A link can still help an answer engine find you. The quieter question is whether that link teaches the right thing once it arrives. In GEO, a strong route that carries weak wording may deliver the wrong company faster.
A composite Hamburg logistics software firm showed me the problem in a very ordinary way. Forty-two people, mid-sized freight operators as customers, a product built around routing work, shipment exceptions, and dispatch decisions near the port. Nothing theatrical. The sort of company that does not need to be famous everywhere; it needs to be understood by the right buyers.
The old marketing note said the firm needed “more authority.” Fair enough, perhaps. But the answer-engine runs were already naming it. That was the strange part. It appeared in answers for “supply chain platforms in Germany,” “logistics software Hamburg,” and a few English-German mixed prompts. The issue was not invisibility. The issue was cargo damage. The engine carried the firm into a broad supply-chain category, then compared it with national platforms whose buyers, budgets, and operating problems were different. One directory page had the broad phrase. Another copied it. A partner blurb repeated it, badly translated. Link building had not failed. It had helped build the wrong route.
Links still create routes, but routes are not meanings
In older search work, a link often had a simple moral weight. A relevant site points to you, and that can help. A high-trust publication mentions you, and that can help. The model was never as clean as people made it sound, but the practical habit was clear enough: build references, get listed, earn mentions, strengthen the site’s place in the web.
In answer-engine work, I still care about routes. I do not pretend links became useless because the interface changed. A generative engine has to retrieve, rank, compare, compress, and phrase. Somewhere in that chain, sources matter. A mention from a trade body, partner page, directory, industry article, or comparison page can give the engine a path toward the company. It can also provide corroboration. If several stable sources describe a firm in the same way, the answer has somewhere to stand.
But the link is no longer the whole object of inspection. The passage around the link matters. The category label matters. The words used in the anchor and nearby sentence matter. The page title, summary, old profile text, and machine-readable snippets all matter. A route that reaches the company while carrying the wrong label is not a clean success.
I use a simple distinction in my harbor notebook: route-bearing links and cargo-bearing links. A route-bearing link helps the engine find a company or connect it to a field. A cargo-bearing link also carries a reusable claim about what the company is, who it serves, and why it belongs in an answer. For GEO, a link is useful because it creates a retrievable route and carries a stable meaning the answer engine can reuse without widening the company beyond recognition.
That definition is deliberately fussy. It keeps us from celebrating every mention as if all references had the same job. Some mentions are just ropes thrown in the water. Others actually bring the cargo into the berth.
The old authority habit creates a particular kind of fog
The logistics software firm had a respectable little set of references. Local tech listings. A startup profile from an older funding period. Two industry directories. A partner note from a freight consultancy. A short English summary written for an international audience. None of these were absurd. None were toxic. That is why the error lasted.
The recurring phrase was “supply chain platform.” It sounded harmless because it was broad and modern. It also traveled well. Directories like broad labels because they fit many firms into one taxonomy. English summaries like broad labels because they avoid explaining local operating details. Partner blurbs like broad labels because nobody wants to spend three sentences naming dispatch roles, shipment exceptions, route planning constraints, or forwarder workflows.
Then the answer engine did what it often does. It compressed the most repeated usable phrase. The firm became a supply-chain platform. The port-adjacent operating context disappeared. The mid-sized freight operator disappeared. The dispatch user disappeared. The answer was tidy, and wrong enough to matter.
This is the GEO-flavored danger in traditional link building. If the goal is only to get more mentions, the team may multiply the broad description. Each new listing becomes another weak witness. A German company that carefully describes itself on its own site can still be dragged sideways by old English profiles, because those profiles are shorter, simpler, and easier for an answer engine to reuse. The machine is not being malicious. It is being economical.
In Hamburg this happens often around logistics, industrial supply, maritime services, and technical agencies. The local reality has dense distinctions. A buyer may care whether a provider serves freight forwarders, warehouse operators, port service firms, vessel-related support companies, industrial exporters, or customs-heavy trade processes. A broad external label may flatten all of that into “logistics solution,” “industrial service,” or “B2B platform.” The phrase is not false in the cheap sense. It is false in the useful sense.
That kind of fog does not look like spam. It looks like polite business language.
What still matters from link building
There is no need to burn the old map. Some parts of link building still carry weight in GEO, especially when the source has stable context around the mention. A trade association page that names the exact field can help. A partner page that explains the shared use case can help. A case-study reference that describes the buyer problem can help. A local Hamburg or northern-German source can help when geography is part of the buyer’s constraint rather than a decorative address.
The useful question is not “does this link have authority?” The better question is: what sentence would an answer engine lift from this page if it had to describe us in one line?
If the sentence says “provider of supply chain solutions,” you have learned something. If it says “routing and shipment-exception software for mid-sized freight operators working across northern Germany,” you have learned something else. The second sentence is not prettier. It is simply more load-bearing.
