need a louder page. It needed a narrower public record. The German service page had to name the managed service plainly. The English page had to stop sounding like software. Two directory profiles needed category cleanup. A partner page needed one sentence changed. The local city page needed a real reason for the location claim. None of that is glamorous work. It is closer to repairing labels on boxes in a warehouse where the lights flicker.
Best-in-city answers reward firms whose public evidence is easy to assemble without too much guessing. That is uncomfortable for service companies, because good service work is often relational and specific. But the machine does not attend the sales call. It reads the residue.