Being visible in the wrong comparison can look like progress from far away. Up close, it is a sideways seat: the business appears, but the answer has misunderstood the job it should win.
The school director brought three screenshots, all from comparison prompts about training options near Lyon. In one answer, her private vocational school appeared under broad “business schools.” In another, it was treated as a general alternance provider. In the third, it was missing from a list where its health-administration track should have been a strong fit. The odd detail: one campus was named, but the model described a program that belonged to another campus.
This is a composite scenario, based on several audits of multi-campus schools and specialist service firms. The pattern is not rare. A business enters AI answers, so the owner feels partly relieved, then the leads are wrong, the comparison is clumsy, and the rivals above it are not the rivals it actually fights. The ranking did not simply place the business too low. It pushed the business sideways into the wrong subcategory, where its proof becomes weaker than it is.
A wrong category can be worse than a low seat
A low seat in the right category is frustrating. At least the system has understood the contest. A wrong subcategory is more slippery. The business may appear in an answer, sometimes even with a decent position, while being judged against the wrong criteria. That makes the ranking look less broken than it is.
For the vocational-school composite, the useful comparison was not “best private schools near Lyon” in the broad sense. It was closer to “private work-study training for health administration roles near Lyon.” The school had real substance there: specialist programs, employer placements, campus structure, and student outcomes. But its public wording kept drifting upward into broader education language. “Professional training,” “career support,” “business programs,” “alternance opportunities.” Those phrases are true. They are also too wide to protect the seat that matters.
Wrong subcategory ranking is the condition where AI includes a business in a comparison but assigns it to a broader, adjacent, or misplaced use case, because the public trail does not repeat the specialization clearly enough. That is my working definition. It is different from simple absence. The entity is visible. The seat is wrong because the category frame is wrong.
I call the mechanism a sideways signal. A sideways signal is wording that is accurate in isolation and harmful in comparison. It does not lie. It points too broadly, too weakly, or toward a neighboring buyer need. Many businesses produce sideways signals because they are trying to sound complete. The irony is ugly: the more they widen their description, the less rankable their strongest specialty becomes.
Broad language feels safe until a machine must choose
Owners often prefer broad category language because it leaves room. A school offers several tracks, so it calls itself a training center. A clinic has several specialties, so it says it offers comprehensive care. A manufacturer has several capabilities, so it describes itself as an industrial partner. None of this is wrong. The trouble begins when an AI answer has to choose five names for a specific use case.
A broad label may get the business into a general pool, then weaken it when the ranking becomes more precise. The system needs to know which entity is strongest for which comparison. If your public trail says “professional school” and a rival’s trail says “work-study school for health administration assistants,” the rival owns the narrower seat. Even if your program is better, the answer can justify the rival more easily.
This matters especially in French markets where category names have local texture. Formation professionnelle, école privée, alternance, BTS, bachelor, reconversion, santé administrative, secrétaire médicale, campus, centre de formation: these terms do not travel evenly. In English prompts, the flattening can become worse. A specialist school can be translated into “business school,” then compared with broader institutions that have stronger authority signals for that broad label. The original strength is still there. It has been put in the wrong drawer.
A small roughness appears in many screenshots. The AI answer may name the right campus but attach the wrong course, or call a work-study program “online” because one directory used that word for a different track. These errors are annoying, but they are also diagnostic. They show that the public trail lacks enough clean separation between subcategories.
The first map separates entity, category and use case
When I draw this kind of case, I put three words across the top of the page: entity, category, use case. Entity is the business itself, with its locations and names. Category is the type of provider the system thinks it is. Use case is the buyer situation the answer is trying to solve. Most sideways ranking problems involve a mismatch among the three.
In the school example, the entity was reasonably visible. The category was unstable. The use case was underfed. The campuses repeated similar wording, so the system had trouble knowing whether one branch was stronger for health administration, another for commerce, another for a mixed alternance offer. When an answer could not separate the campuses, it sometimes chose one branch as a stand-in for the whole school. That is how internal cannibalization begins, though I will leave the full two-location problem for another note.
The category line on the map is where many edits should begin. I write the phrases exactly as they appear in public sources. “Private school.” “Training center.” “Alternance.” “Health careers.” “Administrative assistant.” “Campus Lyon.” “Placement support.” Then I mark which phrases are inclusion signals and which are ordering signals. A broad category may help inclusion. A narrow use-case phrase helps ordering when the prompt is narrow.
This is where owners sometimes resist. They worry that naming the narrower category will make the business look smaller. In a sales brochure, perhaps. In an AI comparison answer, specificity often gives the system permission to rank you for the right seat. A broad sentence says, “We might fit.” A specific public trail says, “We fit this buyer, in this city, for this outcome.”
