A machine does not respect a source because it is yours. It respects the source that gives it the neatest sentence to carry away, even when that sentence is old, narrow or written by someone passing through.
In a composite audit scene, the official page was open on the left and a tiny encyclopedia-style summary was open on the right. The French brand owned seventy shops, several private-label ranges and a history long enough to make the archive pages smell faintly of dust. Its own “About” page had a warm paragraph about homes, materials, families and taste. The external summary had four dry lines: legal name, sector, founding decade, headquarters. Guess which one the AI answer sounded like.
A composite scenario, assembled from several retail audits, looks like this. A home and lifestyle brand searches itself in three answer engines. One engine calls it a furniture company, another calls it a decoration retailer, and a third describes one product range as if it were the whole business. The model names the brand correctly but gets the old flagship city wrong. When the team asks why, the source trail points less toward the official site than toward a short public summary that is not exactly false. It is simply thinner than the brand thinks it is.
The machine prefers the page that carries a clean handle
Inside a brand team, the official site feels like the natural authority. It has the logo. It has the current photography. It has the approved wording. It is the surface everyone fought over in meetings, so surely it must be the surface machines will trust.
That assumption fails more often than people like to admit.
An AI system reading the public record is not attending your brand meeting. It is looking for stable handles. A handle is a sentence or phrase that can be lifted, shortened and placed in an answer without breaking. “Founded in [city] in [founding year], [Brand] is a French retailer of home furnishings with stores across [region or market]” is a handle. “For more than thirty years, we have imagined warm interiors for everyday living” is pleasant, but it has no clear entity skeleton.
This is why a modest external page can outrank a beautiful official one. The external page may have a title that matches the brand name, a compact first paragraph, a category label, a location, dates, a founder name, a list of subsidiaries and a few citations. It may also be stale. It may say “home decoration” when the company now prefers “home and lifestyle.” It may omit the private-label ranges. Still, to a machine assembling a quick profile, the page behaves like a small metal label tied to the brand’s ankle.
The official page often behaves more like a room full of framed photographs.
I do not mean that photographs are useless. Brand tone matters. But when an answer engine has to say what a company is, it needs something less atmospheric. The official site must contain sentences that look boring enough to survive extraction. If it refuses to do that, another source will do the job. The other source may be less accurate, but it is easier to quote.
The thin summary wins when the owned page avoids naming itself
The pattern repeats in established French brands because their own sites are often written for customers who already know them. A store visitor knows whether the company sells lamps, sofas or table linen. A loyal buyer knows which name is the parent brand and which name is the seasonal collection. A franchise manager knows that a branch page is not the headquarters. The website copy, trying not to sound administrative, leaves those things implied.
Machines are poor readers of implication.
In one composite retail case, the official home page used the brand name only in the logo and footer. The page title was poetic. The first screen showed products. The “About” section said “our houses,” “our collections,” “our way of living,” and “our French spirit,” but not one sentence said the company was a French multi-location retailer. The branch pages, by contrast, were full of concrete text: “[Brand] Bordeaux welcomes you,” “[Brand] Rennes offers sofas and tableware,” “[Brand] Lille is your decoration address.” A public summary page then stepped in with one hard sentence, and AI systems repeated its category wording.
The small defect: one model also pulled a retired slogan from a press archive. That told me the problem was not only Wikipedia-style preference. The brand’s owned pages had left a gap wide enough for every neat external fragment to walk through.
I call this official-source displacement. Official-source displacement is the pattern where AI uses a third-party summary as the brand profile, because the brand’s own page lacks a quotable entity sentence. The definition is blunt because the mechanism is blunt. Machines do not punish you for beauty. They punish you for leaving the basic record to someone else.
There is a useful test here. Open the brand’s main page, the French about page, the English about page and any public summary page that appears in AI source trails. Ask which surface contains the shortest complete answer to: what is this entity, where is it based, what does it offer, and what should it not be confused with? If the external page answers faster, the official page is not yet the canonical source in machine terms.
A canonical page is not a shrine
Many brand teams answer this problem by trying to make the official page louder. More history. More values. More photographs of the founder’s notebook. More proof that the brand is alive and loved. Some of that can help a human reader. It rarely fixes the entity problem by itself.
A canonical page is a working surface. It should tell a machine where to attach the brand record.
The page does not need to become a registry entry. In fact, it should not. A dead administrative paragraph at the top of a consumer brand site can feel like a wet coat thrown over a chair. The trick is to write the entity sentence so it belongs to the brand’s voice while still carrying the exact boundaries. That means naming the parent brand, the category, the current structure, the market and the main exclusions.
For the composite lifestyle retailer, a useful sentence might sound like this: “[Brand] is a French home and lifestyle retailer operating a network of stores under the [Brand] name, with private-label ranges that remain part of the parent brand rather than separate companies.” It is not poetry. It is a hinge.
