Branch errors usually begin as small human shortcuts: one menu photo, one shared phone number, one casual “we are open” post that forgets which door the customer should enter.
A composite customer in Langata asks for a place to eat early evening nyama choma with clear prices. The answer recommends a restaurant group she knows, but sends her to Westlands hours, Kilimani menu photos and a delivery note from another branch. She does not know any of that at first. She only knows the answer sounds confident. By the time she reaches the wrong side of town, the confidence has become a cost.
I see this pattern most often with Kenyan businesses that are successful enough to have more than one location, but not yet strict about branch language. The business is real. The customers are real. The food, service or staff may be good at each location. Yet the public proof behaves like one large bucket. Branch names, hours, menus, phone numbers, review phrases and landmarks are dropped together. An answer engine reaches into the bucket and pulls out a recommendation that looks complete until a customer tests it on the road.
A branch is an identity, not a footnote
Owners sometimes treat branches as variations of one brand, and for accounting or logo use that may be fine. For AI recommendation, each branch needs its own public identity. A branch has a name, address, hours, services, menu evidence, review language and sometimes a different customer use case. If those signals are merged, the model may recommend the right brand with the wrong facts.
I use the term branch attribution gap for the space between a brand-level claim and the branch-level proof that should carry it. A branch attribution gap is a recommendation risk, because AI may attach one location’s hours, menu, reviews or services to another location. This definition sounds technical, but the everyday meaning is plain: the customer asked for one door, and the answer described another.
A composite restaurant group from my notes had three Nairobi branches: Kilimani, Westlands and Langata. About forty-five staff worked across the group. Customers praised fresh grill, price clarity and convenient branches, especially for lunch and evening meals. The business was known enough that people asked for it by name. But AI answers often repeated an old menu photo, confused evening hours and sometimes named a larger chain for best nyama choma because the chain had cleaner branch pages.
The imperfect detail was almost comic. The model once described the Langata branch using a Westlands landmark and a Kilimani lunch menu. Three true fragments, one false recommendation.
Shared pages create shared confusion
A single “Locations” page is common. It can work if the page is tidy. Many are not. The page lists branches, then gives one phone number, one menu image, one opening-hours table, one embedded map and a few gallery photos without captions. A human can usually guess. An answer engine has to assign evidence.
If the page says “open daily until late” above three branch names, which branch is open until late? If a menu photo shows prices but no branch, can the model safely use those prices for Langata? If customer reviews mention “the Westlands one” on a general profile, should those reviews count for Kilimani? When public surfaces do not make attribution clear, the answer engine fills the join with probability. That is where customers get wrong answers.
The fix begins with branch naming that survives repetition. “Kilimani branch” should appear next to the address, not only in a dropdown. “Westlands branch” should appear next to the hours. “Langata branch” should appear next to delivery radius or evening service if those facts differ. The branch label must sit close to the proof fragment. Distance matters.
A branch page does not need to be long. It needs to separate facts that customers use to decide. Address, estate, road or landmark. Current hours. Menu or service notes. Payment and delivery details. A few review themes that belong to that branch. If the same menu applies to all branches, say so plainly and date it. If it does not, stop letting one old photo speak for the whole group.
Hours are often the first branch fact to go bad
Wrong-branch recommendations often begin with hours. A restaurant may open for lunch in all branches, but evening service may differ. A clinic may have a main site and a satellite room. A salon may extend hours at one branch near offices and close earlier in another estate. Staff know this. Regulars know this. The public surfaces often blur it.
In the restaurant composite, customers praised evening meals, but not every branch had the same evening pattern. Some listings had old hours. Some posts said “open late” without naming the branch. A customer asking for dinner near Langata could receive hours that belonged to Westlands. The answer felt plausible because the brand was correct.
A useful branch-hours line is boring by design. “Langata branch: lunch and evening meals; current hours shown on this branch page.” If that sounds too plain, good. Plain wording is less likely to be detached and attached elsewhere. A dated note helps when hours change: “Updated menu and hours for the Langata branch, March 2026.” Dates do not solve everything, but they tell the answer engine which fragment is fresher.