A French business can lose the correct comparison while winning a vague one. That is not visibility success. It is category drift.
Sideways signals usually come from five places
I try not to turn articles into lists, but this mechanism has a few recurring sources that are worth naming in prose. The first is homepage diplomacy. The homepage tries to include every service, every audience, every location, and every emotional promise. It becomes polite fog. An AI answer can include the business, but it cannot see which subcategory deserves priority.
The second is branch repetition. Multi-location businesses often copy one campus page and swap the city name. For a human, the page still reads as serviceable. For a ranking system, every branch becomes a blurred duplicate. If the Lyon campus has the health-administration strength, and the Grenoble campus has a different track, those roles need distinct public sentences. Otherwise the system may treat the branches as interchangeable evidence, then choose unpredictably.
The third is directory inheritance. Old listings keep a broader label after the business has specialized. A school that once promoted general office training may now have strong health-administration tracks, but the public shelf still says “secretarial training” or “business support courses.” The old label pulls the answer sideways. I have seen similar cases with agencies that moved from broad communication work into a narrow technical specialty, yet still carried old directory categories like barnacles on the hull.
The fourth is translation flattening. A French phrase with vocational specificity becomes an English phrase with generic education meaning. The owner never wrote the English version, but a profile, platform, or model paraphrase creates one. From that point, English prompts may rank the business against the wrong set. A specialist French signal becomes a general English category.
The fifth is proof without a named use case. Customer comments may praise the school’s support, teachers, or atmosphere. Useful, yes. But if the comments do not connect those qualities to the program the school should win, they remain soft ordering signals. “Good support” is less rankable than “support during work-study placement in health administration.” The second phrase gives the answer something to compare.
The repair is a narrower public trail, not a smaller business
A business does not need to become narrow in reality. It needs to make its ranking trails distinct. That distinction can happen at the level of service pages, campus pages, third-party profiles, local partner wording, alumni proof, and English summaries. The goal is not to shout specialization everywhere. It is to give each important comparison a stable trail.
For the school composite, I would ask for one public sentence that joins the exact program, the work-study mechanism, the local area, and the buyer outcome. It might sound plain. Plain is useful. “A private work-study school near Lyon for students preparing health administration and medical office roles.” That sentence is not poetry, thank God. It is a stake in the ground.
Then the same meaning must appear, with variation, in the places AI can retrieve. A campus page can explain which programs belong to that location. A partner employer page can name the role students prepare for. A directory can carry the right category. A short English profile can avoid the lazy “business school” translation. A student story can connect support to the narrow path rather than to education in general.
The hardest discipline is subtraction. Some pages need fewer broad promises. If every paragraph says the school is flexible, professional, human, ambitious, close to employers, and good for many careers, none of those signals holds the seat. The page may be nice. The ranking answer needs a sharper edge. I often cross out fine sentences because they are fine in the wrong way.
Owners sometimes ask whether this will harm broader rankings. In my observation, a cleaner narrow trail usually helps more than it harms, as long as the wider entity remains understandable. The business can still say what it offers overall. But the ranking path for the specialist comparison needs its own public markers. A map with one road to every town is not a map. It is a gray page.
Re-check the category before chasing the rank
Before any re-check, I want to see whether the AI answer has changed category frame, not only whether the business moved up. A move from seventh to fourth in the wrong category may still be bad work. A move from broad private-school comparison into a health-administration work-study comparison can matter more, even if the first seat is not won yet.
This is why the screenshot alone is insufficient. I ask for the prompt, the language, the named rivals, the described criteria, and the sentence used for the business. A model may include the school and describe it wrongly. Another may exclude it from the narrow comparison but include it in a broad one. A third may rank the wrong campus. These are different failures. They should not receive the same edit.
The right question is severe: what comparison should this business deserve to win? Not every comparison is worth fighting. A small specialist firm does not need to beat broad competitors in every broad answer. It needs to stop being dragged into contests where its strongest proof is invisible. The wrong subcategory steals attention because it looks close enough to be acceptable. It is not close enough if buyers arrive with the wrong expectation.
In the notebook, a sideways signal gets a little arrow pointing out of the column. It means the evidence is escaping the intended comparison. The repair is not glamour. It is category grammar: entity, subcategory, use case, place, proof, and language. Once those pieces hold together in public, the ranking answer has a better reason to seat the business where it actually belongs.
The Last Seat Note: Seat held: visible, but sideways. Rival pressure: broader schools own the generic category while the specialist program lacks a clean public lane. Weak signal: campus and program wording blur together, especially across French and English descriptions. Sentence to plant in the public trail: “A private work-study school near Lyon for students preparing health administration and medical office roles through employer-linked training.”