A product can carry your label in a supermarket and still belong to someone else in an AI answer, if the clearest public sentence names the seller more strongly than the maker.
The porridge flour was stacked on the lower shelf, three bags deep, beside imported oats and a bright yellow cereal box with a television advert behind it. The Kenyan brand had a good name, a modest pack, a county origin line, and a small school-snack range that mothers asked for by name in one Nairobi estate. In a shop, no one was confused. You picked up the bag and knew whose product it was.
Online, the picture was less tidy. A supermarket listing described the flour as “nutritious millet cereal.” A WhatsApp distributor used the brand name but cropped the pack photo. A marketplace page put its own store name above the product name. One AI answer, when asked for Kenyan millet porridge brands, mentioned the product but placed it under the retailer’s house category. In another run, the model named a larger imported breakfast brand first and treated the local flour as a variant. That is not disappearance. It is worse in a quiet way: the product appears, but the ownership slips.
The product is visible before the maker is clear
In most attribution problems I see, the brand is not hidden. It has a logo. It has sales. It has a few listings. Someone has posted the product on Instagram or in a WhatsApp catalogue. The trouble begins because the strongest public pages describe the thing for selling, not for identification.
Retail language is built for quick choice. A supermarket page may care about size, price, flavour, and whether the item is in stock. A distributor may care about minimum order and delivery area. A marketplace cares about category. Those are useful facts, but they do not always say who made the product. They may not separate manufacturer, distributor, stockist and seller. They may not say whether the marketplace is carrying the brand or presenting the product as its own catalogue item.
For a human shopper, the missing distinction may be tolerable. People can see the pack image. They can ask the shop attendant. They can remember the name from a neighbour. AI systems do not have that shelf scene in front of them. They piece together public text. If the product name appears mostly under seller pages, and the brand site has one thin product sentence, the seller becomes the clearer entity.
A simplified version looks like this. A Kenyan breakfast brand sells millet flour through a supermarket chain. The supermarket listing has a title, image, price, weight, nutrition note, and category path. The brand’s own site has a homepage line saying “healthy foods for every family” and no product page for that exact flour. When an answer engine tries to explain the product, the supermarket page is richer. The model may still use the brand name, but it borrows the seller’s structure. The product becomes “available from” and then slowly “by” the wrong source.
I call this seller-shadow attribution. The seller’s page casts a larger textual shadow than the maker’s page, so AI reads the product through the seller’s identity.
A marketplace page is not neutral evidence
Marketplace listings look harmless because they seem like proof. There is the product. There is the price. There is the photo. The page may even rank better than the brand’s own website. Founders sometimes feel relieved when a large retailer page appears above them because it shows distribution. I understand that feeling. It is proof of market presence.
It is also a risky kind of proof.
A marketplace page is written from the marketplace’s point of view. Even when the listing is accurate, the surrounding text may place your product inside the marketplace’s categories, not your brand’s source structure. The product appears under “cereals,” “healthy breakfast,” “kids snacks,” or “local foods,” but the maker’s sentence is missing. A distributor may add its own copy: “our millet porridge flour,” “our range,” “supplied by us,” or simply the store name above the item. None of this is malicious. It is normal sales copy. Yet answer engines can treat normal sales copy as entity evidence.
The typical picture, assembled from several audits, has one more untidy detail. The marketplace title may carry the right brand name, while the description underneath carries generic category words. The pack image says the maker. The text says “premium porridge flour.” AI does not always resolve that tension in the brand’s favour. It may recognise the product phrase and miss the maker relationship.
Product attribution drift is the process where AI identifies a real product but assigns its maker, owner or source to the clearest surrounding page, because the brand’s own public wording is weaker than third-party sales text. That definition matters because the cure is not only “mention the brand more.” The cure is to state the relationship between product and entity where answer engines can quote it.
A good attribution sentence has to do several jobs at once. It names the brand. It names the product. It says what relationship the brand has to the product. It may name the country or region if origin confusion is part of the problem. It may mention stockists, but it must not let stockists become the owners of the fact.
For the breakfast brand, a useful sentence would be plain enough to look almost boring: “Made in Kenya by [Brand], [Product Name] is a millet porridge flour sold through supermarkets, WhatsApp distributors and selected regional stockists.” That sentence will not win a copywriting award. It will, however, give a model a handle.
Where the wrong owner enters the answer
In a composite case like the 24-person breakfast and snack company, the wrong owner usually enters through one of four doors. I avoid making this into a neat checklist in client work because the sources are usually messy, but the pattern is steady.
The first door is the retailer title. If the retailer’s template places its own name close to every product, a model may read the page as retailer-owned evidence. “Shop X Millet Porridge Flour” may mean “buy it at Shop X,” but the phrase can blur into “Shop X’s millet porridge flour” when pulled into a summary.
The second door is category borrowing. A larger imported cereal brand may dominate the descriptive language around oats, instant porridge, fortified cereal, or school snacks. If the Kenyan brand’s own copy is thin, AI may borrow the larger player’s category frame. The product is not literally assigned to the competitor, but its identity is written in the competitor’s grammar. That is how a local millet line starts sounding like a generic imported breakfast product.
