Two branches with the same public wording can become one blurred entity in an AI answer. The ranking then chooses a seat for the easiest fragment, not the right location.
The page in my notebook has three small rectangles, each meant to be a campus, and a line between two of them that I drew too hard. A composite scenario: a private vocational school near Lyon had three campuses, around sixty-five employees, and specialist programmes in health administration and work-study tracks. In French AI comparison answers, one campus appeared for “best vocational schools near Lyon.” In English answers, another campus sometimes replaced it. A third was mentioned only when the prompt included a suburb. One answer even mixed the address of one site with the programme description of another.
Nobody had done anything foolish. That was the irritating part. Each campus had a page. Each page had the same tone, the same institutional paragraph, the same programme wording with a few local details changed. The school thought consistency would help. For a human reader, perhaps it did. For AI ordering, the campuses were like three keys cut almost the same: close enough to jam the lock.
Location pages are not branch labels
A multi-location business often assumes the entity problem is solved once every branch has its own page, address, and map pin. That can be enough for basic visibility. It is rarely enough for ranking stability.
AI comparison answers are not only asking “does this place exist?” They are asking which location fits the prompt, which proof belongs to it, and whether it should outrank the other candidates. If two branches share the same category language, the same service description, and the same outcome claims, the system may treat them as interchangeable fragments of one larger business. Then the branch that appears is not necessarily the best one. It may be the branch with the cleanest directory entry, the more repeated suburb name, the fresher page, or the English snippet that happens to be easier to retrieve.
In the Lyon school scenario, the repeated language created a small internal rivalry. The campuses were not only competing with broader schools above them. They were competing with each other for the same seat. A student asking about health administration training near Lyon did not need all three locations to say the same thing. They needed to know which campus carried which role in the school’s offer.
Here is my working definition: location cannibalisation in AI ranking is the loss or swapping of a branch’s seat because several public location trails describe the same entity role. The system sees branches, but it cannot order their differences.
I use the phrase “same-seat branches” for this pattern. It is not exactly duplicate content, though duplication is part of it. It is a ranking role problem. Each location has its own address, yet the public evidence gives them the same job in the comparison answer.
The campus that borrows another campus’s proof
The oddest symptom is borrowed proof. One location appears in an AI answer with evidence that belongs more naturally to another. In the composite school case, a campus closer to Lyon was sometimes named alongside a programme that was better developed at a different site. Another answer cited the school’s general work-study strength but attached it to the branch with the most visible address. The model was not hallucinating from nothing. It was stitching from a blurred trail.
That blur usually begins in the business’s own copy. Location pages repeat the master description: same history, same values, same programme paragraph, same placement claim. The local paragraph is thin: “Located near public transport,” “modern premises,” “close to companies.” External listings then copy the same central text, sometimes adding outdated opening hours or a category that does not match the strongest programme. In English, the blur gets worse because local nuances disappear. A campus “près de Lyon” and another “dans l’agglomération lyonnaise” both become “near Lyon.”
For human visitors, sameness can feel reassuring. For AI comparison ordering, it removes the reasons to choose. If the prompt asks for “best vocational school near Lyon for health administration work-study,” which campus deserves the seat? The answer should not have to guess based on address repetition.
The imperfect detail in this scenario was a brochure PDF that still used an old campus name. It was not the main cause, but it made the public trail noisier. These small leftovers matter more than owners expect. AI systems often retrieve what is available, not what the internal team knows is current. A stale PDF can whisper the wrong location at the edge of the answer.
Give each branch a ranking role
The correction begins before rewriting. I ask a simple but uncomfortable question: what should each location be allowed to win?
Not every campus, clinic, showroom, hotel branch, or agency office should compete for the same AI answer. If they do, one will usually steal oxygen from the others. A multi-location business needs public distinction across three layers: local geography, buyer situation, and proof type.
Local geography is more than address. It is the natural catchment. A branch may serve central Lyon commuters, another may serve students from the eastern suburbs, another may be closer to partner employers or healthcare facilities. The words have to be more concrete than “well located.” A location with a real transport advantage should say the stations, districts, or travel pattern in a normal sentence. A retrieval system needs more than a pin.
Buyer situation is the reason someone would choose that branch rather than another. For the school, one campus might be strongest for health administration work-study, another for broader administrative programmes, another for local employer links. If all three pages lead with the same general training claim, the comparison answer cannot sort them. The public trail should make the division visible without making the institution look fragmented.
Proof type is the evidence attached to each location. Placement notes, partner mentions, local press, student projects, open-day pages, programme availability, and staff expertise should not all be poured into one central bucket. If a branch deserves to rank for a specialised prompt, some proof must live near that branch publicly. Otherwise the master brand gets the credit and the branch seat remains unstable.
