A service page may look full because the target phrase appears everywhere. To an answer engine, that same page can feel almost empty if no sentence states who does what, where, with which evidence.
The page had the phrase in the title, the H1, two subheadings, the first paragraph, the image alt text and the final call to action. It had worked for years. A tidy French service page, nothing obviously shameful. Yet when I put the same service query into an answer engine, the company disappeared. The model described two competitors and a national directory. It did not even get close enough to be wrong.
The composite scenario is familiar: a 28-person accounting and export-documentation firm near Nantes ranks for local accounting and VAT searches. The page repeats the right words: expert-comptable Nantes, accompagnement TVA, conseil entreprise. But the important part, the export-documentation work for French SMBs trading inside the EU, appears only as a soft phrase near the bottom. “Support for your international growth.” Nice enough for a human skimming fast. Useless as proof.
The page is loud, yet the claim is quiet
Keyword repetition has a strange effect on the eye. It makes a page look worked on. Someone has touched the title, the intro, the block headings, the internal links. There is visible effort. For a human buyer, that effort may be enough to signal relevance. For a machine trying to describe the business, repetition without definition is like receiving the same blurred photocopy five times.
In older SEO work, I often accepted a certain amount of this. Not because it was beautiful, but because it helped a page sit in a query category. The phrase marked the shelf where the page belonged. A service page for local tax support had to say local tax support. A page for export paperwork had to say export paperwork. Fine.
The problem begins when the phrase becomes the only real content.
A keyword-stuffed service page is a page where repeated search phrases replace factual service statements, because the copy signals topic without defining the business action. That is the working definition I use in audits. It is not a moral judgment. Some keyword-heavy pages were written by careful people under the rules they had at the time. But inside AI answers, those rules are less forgiving.
The answer engine does not only ask, “Is this page about the query?” It also has to answer, “Can I safely reuse this sentence without inventing the missing part?” If the page says “our accounting firm supports companies with their international ambitions,” the model must guess the boundary. Does that mean VAT? Customs forms? Payroll for foreign staff? EU sales declarations? Advice only? Filing too?
A keyword can place the page in the room. It cannot testify for the business.
Three kinds of empty repetition
When I mark a page ledger, I usually see three kinds of repetition that look useful until I reduce them to factual sentences. I call them label repetition, promise repetition and place repetition. The names are plain because the problem is plain.
Label repetition is the repeated phrase that names the category but never opens it. “Accounting services in Nantes.” “Business accounting in Nantes.” “Your accounting partner in Nantes.” The label changes coat but keeps the same bones. There is still no sentence saying which accounting tasks are handled, for which business type, and where the firm’s competence stops.
Promise repetition is softer and more dangerous. “We help you grow.” “We simplify your administration.” “We support your projects.” These sentences feel buyer-friendly. They may even be true. Yet they cannot be cited as evidence, because they do not contain an extractable fact. An answer engine can paraphrase them into almost anything, so it often prefers a directory line or competitor page that names the service more bluntly.
Place repetition is the local SEO habit of naming a city again and again without explaining the actual service area. “Near Nantes,” “in Nantes,” “for Nantes businesses,” “your Nantes accounting firm.” In the composite accounting example, this gave the page local colour but left the export work floating. The model could see Nantes. It could see accounting. It could not see “French SMBs trading inside the EU” as a clean service boundary.
The three repetitions create a page that appears dense from far away. Up close, it has holes.
A useful service page does not merely repeat the query; it gives the model a sentence it can carry away intact. That sentence needs an entity, a service, a buyer or use case, a location or market boundary, and some cue that the claim belongs to this business rather than the whole category.
The rough part is that this often makes the first rewrite feel less elegant. “The firm prepares French VAT, customs and export-documentation support for SMBs trading inside the EU” is not perfume copy. It is a working sentence. It has elbows. It can be lifted.
