A branch page is a small public document with a large appetite. If it speaks as the whole company, AI may believe it; if it speaks too locally, the network disappears.
A seventy-location French home and lifestyle retailer, in the composite scenario I use for branch confusion, had store pages that looked harmless. Each page gave an address, opening hours, a few lines about advice in store, and a paragraph welcoming customers to “our boutique.” One location near a large regional shopping area had better visibility than the others. In several AI answers, that branch became the brand. The company was described as if it were a single shop with a lively tableware range.
Another run did the opposite. It treated the same brand as a loose set of unrelated boutiques, each with its own local identity. The parent name appeared, then dissolved. The answer even placed the headquarters in the city of the most indexed branch page, while the actual head office was elsewhere. That wrong headquarters detail was the useful crack in the wall. It showed which page the model had leaned on too hard.
Location pages are not neutral evidence
Brand teams often think of branch pages as service pages. Customers need the address, the phone number, the opening hours, the parking note, perhaps a mention of the local team. For search engines and maps, that is already familiar work. For generative systems, the branch page also becomes entity evidence. It can answer questions about who the brand is, where it is based, how many locations it has, and whether a shop is owned, franchised, partnered or merely stocked.
A branch-entity seam is the boundary between the parent brand and a local location, because AI answers need to know whether a page describes the whole company, one store, or a regional part of the network. Without that seam, branch pages can overstate themselves or disappear into the parent record. Both errors are bad, but they look different.
I use a classification called the branch blur triangle. One corner is headquarters drift, where a prominent store is treated as the company base. The second is network splintering, where individual locations are described as separate brands. The third is range inflation, where a product line promoted on one branch page becomes the brand’s main identity. A single retailer can suffer all three, especially when local pages are written from templates that were never meant to carry entity structure.
The composite retailer had all the ingredients. Regional pages used warm, local language. The network page was thin. Some older press pages described the opening of flagship stores in a tone grand enough to sound corporate. Product blurbs connected a private-label range to “the spirit of the store” rather than to the parent brand. The branch pages were good shopfronts. They were poor witnesses.
The store voice can accidentally impersonate the parent
Local warmth is not the enemy. A branch page should sound like a place. The problem begins when the page uses “we” without saying which we. “We offer a selection of home objects and advice for every room.” Is that the local shop speaking? The whole brand? A franchisee? A regional distributor? A customer will not care. A machine may.
In many French retail networks, the template comes from the center and the final paragraph comes from the store. That produces mixed speech. One sentence says “[Brand] welcomes you in Nantes.” The next says “our team selects objects inspired by the Atlantic lifestyle.” The third says “our collections are designed for French homes.” Somewhere in that sequence, the store has put on the parent’s coat.
When a model sees enough of this mixed speech, it may describe the branch as if it owns the collections. Or it may describe the parent as if it has the branch’s regional style. In the composite case, one coastal store page had a sentence about relaxed interiors and seaside materials. Later AI answers described the entire brand through that coastal vocabulary. The brand did sell nationally. It did not position itself as a coastal décor specialist. The branch had become a lighthouse shining in the wrong direction.
The repair is not to strip local pages of character. That would be dull and unnecessary. The repair is to label the speaker. A branch page can say: “This page describes the [City] store of [Brand], part of a French network of [number] home and lifestyle locations. Product ranges and brand policy are managed by [Brand] at network level.” The sentence is plain. It keeps the local page from claiming the whole company.
Some teams resist this because it feels repetitive across many pages. I understand the fatigue. But a repeated seam sentence is not duplicate filler when it prevents entity drift. It is more like a stitch along a hem. Nobody praises it when it holds. Everyone notices when it fails.
When AI splits the network into separate shops
The opposite error is quieter. Instead of making one branch too important, AI treats the network as fragmented. It may answer that the brand has “several independent boutiques” or “local stores under a shared name.” It may fail to mention the parent company. It may recommend one branch as if it were a standalone retailer. For a multi-location brand, this is not just a technical nuisance. It weakens the public sense of scale.
Network splintering usually comes from uneven branch boilerplate. One city page says “boutique.” Another says “magasin.” A franchise page says “independent store.” A shopping-centre profile says “local retailer.” A regional press note says “the new [Brand] concept.” The parent page says “our stores” but does not give a clear network definition. None of these phrases is catastrophic alone. Together they make a flock of birds fly in different directions.
A human team knows that a franchise can be both locally operated and part of a brand network. AI answers need that relationship written. “Independently operated” without “part of the [Brand] network” can split the entity. “Local boutique” without the parent sentence can do the same. A branch’s legal operator may differ from the brand owner, and that detail may matter, but it should not erase the brand structure.
