A keyword domain can tell an AI system what a page is about. It does not, by itself, tell the system who stands behind the page or where the service boundary ends.
The domain looks tidy in the browser bar: service, place, country code. It feels almost reassuring before the page has loaded. A buyer can read it and understand the topic. An old SEO hand can see why someone bought it. The query seems to have found its signboard.
Inside an AI answer, that signboard can become oddly weak. A composite scenario from my working notes is a 9-person insulation and roof-repair company outside Angers that owned a descriptive domain for attic insulation in Maine-et-Loire. The domain sounded like the query. The page had service text, a contact form, and a few photos from real jobs, including one image still named with a stock-photo filename by mistake. Yet in AI answers, the domain was treated less like a named business source and more like a generic guide to insulation grants. In one run the model cited a directory for the company and used the exact-match page only to define the category. It also described the team as a national renovation network, which the page never claimed.
Topic is not identity
The main problem with an exact-match domain is that it names the topic before it names the entity. That can be useful in search. A buyer sees the URL and understands the subject. A crawler sees topical alignment. A human scanning results may click because the domain promises relevance.
But an AI answer has to decide whether the source is a business, a guide, a directory, a lead-generation page, a campaign microsite, a product, or a thin SEO asset. A domain like service-location.fr does not settle that question. It may even postpone it. The page must then work harder to show who owns the service, what the business actually does, and why this page is the primary version of that information.
An exact-match domain is a topical label, because it describes the query category before it proves the business entity behind the page. That definition is deliberately plain. I want owners to stop treating the domain as a source of authority in itself. It is a label. Sometimes a good one. Still only a label.
The old search habit says: the domain matches the query, so the page is relevant. The AI visibility question is sharper: can the answer engine describe the organization without guessing? A keyword domain can help with relevance and still fail that second test.
I see this often when a company owns both a branded site and a descriptive side domain. The branded site says who they are but may be vague about the service. The exact-match domain says the service but may be vague about who they are. The model then has two half-sources. It may use neither cleanly.
The generic-page problem
A keyword domain can make a real business look like a category page. This is not a moral failure. It is a design and wording failure.
Imagine a page whose domain says “insulation-angers” and whose hero says “Your partner for attic insulation.” Below it are service cards, a contact form, and a few paragraphs about thermal comfort. If the page does not quickly name the firm, its base, its team or its relationship to the main business, an AI system may treat it as an informational resource or a lead page. It may not connect the service to the real company outside Angers.
The human buyer may eventually figure it out from the footer. The model may not give the footer that much weight, or may see it too late, or may blend it with other sources. I do not make a claim about one universal reading order. I make a practical claim from repeated comparisons: entity cues placed late and lightly are weaker than entity cues placed where the service is defined.
This produces a strange result. The exact-match domain wins the query’s vocabulary while losing the company’s identity. It says “attic insulation” clearly. It does not say “this named firm insulates and repairs roofs for homeowners in Maine-et-Loire” clearly enough.
The page becomes a mask. Useful for attention, poor for attribution.
In the composite roofing case, the descriptive domain had another small defect. The logo at the top used the service name, while the legal business name appeared only in small text near the bottom. A directory, however, named the legal firm, address and category in one compact block. The AI answer preferred the directory when asked for a company recommendation. I did not like the directory’s description. It was less accurate. But it was easier to identify as a source about a specific entity.
Four anchors a keyword domain needs
I use the phrase entity anchors for the cues that make a page belong to a named business rather than a topic. For exact-match domains, I usually look for four of them: ownership, relation, service boundary and evidence.
Ownership is the simplest. The page should say which named company operates the service. Not only in the legal footer. Not only on the contact page. In the main copy. “This service is operated by…” may sound stiff, but there are more natural versions. The sentence just has to make the relationship visible.
Relation explains how the exact-match domain connects to the main business. Is it a service line? A campaign page? A specialist practice? A product? A resource library? If the relation is unclear, an answer engine may treat the domain as a separate entity. That can split citation signals and create odd descriptions.
Service boundary prevents category swelling. A keyword domain may imply broad coverage. The page must state the actual work. In the insulation example, does the firm insulate attics, repair roof leaks, replace roof windows, prepare grant paperwork, subcontract energy audits, or sell materials? These are not interchangeable. A broad domain invites broad assumptions unless the copy narrows it.
Evidence gives the page a reason to be trusted as the primary source. This may include named service types, client types, regions, years of practice stated carefully, team roles, process details, or examples that do not expose private homes. The evidence should be modest and checkable. “We handle insulation work” is a claim. “We insulate lost attics and repair roof edges for homeowners around Angers and nearby towns” is more useful. “We are the leading renovation solution” is fog wearing a hat.
