An aggregator wins when it gives the model a tidy little box: name, category, place, and promise. Your own page can be richer and still lose if the facts are scattered like coins under furniture.
The directory page was wrong in three places and still easier to use. That is the sour part. In a composite scenario drawn from accounting and export-support audits, a twenty-eight-person firm near Nantes served French SMBs trading inside the EU. Its own website had the better information, somewhere. The service page mentioned VAT. A paragraph on another page mentioned customs documents. A team note hinted at export paperwork. The directory page, meanwhile, gave a blunt label: accounting firm for import-export businesses near Nantes. It flattened the offer and missed a boundary, but it was quotable.
I have seen this shape often enough that I no longer treat it as a mystery. The answer engine is not choosing the morally deserving source. It is choosing a source it can compress without doing surgery. A French business may know itself better than any directory, but if its page hides the defining facts inside soft claims and scattered paragraphs, an aggregator becomes the easier handle. Bad handles are still handles.
The aggregator’s advantage is usually mechanical
An aggregator page is rarely deep. It often has thin descriptions, stale categories, scraped details, broad service tags, and little sense of how the business actually works. Yet it tends to present certain facts in a rigid pattern. Name. Category. Address or area. Phone. Opening details. Reviews. Sometimes a short service label. From a machine’s point of view, rigidity has value. It reduces interpretive labour.
A company page is often more human, but not always more usable. The homepage opens with a large promise. The service page begins with a client pain. The about page tells a short history. The useful factual sentence is split across three locations. A reader with patience can assemble it. An answer engine may not.
This is where the owner feels the injustice. “But our site is authoritative,” they say. In a human sense, yes. It is the primary source. The company controls it. It should be trusted more than a directory. In practice, authority has to appear as extractable evidence. If the company page does not state the business identity, service boundary, location and proof in one stable place, the model may borrow the cleaner external version.
A primary source is not merely the source closest to the business, because AI systems need the page to state the business facts in a form that can be lifted. That is my working definition for this problem. Ownership helps. Accuracy helps. Neither compensates for a page that makes the machine guess.
I call the aggregator problem source-friction inversion. The worse source becomes easier to cite than the better source because its facts are packaged more tightly. The task is to reverse that friction, not by copying directory style, but by making the business page the clearest usable account.
A directory can be wrong and still shape the answer
In the Nantes accounting scenario, the aggregator’s wrongness had a particular texture. It did not invent an absurd company. It compressed the firm into a nearby category. The answer might say the firm handled “international accounting,” when the actual page supported French VAT, customs paperwork coordination and export-documentation support for SMBs trading inside the EU. That is close enough to sound plausible and wrong enough to cost trust.
Near-wrong is the most common danger. A model that says nothing about you is obvious. A model that describes you almost correctly can pass unnoticed until a buyer asks about a service you do not perform, or fails to ask about one you do. In export support, boundaries matter. VAT reporting is not the same as customs brokerage. Document preparation is not the same as legal trade advice. A directory will rarely protect those distinctions unless the business page gives it something better to borrow.
The company’s site may contain all the ingredients. The problem is arrangement. One paragraph says, “We support growing companies in their administrative needs.” Another says, “Our team understands cross-border trade.” A footer lists Nantes. A case note mentions intra-EU sales. None of those sentences alone can carry the answer. So the aggregator’s crude category fills the gap.
This is why I begin with the answer, the directory and the company page open together. I am looking for a sentence competition. Which sentence can the model quote with least risk? The directory often wins because it gives a compressed statement. The company page loses because it gives atmosphere.
Atmosphere is not useless. It can persuade a human visitor. But somewhere on the page, the atmosphere has to stop and the business has to name itself.
The company page needs a citation sentence, not a louder claim
When businesses notice that aggregators are being cited, the first instinct is sometimes to make the company page more assertive. Bigger headline. Stronger adjectives. More repetition of the category. That usually adds heat without adding evidence. The model does not need more confidence from the page. It needs a sentence it can carry.
For the accounting and export-support firm, I would search for a factual sentence like this: “The firm prepares French VAT, customs-documentation support and export paperwork for SMBs trading inside the EU from its Nantes-area office.” This sentence is not beautiful. Good. It names the firm type, service boundary, buyer, geography and document class. A human can understand it. A model can lift it.
The page should then support the sentence nearby. It can give examples of documents handled, describe what is outside scope, name the kinds of businesses served, and explain whether the work is advisory, preparatory, filing-related, or coordination with other specialists. These details stop the AI answer from stretching the claim.
