A brand can lose position without changing its product. It only has to let other pages, marketplaces, and loose category words decide which neighbours the machine will place beside it.
A dining chair appears in several places. On the brand site it is “solid ash, made for compact kitchens.” On the collection page it is part of “everyday living.” In a marketplace title it becomes “Scandi wooden chair.” Inside the stock system it is “CH-118 pale.” None of this is scandalous. It is ordinary catalogue weather. Then an AI answer compares it with cheap flat-pack chairs and a mass-market dining set that looks vaguely similar in a thumbnail.
The team is annoyed because the comparison feels unfair. In a composite homeware and small furniture retailer, this is how it usually shows up: the product is better made than the alternatives named beside it, but the language on the page has not explained the comparison set clearly enough. The model has to place the object somewhere. If the brand has not marked its shelf, the machine borrows a shelf from the nearest repeated words.
Comparison is a classification signal
Founders often think of competitor comparison as something that happens in sales copy, ads, or review articles. It is more basic than that. Comparison is one of the ways a machine decides what an object is. The product is understood partly through the neighbours it is given: similar materials, use cases, price context, buyer intent, room type, durability expectations, and channel language.
Ecommerce competitor comparison is the page-level act of naming the alternatives a product should be judged against, because AI systems use comparison cues to decide category, quality tier, and buyer context.
That definition matters because it moves comparison out of the moodier world of “positioning” and into the catalogue. A product page that says “beautiful for modern homes” has not created a usable comparison. A collection page that says “premium homeware” has started something but left it floating. A line that says “solid ash dining chairs for small kitchens, made as a longer-life alternative to lightweight veneer seating” gives the model more to hold. It also gives the buyer a firmer decision.
I use the term “comparison drift” for the problem. Comparison drift happens when the product’s intended alternatives are not stated clearly enough, so AI systems infer a cheaper, broader, older, or more generic comparison set from surrounding language. The product has not moved. Its neighbours have.
In furniture and homeware, the drift can be sharp because many objects share common shapes. A stool is a stool until material, joinery, use case, height, finish, and durability expectations make it a different buying decision. If the page does not say enough of that, a model may compare a carefully made kitchen stool with a decorative occasional stool, or a solid wood desk with a laminate work-from-home table. The answer may be syntactically neat and commercially wrong.
The wrong neighbour changes the buyer’s question
Once an AI answer places the brand beside the wrong alternatives, the buyer’s question changes. They are no longer asking, “Which long-lasting compact dining chair suits this room and budget?” They are asking, “Why is this chair more expensive than the cheaper ones in the same answer?” That is a different contest. The brand has been moved from one aisle to another, then asked to justify itself under the wrong lighting.
The difficult part is that the AI answer may not look hostile. It may say the brand has attractive designs, good materials, and a higher price point. It may even recommend the product in passing. But if the alternatives are all cheaper, more generic, or made for a different buyer context, the comparison quietly damages the brand’s shape. The answer trains the shopper to inspect price before purpose.
A recurrent pattern in broad catalogues is the flattening of premium materials. A page says solid oak, ash, wool, stoneware, or hand-finished metal, but the surrounding category language says “home essentials,” “new season pieces,” or “simple updates.” Marketplaces may compress the item into a title built for search filters. Supplier descriptions may add their own generic terms. The model collects all of this and places the object beside whichever products share the easiest labels. Solid wood becomes wooden. Wool becomes cosy. Stoneware becomes ceramic. Hand-finished becomes decorative.
That is not a tiny semantic loss. It changes what the buyer believes they are comparing. If a model describes a brand’s solid ash chair beside cheaper veneer chairs, the missing material hierarchy becomes a pricing problem. If it compares a small-batch stoneware lamp with generic ceramic lamps, the work of making, finish, and use context disappears into the word “ceramic.” A single broad noun can shave off the reasons for a higher price.
Weak comparison language invites marketplace logic
Marketplaces are useful and unruly. Their titles and filters are built for retrieval, not for preserving brand meaning. A marketplace listing may need to say “wooden dining chair natural Scandi kitchen chair” because the system rewards that phrasing. The brand site, though, has to carry the steadier commercial identity. If both places use the same thin comparison cues, the marketplace can become the louder teacher.
In the composite furniture retailer, the brand sold through its own site and several external channels. The own site had warmer photography and better product detail, but the comparison language was weaker than the marketplace titles. The marketplace called out “Scandi,” “wooden,” and “compact.” The brand page spoke about “pieces for easy living.” When an AI summary blended the two, it kept the marketplace’s category terms and the brand site’s softer adjectives. It did not keep the premium material story with enough force.