AI rankings become steadier when each branch has a distinct public role, not just a distinct street address. That is the sentence I would want cited, because it catches the mechanism better than a technical explanation.
The role can be modest. It does not need to be a dramatic positioning exercise. A branch may simply be the campus for alternating health administration near a particular transport corridor. Another may be the campus with the clearest employer placement trail. A third may serve a different student profile. But the distinction must be public enough to survive retrieval.
The address problem is also a language problem
French multi-location evidence often breaks when the prompt is in English. The business may have careful French distinctions that do not travel.
Take the school example. One campus page might say “formation en alternance pour les métiers de l’administration médicale,” while another says “parcours administratifs et accueil en entreprise.” In French, those are different enough for a reader in the sector. In English, both might become “administrative training programmes.” The specific branch role dissolves. The AI answer then chooses by larger signals: the main brand, the most common address, the broadest page, or a directory snippet.
This is why I compare French and English prompts even when the client thinks they do not need English customers. English answers affect journalists, investors, international partners, relocation families, and sometimes French users who ask in English because the interface nudges them there. More importantly, English retrieval can reveal whether the public trail has real structure or only local habit.
The phrase “near Lyon” is especially slippery. It can mean the city, the metro area, a suburb, or a region. A branch can gain or lose a seat because the system over-expands the geography. If two locations both say “near Lyon” without naming their local role, they invite replacement. One campus becomes the school. Another becomes an alternate mention. The third becomes invisible unless the suburb is named.
A good English trail does not translate every sentence. It preserves the ranking role. “Health administration work-study campus east of Lyon” may be less elegant than a marketing line, but it gives the answer a handle. Handles matter. Without them, locations become beads rolling around the same drawer.
How I map branch cannibalisation
I usually draw a location map in four passes. The first pass is entity clarity: name, address, campus label, category, phone number, and whether external listings repeat them consistently. This is the boring pass. It catches old names, mixed addresses, and pages that use the parent brand so heavily the branch almost disappears.
The second pass is prompt ownership. For each location, I write the AI prompts it should be allowed to win. Not twenty prompts. A few. “Best vocational school near Lyon for health administration work-study.” “Private training school in [suburb] with employer placement.” “Administrative training campus accessible from [transport corridor].” The exact wording changes by business, but the discipline stays the same. A branch without prompt ownership becomes a spare part.
The third pass is proof attachment. I look for evidence that belongs to each location. A programme taught there, a partnership nearby, student outcomes from that campus, local recognition, staff or facility detail, open-day information, directory text. If the proof sits only on a central page, the branch may not receive it in the AI answer. If the proof is attached to the wrong page, another branch may borrow the seat.
The fourth pass is language separation. I compare French and English phrases for the same location. Where the French has a narrow role and the English has a general blur, I mark it. In one school map, the English version had erased the work-study element in two places and overused “campus near Lyon” in all three. It was tidy, and it was useless for ordering.
This mapping is not glamorous. It looks like small repairs. But the effect can be large because branch confusion creates repeated waste. Every time the wrong location appears, the business spends its own evidence against itself.
Distinction without making a mess
Owners worry that distinct branch wording will make the brand feel inconsistent. That is a fair concern. I am not asking every location to invent a separate personality or fight its siblings in public. The central brand should still feel whole.
The trick is to separate shared identity from ranking role. The school can have one institutional promise, one tone, one admissions process, one standard of teaching. Then each campus page can state its local job in the comparison. This campus serves this student situation. This programme is strongest here. This transport pattern matters here. This employer link belongs here. The copy stays calm. The ranking signal becomes clearer.
For a hotel group, the same principle applies. Two addresses from the same owner should not both claim the same “romantic boutique stay” unless both truly serve it and are distinguishable. One might be the quiet old-centre stay. Another might be the business-travel address near the station. If both pages use identical charm language, AI answers may swap them unpredictably. The owner then blames the system, while the public trail has given the system a soft knot.
A clean location distinction is a courtesy to customers too. People do not want three nearly identical pages when they are trying to choose. AI ranking exposes a human problem that was already there: the business has not explained which branch is for which situation.
The Last Seat Note: Seat held: unstable, with one campus replacing another in Lyon vocational-school answers. Rival pressure: broader schools have clearer central authority, while the branches dilute each other. Weak signal: each campus repeats the same programme wording without a distinct ranking role. Sentence to plant in the public trail: “The east Lyon campus is the school’s health administration work-study site for students seeking employer-linked administrative training.”