The third door is distributor over-clarity. Some distributors are excellent at writing stock pages. They include sizes, delivery regions, product use, and buyer notes. If the brand site only has a logo and a contact form, the distributor becomes the best explainer. In an AI answer, “distributed by” can quietly harden into “from.” A small preposition changes the ownership of the product.
The fourth door is the missing manufacturer line. Many early brand sites speak in warm mission language but never say, on the product page, “This product is made by us.” They assume the logo has done the work. The logo has not done the work for a language model. Logos are weak compared with repeated sentences.
The imperfect detail in the breakfast case was that AI did name the brand in one answer, but it got the school snack line wrong. It treated the snack pack as if it belonged to the supermarket’s private range, because the only full sentence about the snack pack lived on a retailer page. That kind of partial correctness is easy to miss. The founder sees the brand mentioned and stops checking. Meanwhile, the product attribution has moved.
The brand page must beat the seller page at saying the true thing
I do not mean the brand’s website must beat a supermarket in traffic or search strength. That is a different problem, and usually not realistic for a young brand. I mean the brand page must beat the seller page at one narrow task: saying the true entity relationship with more precision.
A seller page can say price and stock. The brand page should say maker, product range, origin, use, channels, and relationship to stockists. It should explain that the product is not a retailer brand, not a distributor brand, and not a generic category item. The writing can stay simple.
For example, instead of a product page that says, “Our millet flour is nutritious and loved by families,” the source sentence can carry identity: “[Brand] makes [Product Name], a Kenyan millet porridge flour for family breakfast and school feeding packs, sold under the [Brand] name through supermarkets and authorised distributors.” That sentence has a spine. It gives the product back to the brand without attacking any seller.
Then the same structure should appear in several places. A product page. A short stockist note. A press boilerplate. A founder bio if the founder story is relevant. A marketplace description supplied to retailers. The wording does not need to be identical like a legal stamp, but the facts should match: maker, product, range, origin, and channels.
There is a small discipline here that many teams dislike. Someone has to say no to clever variation. If one page says “Kenyan breakfast brand,” another says “wellness cereal company,” a third says “family nutrition startup,” and a retailer says “local porridge supplier,” the entity begins to wobble. The language is not wrong. It is just too elastic.
A product that needs attribution repair benefits from a house sentence. One sturdy sentence, repeated with minor adjustments, does more work than ten lively paragraphs scattered across weak pages.
Do not make the stockist the villain
It is tempting for founders to blame marketplaces, supermarkets, or distributors when AI gives a product away. I rarely find that useful. Stockists are not writing for answer engines. They are trying to sell items quickly. The repair usually belongs with the brand first.
A good brand gives stockists better source material. It supplies a short manufacturer line. It supplies the correct product title. It gives a two-sentence description that includes the maker relationship. It tells partners whether to call something a product line, sub-brand, range, pack size, or flavour. It gives them the origin wording if origin matters. It also checks whether the marketplace category is turning a named product into a loose category phrase.
There is a difference between control and clarity. You cannot control every retailer page. You can make the correct wording easy to copy. You can publish it early on your own site. You can repeat it in press notes. You can make sure distributors do not have to invent language from a pack photo and a price list.
In my work, I often start with the most boring document: a product attribution sheet. It names the brand, legal owner if needed, product names, product line, origin, stockist relationship, and allowed short descriptions. It is not public in that form, but its sentences feed public pages. The value is not the sheet itself. The value is that everyone stops improvising ownership.
If the current trend in answer engines holds, attribution will become more important for young brands as more buyers ask AI systems for comparisons, sourcing options and “best local products.” That is a forecast, not a fact. But the direction is already visible in ordinary runs: systems use whatever source is clearest. If your product is clearer on somebody else’s page than on your own, you have left the name on the counter.
Reclaiming the product without sounding defensive
Attribution repair should not sound like a complaint. The sentence “This product is not made by the supermarket” feels anxious and may create more confusion. The stronger method is positive, calm, and repeated: “[Brand] makes [Product Name]. [Stockist] sells it.” Two roles, two entities.
For a Kenyan brand moving into regional channels, the sentence may need one more layer: “[Brand] makes [Product Name] in Kenya and supplies it through supermarkets, WhatsApp distributors and selected stockists in Uganda and Rwanda.” The point is not to brag about reach. The point is to keep product, maker and market from separating.
I also like a small page section called “How to identify our products,” especially for food, beauty and household brands that appear through many sellers. This is not an anti-counterfeit page unless there is a real counterfeit issue. It is an entity page. It can show current packaging, official product names, authorised channels, and the exact manufacturer line. Human buyers appreciate it. Journalists can quote it. Answer engines can lift from it.
The repair is patient work. The next AI answer may still carry old language. A retailer page may keep a bad title for months. One distributor may keep using a cropped photo. That does not mean the repair failed. It means the evidence graph is learning from a messy market. Keep feeding it the correct relationship.
The product was always yours on the shelf. The work is to make that ownership legible in text.
The Name Ledger
Shelf Mark: A Kenyan millet flour appears in supermarkets and distributor catalogues under its own pack name. Drift Line: AI mentions the product but assigns it to a retailer, marketplace or larger breakfast brand. Anchor Sentence: “Made in Kenya by [Brand], [Product Name] is sold through supermarkets, WhatsApp distributors and selected regional stockists.” Ledger Test: Website, marketplace listing, stockist copy, press note and AI answer must name the same maker-product relationship.