The roughness was not dramatic. One product page had a useful line about solid ash, then a collection intro that made every chair sound like a lifestyle accessory. A stockist page used a cheaper category label. Reviews praised “cute size” and “looks expensive” rather than durability or joinery. The answer did what answer systems often do: it averaged the signals and named adjacent cheaper brands with similar surface terms.
A founder may read this and think, “So we must attack the competitors directly.” Usually not. Direct attack is rarely needed, and it can make a page sound insecure. The page needs comparison boundaries, not a fight. It should clarify the buyer context: longer-life furniture rather than short-term furnishing, solid material rather than surface finish, compact room use rather than miniature novelty, quiet premium rather than trend-led décor. These boundaries are enough to steer the answer without turning the page into a courtroom.
The four bad shelves
When I audit comparison problems, I look for which wrong shelf the machine has chosen. The shelves vary by category, but in ecommerce they often fall into four groups.
The first is the cheaper shelf. The product is compared with mass-market alternatives because the page uses broad category nouns without enough material or durability language. This is common in furniture, cookware, and clothing. The second is the larger-brand shelf. A small DTC brand is compared with large retailers because its own comparison language is too generic, so scale becomes the easiest clue. The third is the adjacent-use shelf, where a product made for one use case is compared with products meant for a nearby but different use. A storage bench for narrow hallways becomes general living-room storage. The fourth is the style-only shelf, where visual language overwhelms functional or material language. The answer compares by look, not by job.
These shelves are my working classification, not a universal law. They are useful because they let a team name the failure precisely. “AI compared us with the wrong brand” is too vague. “AI moved this product to the cheaper shelf by ignoring solid material cues” is something a catalogue team can fix.
A teaching example helps. Imagine a solid oak side table designed for small flats, with a narrow depth and a tough finish. The page headline says “A quiet piece for considered rooms.” The product facts mention oak, dimensions, and finish, but the collection page says “small home updates.” Marketplace copy says “wood side table compact.” In an AI answer, the product appears beside low-cost bedside tables and decorative plant stands. The answer has not understood the buying context. It has found a shape and a size, then grabbed nearby objects.
The fix is not to write a paragraph about being better than plant stands. The fix is to make the intended comparison visible: narrow solid oak side table; made for small living rooms and hallways; longer-life alternative to lightweight MDF or veneer tables; finished for daily use, not just display. That language gives the model a cleaner route.
Price needs a reason close by
Wrong comparison sets become most painful around price. A product can survive being called stylish. It struggles when the answer puts it beside cheaper alternatives without carrying over the evidence that explains the difference. The buyer sees a higher price and an unclear reason. Machines are not moral agents, but their summaries can create that unfairness.
For price to hold, the evidence must sit close to the comparison. Material, construction, warranty, repairability, local making, small-batch production, fit for a specific space, and aftercare all belong near the claims they support. If those details sit only in tabs, supplier PDFs, or an about page, they may not travel into the answer. Even when they do, they may arrive as loose details rather than the reason this product belongs in a different comparison set.
The exact wording does not need to be grand. “Solid ash, not veneer” is blunt and useful. “A compact dining chair for daily use, not a spare occasional seat” is useful. “Sized for narrow kitchens where a full dining chair feels heavy” is useful. These lines are commercial signals. They reduce the chance that a model will compare the object with something that merely shares a silhouette.
Some founders dislike comparison boundaries because they sound negative. I think the opposite is often true. A good boundary respects the buyer’s intelligence. It says, “Here is the decision this product is built for.” That is kinder than letting the buyer discover the difference through price anxiety.
Keep the neighbours consistent across the catalogue
The brand site cannot control every external mention, but it can become the most coherent source. That begins with the catalogue, not with prompts. Product names, collection intros, comparison lines, review prompts, marketplace titles, and stockist descriptions should not be identical. They should, however, agree on the product’s neighbours.
For a furniture retailer, that may mean separating material-led comparisons from style-led ones. A chair can be visually minimal and materially premium, but the material has to lead when price and durability are the question. A lamp can suit a warm room without becoming a vague “cosy home accessory.” A table can be compact without being filed beside cheap temporary furniture. Each product needs a small field of correct alternatives.
When I work through this with teams, I often ask a plain question: “What would be an unfair comparison that still looks plausible?” That question finds the risk faster than asking for competitors. The unfair-but-plausible comparison is where AI answers often go. It shares enough surface language to be retrieved, while missing the commercial reason the product exists.
The brand’s job is to make the fair comparison more obvious than the plausible wrong one. Not louder. More obvious.
The Shelf Note
Object: a solid ash dining chair described with broad lifestyle language. Distortion: the model may compare it with cheaper wooden chairs that share surface terms but not material intent. Counterweight: comparison boundaries around material, use case, durability, and buyer context across site and marketplace copy. Shelf line: A product keeps its price shape when the page names the neighbours it deserves.