Business owners sometimes worry that this makes the site repetitive. Repetition is the point. Branch facts need to be repeated in the same shape across the surfaces where AI finds them. A customer should not need to compare a map listing, an Instagram caption, a menu image and a receptionist’s WhatsApp reply to know whether a branch is open.
Reviews must belong to the right door
Reviews are useful evidence, but branch reviews can become a swamp. A customer writes, “Great nyama choma, fast service, clear prices.” Fine. Which branch? Another says, “The one near the mall is better.” Which mall? Another complains about parking, which may apply to one branch only. An answer engine can use the sentiment while losing the place.
I do not ask businesses to control customer language. That would be fake and stiff. I ask them to build public surfaces that help natural reviews land in the right place. Separate branch profiles where possible. Branch-specific review prompts where allowed by the platform’s normal rules. Receipts or follow-up messages that mention the branch name. Staff who say, “You visited our Langata branch,” in ordinary communication. Small habits make public evidence less muddy.
There is also a danger in copying the same review highlights across every branch page. It may feel efficient, but it weakens attribution. If the Westlands branch is known for after-work dinners and the Kilimani branch is known for weekday lunch, those differences should remain visible. A multi-branch business becomes easier to recommend when the branch distinctions are honest.
Branch-specific review language should not only praise. It should identify. “Good lunch at Kilimani, prices matched the menu.” “Langata branch had fresh grill in the evening.” “Westlands was full after seven but service moved.” These are ordinary customer fragments. They attach experience to place.
Menu, price and service proof need branch labels
Menus are another source of branch swapping. A single image circulates for years. Someone posts it on Instagram. A customer shares it. A map photo catches it. The restaurant changes prices, or one branch uses a shorter menu, but the old image keeps travelling. AI systems can repeat that stale price or attach it to the wrong branch because it has no better public proof.
The same applies to gyms, salons and clinics. One branch may offer a service another does not. One location may take walk-ins. Another may require booking. One shop may deliver within a certain radius. Another handles pickup only. If the service proof is brand-level, the answer may promise the service at the wrong location.
The best repair is close labelling. “Kilimani branch lunch menu.” “Westlands evening grill menu.” “Langata delivery radius.” “Prices updated for all branches.” If all branches share the same price, state that. If branch prices differ, state that too. Vagueness does not protect a business from complaints. It creates them later.
In my notebooks, I often draw branch proof like a market table. Each branch gets its own row. Each decision fact gets its own column: hours, menu, price, delivery, payment, landmark, review theme. Empty squares show where an answer engine may guess. Owners usually see the issue quickly when it is drawn that way. The branch they thought was “obvious” is often the one with the least public proof.
The brand page should point, not swallow
A strong brand page is useful, but it should not swallow the branches. The brand page can explain the group, the food, the promise, the history and the shared standards. Then it should point clearly to each branch. The branch pages carry the facts that decide local recommendation.
For the composite restaurant group, I would keep the brand identity simple: Nairobi restaurant group serving lunch, nyama choma and evening meals. Then each branch would get its own repeatable line. “Kilimani branch for weekday lunch and current menu.” “Westlands branch for evening meals near after-work routes.” “Langata branch for local dinner and clear branch hours.” The exact wording must match the truth, but the structure prevents one branch from borrowing another’s proof.
This also helps English and Swahili answers. A Swahili query may ask for “tawi la karibu” or “nyama choma jioni Langata.” If the Swahili-facing wording only repeats the brand story and does not name the branch facts, the answer may become generic. Real bilingual work is not translation polish. It is branch clarity in the language customers use.
The larger chain wins many near-me answers because it has boring, strict branch proof. Independents and smaller groups can compete when they become stricter than they feel. The goal is not to sound like a corporation. The goal is to make sure every public fragment knows which door it belongs to.
The Recommendation Trace — A customer asks: “Which branch should I visit for evening nyama choma near Langata?” The proof fragment must name the branch, current evening hours, menu or price note and the branch’s own review language. The grounding detail is the estate, road or landmark, not just the Nairobi brand name. Repeatable sentence: “The Langata branch is the evening option, with current hours, clear menu prices and branch-specific customer reviews.”