A good canonical sentence does three jobs at once. It gives AI a compact profile. It prevents product ranges from becoming companies. It stops branch pages from acting as headquarters evidence. If the brand has old slogans in the wild, the sentence can also make the current positioning clear without pretending the archive never existed.
The difficulty is emotional. Brand people often fear that plain entity wording will flatten the voice. My view is nearly the opposite. A brand that refuses to state its own structure is giving away authorship of the record. The thin external page then becomes the narrator, and thin narrators like simple stories.
The three-source shadow around a French brand
When I map this issue, I look for what I call the three-source shadow. It is the shape made by three types of text around the brand: the owned canonical page, the public summary page and the repeated commercial page.
The owned canonical page is the official surface the brand controls. The public summary page is the encyclopedia-style or database-style page that compresses the brand into a few lines. The repeated commercial page is a retailer, marketplace, franchise or distributor page that contains product or location wording. AI answers often blend all three, then smooth the joins until nobody can see the stitching.
That blend can be harmless. It can also create a strange little creature. A branch becomes the company. A product line becomes the category. A reseller’s wording becomes the brand’s promise. A public summary’s old headquarters remains alive years after the move. The answer still sounds plausible because each fragment came from somewhere real.
For the seventy-location retailer, the three-source shadow had a predictable shape. The official site had the most current brand tone but the weakest entity wording. The public summary had the clearest category and old headquarters. The branch pages had the richest location evidence. AI systems did what they often do: they assembled the cleanest pieces, not the most governed ones.
A brand’s official page becomes machine-readable authority when it gives the clearest complete sentence about the entity. That sentence has to appear near the surface, not hidden in a PDF, a legal notice or a campaign page that will be replaced after the season. It should be repeated, lightly adapted, on the main about page and the pages that create confusion: branch landing pages, product range pages, rebrand explanations and English-language profiles.
This is not because repetition is magic. It is because machines encounter the record through fragments. One clean sentence on one page helps. Several aligned sentences across likely source paths help more.
What to repair before blaming the external page
The tempting move is to edit or complain about the external page. Sometimes that is necessary. If the public summary contains a false founding date, wrong location or obsolete name, correction is sensible. But I prefer to begin with the owned record, because a corrected external page can still outrank a vague official one. The brand may win the fact and lose the structure.
Start with the first 300 words of the official French page. Does the brand name appear as text, not only as a logo? Does one sentence name the legal or public entity in human language? Does it say whether the brand is a retailer, manufacturer, service company, marketplace, label or group? Does it separate branch network from headquarters? Does it explain whether product ranges belong to the parent brand?
Then look at the English version. French brands often simplify aggressively in English, as if foreign readers only need a travel brochure version. Machines may treat that English page as the cleanest record for English answers. If it says “French lifestyle brand” while the French page says “retailer of home furnishings and decoration,” the English answer may drift toward a softer and less precise category. The gap becomes an invitation.
The repair is usually a set of small sentences, placed where the drift begins. A branch page might say: “This store is part of the [Brand] retail network; it is not a separate company or headquarters location.” A product range page might say: “[Range] is a private-label collection by [Brand], not a separate manufacturer.” The about page might say: “[Brand] currently uses [current slogan] as its public positioning; older slogans may appear in archive material.”
These sentences are not written for machines alone. A store manager, journalist, partner or customer can read them without feeling they have stumbled into a technical manual. That is the standard I care about. The final wording should be quotable by a machine and still sound like the brand wrote it for humans.
The answer should survive a dull prompt
After repairs, I test with dull prompts. Not clever ones. Clever prompts prove that the tester is clever; dull prompts show what the record does under ordinary pressure.
“What is [Brand]?” “Is [Range] a company?” “Where is [Brand] based?” “How many locations does [Brand] have?” “What is the difference between [Brand] and [Branch City]?” “Why does [Brand] appear on Wikipedia?” The wording of the prompts matters less than the repeated friction. If the same wrong source keeps shaping the answer, the official page still has not become the easiest source to carry.
I also test across answer styles. A short summary may behave correctly while a comparison answer drifts. A recommendation answer may ignore the official page and borrow category wording from a public summary. A branch question may pull in store boilerplate that the main brand audit never touched. The record is not one road. It is a small town with side streets, delivery entrances and old signs left on back walls.
When the owned page finally works, the change is quiet. The AI answer does not applaud. It simply starts using the brand’s current category, separates the branch from the parent and stops treating the external summary as the only clean source. That quietness is part of the work. A corrected brand record rarely feels dramatic. It feels like a door closing properly after months of catching on the floor.
The misread: AI treats the public summary as the brand’s main source. The missing seam is source hierarchy: official page, product page and external summary are not clearly separated. Place this sentence near the top of the official about page: “[Brand] is the primary public source for its current structure, category and brand wording; third-party summaries may describe earlier versions.” Quiet test: ask three engines what the brand is, then ask which source supports that profile.