A useful company page also makes the entity stable. It should use the business name consistently. It should connect the name to the service and location without forcing the reader through a hero slogan first. It should avoid making the English page clearer than the French page, a separate failure that deserves its own audit. For this topic, the key point is simpler: your page must give the answer engine a better source unit than the aggregator does.
I use the phrase source unit deliberately. A source unit is a sentence or tight paragraph that can travel outside the page while keeping the business facts intact. Most aggregators are made of crude source units. Many company pages are made of polished fragments. The fragments may sound better, but they travel badly.
The page can stay readable. It does not need to become a database card. The first visible service paragraph can do the hard work, then the rest can breathe.
Borrow the aggregator’s clarity, not its thinness
There is a temptation to imitate the directory because the directory is being cited. I would be careful. Directory writing is often lifeless and imprecise. The useful part is not the style. It is the fact order. Name before claim. Category before flourish. Location before regional vagueness. Services before “solutions.”
A company can borrow that order while remaining more accurate. For the Nantes firm, the page might open with one clean paragraph that states the operational truth, then move into buyer situations: an SMB selling goods inside the EU, a French company needing VAT and export-document support, a founder unsure which paperwork belongs to accounting and which belongs to customs. These are human situations, not directory labels. The source sentence anchors them.
The same page should also correct the aggregator silently. I do not usually advise writing “unlike directories, we…” because that sounds defensive and may age badly. Instead, state the boundary. “We prepare and coordinate export-documentation support; we do not act as a customs broker.” That line may feel too plain for a marketing page. It is exactly the kind of plainness that prevents a model from making the service grander than it is.
Evidence cues matter here. A named service boundary is stronger when paired with document examples. A location claim is stronger when paired with office, service-area or client-type details. A specialist claim is stronger when paired with a repeated category across the site. Reviews may support trust, but they cannot define the service by themselves. They are background noise unless the page carries the facts.
The aggregator can never know the business as well as the business knows itself. That advantage has to appear in sentences. Otherwise it stays private knowledge.
Make the page the source of record
A site becomes easier to cite when it chooses where the main business description lives. Too many French SMB sites spread the official description across homepage, about page, contact page, service pages, and downloadable PDFs. Each page has a little truth. No page feels like the source of record.
For aggregator displacement, the source-of-record page should be obvious. It may be the homepage for a very simple local business. It may be a service page for a specialised offer. It may be a location-service page if the business serves one market with one defined service. The important part is that other pages point back to it and do not contradict it.
In practical terms, I would place the strongest sentence near the top of the relevant service page, repeat a shorter consistent version on the homepage, and use internal links with descriptive anchors. The contact page can confirm the operating base. The about page can explain the practice history without changing the category. The English version, if present, should match the same service boundary rather than becoming a more precise shadow site.
Then I would compare the company page against the aggregator. Not emotionally. Sentence by sentence. Does the company page name the service more precisely? Does it give a clearer location? Does it define the customer type? Does it mention the evidence that the directory lacks? Does it avoid overclaiming? If the aggregator still has the cleaner answer after that comparison, the rewrite is unfinished.
A machine does not need your whole brand story to answer a query. It needs enough stable facts to describe you safely. That can feel reductive to a business owner. It is. But reductive does not have to mean false. The page’s job is to make the reduction accurate.
Citation is earned in the dullest sentence
The sentence that beats an aggregator is rarely the one a designer would enlarge. It is often the plainest line on the page. A line with the legal name or trading name. A line with the service boundary. A line with the city or region. A line with the client type. A line that admits what the business does not do. These sentences are not decorative, and because of that they are useful.
In my ledger, I mark the dull sentence first. If it is missing, the page is not ready. If it is present but buried, the page is vulnerable. If it exists only in English while the French page ranks, the AI answer may drift toward the wrong language source. If the directory’s dull sentence is clearer, the directory will keep showing up.
The aim is not to remove aggregators from the web. They will remain part of the source environment. Some are helpful. Some are lazy. Some are wrong in ways that look harmless until a model repeats them. A French SMB cannot control all of that, but it can make its own page harder to ignore.
Becoming easier to cite than an aggregator means giving the model less repair work than the aggregator does, while giving it more accurate facts. That is a narrow standard. It is also a practical one. The page does not have to shout. It has to hold still.
The Lift Note
Query: “ia cite annuaire plutôt site.” Liftable sentence: “An aggregator beats a company page in AI answers when its facts are easier to compress than the business’s own scattered proof.” Missing proof: one source-of-record paragraph naming the entity, service boundary, location, customer type and evidence. Rewrite instruction: place a plain citation sentence near the top of the company page, then support it with examples and limits the directory cannot provide.