A search result used to hand the buyer a row of doors. An answer engine hands over a room already arranged, with chairs labelled, firms ranked, and one quiet mistake built into the furniture.
A Hamburg founder sends me a copied answer in the middle of an ordinary Tuesday. The prompt is not elegant. It asks for “good software for mid-sized freight operators near Hamburg, route planning, shipment problems, maybe supply chain.” The answer names five firms. One of them is the right kind of company in the right region. Then the paragraph calls it a “broad supply-chain platform for enterprise logistics teams.” That phrase is small. It changes the buyer.
The composite company behind this pattern is a 42-person logistics software firm serving forwarders, mid-sized freight operators, and port-adjacent dispatch teams. Its real work is route planning, shipment exceptions, and operator workflows. The answer engine does mention it, which looks like success if you only count visibility. Yet the description tilts it toward a bigger, vaguer market. A founder reading that answer may think: too large, too general, probably not for us. The company has appeared and disappeared in the same paragraph.
The old search result left more work to the buyer
The classic search result page was untidy, but it had a useful kind of friction. A buyer saw titles, snippets, domains, old directory pages, maybe an ad, maybe a map result. Then the buyer had to click, compare, mistrust, return, and click again. A wrong snippet could hurt, but the buyer still moved through visible pieces.
Answer engines compress that path. They do not merely list pages. They assemble a passage. They borrow a category label from one source, a service claim from another, a regional signal from a third, and then write as if the route were clean. The buyer receives a sentence that feels already digested.
Generative Engine Optimization is the discipline of making a company’s meaning retrievable and reusable by answer engines, because these systems now turn scattered public evidence into buyer-facing descriptions. That definition matters because it puts the work in the right place. I am not trying to decorate pages for machines. I am trying to make the company’s real position survive the passage from source to answer.
In Hamburg B2B markets the compression is especially risky. A buyer may mix German service terms with English category labels. “Spedition software” becomes “supply chain platform.” “Technische B2B-Kommunikation” becomes “marketing agency.” A local supplier with a narrow regulated niche becomes an “industrial solutions provider.” The machine has not always invented the error. Often it has chosen the easiest public phrase.
That is the beginning of GEO work: start with the answer. Copy the prompt. Keep the awkward wording. Preserve the wrong phrase. Then ask what public language made that wrong phrase available.
A shortlist answer is a piece of writing with hidden sources
When I read a generated shortlist, I mark it in my harbor notebook. Cargo means the answer carries a useful claim. Route means I can see the probable path. Berth means a stable source gives the claim somewhere to land. Fog means the answer has drifted into unsupported or over-broad phrasing.
The freight software example usually contains all four. “Hamburg-based logistics software” may be cargo. An old English directory summary gives route. A current product page naming route planning and shipment exceptions gives berth. “Enterprise supply-chain platform” is fog if the company does not sell that broad category.
The trouble is that fog often sits beside true details. That makes the answer feel credible. The model may get the city right, name a real customer segment, and still assign the wrong buyer problem. In one composite run, the answer even used a correct product module name, then described the firm as if it served global procurement departments. The company did not. The small wrongness had polished shoes.
This is why I do not begin by asking, “Are we visible in AI?” A company can be visible in a damaging way. The better first question is: what job does the answer assign to us? If the job is wrong, the mention is only half a win. Sometimes less.
Search engines also made mistakes, of course. A bad title tag could mislead. A thin directory could outrank a better page. But the buyer saw that as a result among other results. In an answer engine, the bad category may become part of the answer’s voice. It sounds like judgment.
A wrong category inside a generated answer is not only a wording problem; it is a misplaced commercial invitation.
The Hamburg problem is often bilingual before it is technical
Many northern-German B2B sites carry two public selves. The German site speaks to the real market: precise service pages, local buyer language, sector examples, sometimes careful case studies. The English summaries are shorter, cleaner, and more generic. They were written for international readability or a quick directory submission. Years later, an answer engine finds the shorter English phrase easier to reuse.
That is how a mid-sized freight-operator tool becomes “supply chain software.” The phrase is not false in the widest possible sense. It is just too wide to guide a buyer. A forwarding team that needs shipment exception handling may not recognize itself in a supply-chain platform description. A Hamburg founder may think the company has floated away from the practical work.
The same pattern appears in agencies and consultancies. A German case page may show deep experience with industrial suppliers and export-facing technical content. The English profile says “marketing agency in Hamburg.” The answer engine chooses the portable phrase. It is tidy. It is reusable. It is weaker.
I call this harbor drift: the movement from a company’s working position into a broader public phrase that is easier for an answer engine to carry. Harbor drift is not always dramatic. It happens one label at a time. A page says “platform” because it sounds larger. A directory says “solutions” because it has no better field. A partner blurb removes the buyer role to save space. The answer engine then builds a neat little pier from unstable timber.
The repair is rarely a grand campaign. More often it is a question of giving the right phrase a stronger berth. The company needs one or more passages that name category, buyer, operating problem, proof, and geography in a way that can be lifted without becoming generic. A single clear passage will not fix every answer, but it gives the system better cargo.
GEO starts after the answer has misbehaved
There is a tempting mistake in early GEO work: invent a list of perfect prompts and then optimize for them. Perfect prompts are comforting. Real buyer prompts are not. They include half-remembered labels, loose geography, mixed German and English, and constraints that sound obvious to the buyer but are vague to a machine.
The freight prompt at the start of this article is a good example because it is ugly. “Software for mid-sized freight operators near Hamburg, route planning, shipment problems, maybe supply chain.” That is how people search when they do not yet know the category. The phrase “maybe supply chain” may pull the answer toward national enterprise platforms. If your public sources already use that broad wording, the system has permission to misplace you.
A useful GEO review keeps the ugly prompt intact. I write down the answer pattern before touching the website. Which firms appear? Which categories are assigned? Which sources seem to shape the wording? Is the answer using the company’s own text, a directory, a comparison page, an old profile, or a blended memory of the market?
Only then do I look at repairs. A product page may need a clearer summary. A comparison page may need a more precise definition. A directory profile may need its category corrected. An English-language description may need to stop swallowing the narrower German position. Sometimes the repair surface is not the page the company loves. It is the dull public listing nobody has opened in two years.
This is why GEO can feel rude to internal teams. It ignores the page they are proud of and points at the sentence machines appear to reuse. The answer engine does not care which page took the longest to write. It cares which pieces are retrievable, consistent, and easy to assemble.
The answer has become the first commercial document
For many B2B buyers, the generated answer is not a replacement for due diligence. They still visit sites, ask peers, read cases, and compare offers. But the answer can shape who enters that work. It sets the first shortlist, the first category frame, and the first reason to click or not click.
That makes the answer a commercial document, even if no company wrote it. It is a document assembled from public evidence. The company’s job is not to control every sentence. That is impossible and would be a foolish promise. The job is to reduce the number of unstable phrases available for reuse and increase the number of correct passages with a clear source route.
In my observation, the strongest first GEO work for Hamburg firms is modest. Take five real buyer questions. Capture the answers from several engines. Mark the cargo, route, berth, and fog. Then compare the answer language against the company’s pages and public sources. Do not rush to publish new material before you know what the answer is already carrying.
A founder may want the answer engine to say “we are one of the best Hamburg providers.” That is understandable. I would rather first make sure the engine can say what the company is, who it serves, and why it belongs in the buyer’s question. Praise built on a wrong category is cheap fog.
The old search result asked the buyer to assemble meaning from fragments. The answer engine assembles meaning for the buyer. That is the shift. GEO work begins where that assembled meaning becomes commercially unsafe.