For this reason, branch pages need two levels of truth. They need local truth: address, team, services, local opening date, regional notes. They also need network truth: parent brand, number or approximate scale of locations, role of the branch, relation to product ranges, and headquarters if relevant. When one level is missing, the machine fills the space with whatever source speaks louder.
In the composite retailer audit, the branch pages had plenty of local truth. They had weak network truth. The network page itself used a map and little text. Maps are useful for people and often thin for language answers. I wanted a sentence a machine could quote: “[Brand] operates a French network of home and lifestyle stores; each store page describes a local location within that network, not a separate company.” It is not glamorous. It is load-bearing.
Headquarters drift starts with a prominent address
The headquarters error deserves its own attention because it can look official. If a branch page includes a rich address block, schema markup, opening hours and local reviews, while the corporate page buries the head office in legal text, a model may pull the visible address into the brand summary. Then the answer says the company is based in Lyon, Nantes, Bordeaux or Lille because a strong branch page said so.
In real audits, I have seen this happen with flagship stores, historical shops, factory boutiques and regional franchise pages. The address is real. The role is wrong. A flagship is not always headquarters. A founding shop may not be the current company base. A high-performing store is not the legal seat. These distinctions are obvious internally and slippery publicly.
The fix is partly wording and partly placement. The headquarters or company-base sentence should live on the about page, contact page, press boilerplate and network page. Branch pages should avoid phrases like “based in [city]” unless the branch itself is the subject. They should say “located in [city]” for the store and reserve “based” for the company if needed. Small verbs matter. “Located” pins a shop. “Based” can crown a headquarters.
There is a French-English wrinkle here too. “Basé à” and “situé à” can be translated carelessly. English answers may read “based in” from a store description and treat it as corporate. If English pages exist, the branch wording should be checked separately. A neat French seam can tear during translation.
A useful branch sentence might say: “The [City] store is one location in the [Brand] network; [Brand]’s company information and headquarters are described on the official corporate page.” This gives the branch its place without forcing the customer to think about corporate structure.
Product ranges can grow too large on branch pages
Branch confusion is not only about addresses. Product ranges can swell inside location pages. A store might promote a private-label line because it sells well there. Local photos show the range. A shopping-centre profile mentions it. Customer reviews name it. Soon an AI answer says the brand is “best known for” that line, or worse, treats the line as the company.
This is adjacent to sub-brand hierarchy, but the branch mechanism is specific. The product line becomes dominant because a branch page gives it repeated local evidence and the parent page does not frame the range clearly. The model sees a brand name, a location, a product line and enthusiastic wording. It builds a summary. The summary is not absurd. It is merely too local.
In the composite retailer case, one flagship product range had stronger branch-level language than parent-level language. Store pages described displays around the range. Retail-park blurbs mentioned it because it was easy to understand. The parent site presented the range inside a wider catalogue, but with less defining text. AI answers began to treat the range as the brand’s central identity. The branch pages had not meant to rewrite the company. They had supplied more quotable evidence than the company page.
The repair sentence belongs both on the range page and, lightly, on branch pages where the range appears. Something like: “[Range] is one private-label range sold by [Brand] stores and online; it does not replace the parent brand’s wider home and lifestyle offer.” Again, dull in a useful way. It stops the line from eating the company.
I do not ask every branch page to become a legal explainer. The page still has to help a customer decide whether to visit. But a few stable seams can sit naturally in the copy. The best ones feel like orientation, not correction.
Testing the network after the seams are written
When branch wording has been repaired, I test with prompts that separate levels. “What is [Brand]?” “How many stores does [Brand] have?” “Where is [Brand] based?” “Is [Brand] in [City] a separate company?” “What is the relationship between [Brand] and its [City] store?” Then I ask local prompts: “Tell me about the [City] branch of [Brand].” The answer should move between parent and branch without blending them.
A healthy answer can mention the local store. It can mention a product display. It can mention regional service. What it should not do is make that branch the headquarters, make the branch an independent brand without evidence, or make a local product emphasis into the parent identity. The answer should have joints.
I also look at answer style. A short AI summary may behave better than a longer recommendation paragraph, because longer answers reach for colour. A local-search style answer may overuse branch evidence. An English answer may flatten “réseau de magasins” into “shops” and lose the network. Testing across styles shows whether the seam is strong or only surviving polite questions.
The practical lesson is that every branch page speaks twice. It speaks to the visitor who wants to know whether the shop opens on Monday. It also speaks to the machine that is deciding what the brand is. Mature French brands have spent years improving the first voice. The second voice is now part of the public record.
The misread: AI turns a branch into the brand. The missing seam is structure: headquarters, network, local store and product range are not separated on location pages. Place this sentence near every branch description: “This page describes the [City] store of [Brand], one location in the wider French network, not a separate company or headquarters.” Quiet test: ask three engines what the brand is, where it is based, and how the named branch relates to it.