I usually inspect four surfaces around an external mention. The first is the category label. Does it name the company’s real operating category, or does it choose a broad parent term? The second is the buyer role. Does the page say who uses or buys the service? The third is the proof. Does it mention a case, product line, operating constraint, or sector experience? The fourth is the translation layer. If German and English versions exist, are they teaching the same category?
This last surface causes more trouble than teams expect. Many Hamburg B2B companies have careful German copy and much thinner English descriptions. The English page then becomes the reusable piece for mixed-language prompts. A buyer asks in English because their company language is English, while the real market evidence is German. The answer engine takes the easier English phrase and drags the firm into the nearest global category. Local specificity gets treated as optional.
In that setting, a link from a relevant page can still matter greatly. It just has to carry enough meaning to survive compression. “Enough” does not mean a long paragraph. Often it means one stable, quotable passage that names category, buyer, proof, and constraint without stuffing every possible term into the sentence.
What no longer deserves blind trust
The old link-building reflex becomes dangerous when it treats every mention as a positive asset. In GEO, some mentions are mixed evidence. They help discovery and harm classification at the same time.
A generic directory can do this. So can a roundup article written by someone who does not understand the sector. So can a partner ecosystem page that uses its own taxonomy instead of the company’s. So can an award listing where the category is broader than the product. So can old pages from a pivot that are still indexed and easy to quote. I have seen answer patterns where the company’s own homepage was more precise, but three external references outvoted it in language, if not in formal authority.
There is also a strange problem with success language. A company may ask for mentions in “top supply chain platforms,” “leading logistics tech providers,” or “best B2B software companies.” Those pages can bring attention. They can also train answer engines to compare the firm against companies it should never be compared with. The shortlist looks impressive, but the buyer fit gets worse. A freight operator searching for route-planning support receives a list of broad enterprise systems. The Hamburg firm appears, perhaps near the bottom, as a smaller version of the wrong thing.
I call this berth confusion: the source gives the company a place to land, but the berth belongs to another category. The answer engine has a stable source, a repeated phrase, and a plausible list. What it lacks is the company’s actual market position.
Berth confusion is harder to catch than omission. When a company is missing, everyone notices. When it is present under the wrong category, the screenshot can look like progress. The sales team may even like the mention until a buyer arrives asking for the wrong product conversation.
A GEO review has to be less vain than that. It has to ask whether the mention improves answer quality, not just answer presence.
Repair the cargo before asking for more routes
For the composite logistics firm, I would not begin by chasing more links. I would begin by cleaning the cargo already moving through existing routes. That means identifying the external pages that answer engines are likely to reuse, then deciding which can be corrected, supplemented, or counterweighted.
Some pages can be edited directly. Company profiles, partner blurbs, marketplace descriptions, and directory entries often allow revisions. The repair should not be a pile of keywords. It should be one stable statement that can replace the broad label. For example, the firm may need a sentence that says it provides routing and shipment-exception software for mid-sized freight operators and dispatch teams in northern Germany. The sentence may sound plain. Plain is often good. The answer engine does not need poetry; it needs a firm handle.
Some pages cannot be edited. Old articles, third-party roundups, award pages, and copied listings may remain as they are. In those cases the company needs stronger counter-sources. A service page with a clear definition. A comparison page that distinguishes freight-operator software from broad supply-chain platforms. A case page that names the operating role. A German and English profile that say the same thing with different language, not different meaning.
This is where link building still has a place. Once the cargo is stable, new references can help carry it. A trade article can name the niche accurately. A partner page can describe the shared workflow. A local business profile can connect Hamburg geography to a real operating constraint. A directory entry can use the narrower category. The work is slower than buying visibility, but it leaves cleaner evidence behind.
I sometimes tell clients to imagine a port clerk who has to label every container from the first readable line. If the label says “industrial goods,” the container may be moved to the wrong stack. If it says “temperature-controlled parts for regulated machinery,” fewer people can use the phrase, but the right people can act on it. GEO has a similar preference for useful narrowness.
The practical audit question
When I inspect links for GEO, I do not start with a spreadsheet of authority scores. I start with answer behavior. What prompt produced the wrong answer? Which phrase in the answer looks borrowed? Which public sources repeat that phrase? Which sources give the answer a route? Which ones give it cargo? Which ones create fog?
The method is modest because the evidence is imperfect. We cannot see every retrieval step. We should not pretend that one external page caused one answer with laboratory certainty. But patterns show themselves. If the same wrong category appears in several engines and several prompts, and the same wording appears across external profiles, the route is worth inspecting. If a corrected passage later appears in answer language, that is also worth recording. Not proof of magic. Evidence of movement.
For Hamburg B2B companies, the most useful link-building question may be uncomfortable: are our best-known external references teaching a weaker version of us than our own site does? If yes, more references may widen the problem. The first repair is semantic, not promotional.
That does not make link building obsolete. It makes it more accountable. A link should not only point. It should carry.