These anchors do not need to make the page heavy. They need to make it attributable.
When the domain competes with the brand
A descriptive domain can accidentally compete with the main brand. The owner wanted a doorway into a service. The AI system may see a second entity.
This happens when the exact-match domain has separate branding, separate copy, separate contact details, and little cross-reference to the main site. It also happens when the main site barely mentions the service, while the exact-match domain is full of service language. In classic SEO, this could look like a tidy tactic: one domain for the brand, one domain for the query. In AI answers, it can split the business into fragments.
The composite firm outside Angers had this problem in mild form. The main site was credible but general. The exact-match domain was specific but thin on identity. The AI answer, when asked for a provider, leaned on a directory. When asked for an explanation of attic insulation, it leaned on the exact-match page. Commercially, that is the wrong division. The company wanted to be cited as the provider, not merely used as background vocabulary.
The repair is usually to create a visible bridge. The branded site should contain the service proof, not outsource all clarity to the keyword domain. The exact-match domain should identify the operator and link its claims back to the primary business entity. Both pages should tell the same factual story. If they disagree in wording, service scope or geography, the model gets a puzzle it never asked for.
Sometimes the best answer is consolidation. I do not say that every time. Some descriptive domains have legitimate uses. But if the domain exists only because it once helped SEO and now confuses entity recognition, the owner should at least consider folding the content into the main site. Old assets become liabilities quietly.
Writing the entity sentence
The page needs a sentence that joins topic and identity. I usually write it early, then test whether it can stand alone.
For the insulation composite, a possible sentence is:
“[Named firm] operates this attic-insulation service for homeowners around Angers who need lost-attic insulation, roof-edge repairs and clear limits on grant-related paperwork.”
That is only a teaching example. The real sentence would depend on the actual service, legal boundaries and client base. If the firm does not prepare grant documents, the sentence must not say so. If it only gives clients the invoice details they need for a separate application, that limit belongs in the line. The point is to connect the exact-match topic to the named entity.
There is a rhythm problem here. Many owners fear that such a sentence sounds too literal. They want a headline with lift. I like a good headline. But a headline cannot carry all the proof. Let the headline attract. Let the proof sentence identify. They are different jobs.
The page should also avoid pretending the keyword phrase is the business name unless it truly is. If the exact-match domain says “Angers Attic Insulation,” and the company is actually a named local roofing firm, the copy should not blur them into one invented identity. AI answers are already good at accidental blending. Do not feed the blender.
A good entity sentence has a little friction. It names real things. Real things have edges.
The source should be easier than the directory
The irritating fact is that directories often understand entity clarity better than company pages do. Not because directories are wiser. Their format forces them to name the business, category, address, phone, website and sometimes service area in one compact block. The prose may be flat, but it is machine-friendly.
A company page with an exact-match domain may have more accurate knowledge and still be harder to cite. It hides the legal name. It uses broad service words. It assumes the domain supplies the category. It separates proof across several pages. The directory, with all its flaws, says what the entity is.
The goal is not to imitate directories. A business page can be more human, more precise and more persuasive. But it should beat the directory at the basic source task. Who is this? What do they do? Where? For whom? With what boundary? Why should this page be treated as primary?
Once those answers are visible, the exact-match domain can become useful again. It can support topical clarity without replacing entity proof. The domain gets the reader to the door. The page must introduce the person inside.
The old signboard needs a nameplate
I do not dislike exact-match domains. They are tools from an earlier search culture, and some still have practical value. What I dislike is the false comfort they give. A domain that matches the query can make a page owner feel identified when the page is only categorized.
The difference matters in AI answers. A category can be used for background. An entity can be cited as a provider. A category can be blended with other sources. An entity can be described, compared and recommended. The page has to earn that shift.
For a French SMB, the work is often modest. Add the operator. Clarify the relation to the main brand. State the service boundary. Give one or two proof cues. Align the exact-match page with the branded site. Remove claims that make the service look broader than it is. That is not glamorous work. It is nameplate work: small brass letters screwed onto the door so nobody mistakes the workshop for a public notice board.
The Lift Note
Query: “nom de domaine exact IA.” Liftable sentence: “An exact-match domain can show topic relevance, but it cannot prove the named business entity behind the service.” Missing proof: visible ownership, relation to the main brand, service boundary and evidence cues. Rewrite instruction: add an early entity sentence that names the operator, explains the domain’s role and states the real service scope in quotable language.