What answer engines do with vague service copy
I do not pretend to know the private internals of every answer engine. I can only work from repeated runs, source comparisons and the behaviour visible on the page. The pattern is steady enough to be useful. When a service page is vague, the model looks for a denser substitute.
That substitute may be an aggregator. It may be a competitor’s service page. It may be an English summary, especially when the French page is elegant but thin. It may be a Google Business Profile fragment mixed with a directory category. The answer that comes out can sound reasonable while quietly removing the company’s real specialty.
In the Nantes accounting composite, the AI answer did not deny that the firm existed. It simply failed to describe the export-documentation work. One run called the firm “a local accounting office for small companies.” Another mentioned VAT in a generic list, then cited a directory that did not know the firm’s export practice. In one awkward run, the model placed the firm in the right region but implied that customs paperwork was handled by a partner. That was not on the page. The model had stitched around the absence.
This is why I distrust pages that “say the right thing somewhere.” Somewhere is not a source strategy. A sentence buried in a testimonial carousel or hinted in a hero slogan cannot carry much weight. Answer engines tend to prefer stable, nearby, explicit statements. The page has to make the important fact easy to extract, not merely possible to infer.
Classic SEO tolerated implication more often. A human could land on the page, click the navigation, read the about text, infer the offer from context, maybe call the firm. A generated answer has to compress that path into a few sentences. When the page refuses to name the work, the model borrows language from elsewhere.
The business owner then says, “But we do that.” I believe them. The page does not.
Rewriting without turning the page into a database
The temptation, after seeing this, is to overcorrect. Some pages become stiff little inventories. Every sentence carries a noun, a service, a location, a credential. Nobody wants to read them. A service page still has to sound like a business speaking to a buyer, not like a customs form left in the rain.
The better move is to place a few factual sentences where they can do the most work. I usually begin with the first service paragraph, one mid-page proof block, and one section heading. Those are the places where an answer engine and a human reader both expect clarification.
A weak paragraph says: “We support companies of all sizes with accounting, tax and administrative needs, offering tailored advice for every stage of development.”
A stronger paragraph says: “The firm prepares French accounting, VAT filings and export-documentation support for SMBs near Nantes that trade with customers inside the EU.”
The second sentence is still human. It is just less evasive. It names the entity type, the tasks, the location and the customer situation. It does not claim to be the best. It does not wave at “international growth.” It says enough.
Then I look for what I call the proof hinge. This is the point where a broad claim turns into something that can be checked. For the accounting firm, a proof hinge might mention the documents handled: French VAT returns, EU sales listings, customs-support paperwork, invoice evidence for cross-border trade. For another business, it might be equipment models, service intervals, regulated forms, named departments, or measurable service limits.
A proof hinge is the sentence where a service claim stops floating and attaches to visible work. It does not need to reveal private client data. It does need to give the claim a surface.
The page can still have softer copy around it. Warmth is allowed. Rhythm is allowed. The problem is not style; the problem is absence. One clean factual sentence can make the surrounding prose safer because it gives the model a centre of gravity.
The sentence ledger test
My ledger is dull by design. I take each important sentence and ask three questions. What fact does it state? What source cue supports it? What extraction risk does it create?
Take “We support companies with their international development.” The fact is unclear. The source cue is absent. The extraction risk is high, because a model may turn “support” into legal, tax, customs, translation, logistics or strategic advice. The sentence may stay on the page as atmosphere, but it cannot be the service proof.
Now take “The firm prepares French VAT and export-documentation support for SMBs trading inside the EU.” The fact is clear. The source cue can be strengthened by nearby examples of documents handled. The extraction risk is lower, though I would still check whether “support” means preparation, review or advisory only. This is where the rewrite becomes exact.
Sometimes the best fix is a verb. “Supports” becomes “prepares,” “reviews,” “files,” “maintains,” “installs,” “repairs,” “documents,” “audits.” Verbs are small machines. They tell the answer engine what action belongs to the business.
Sometimes the fix is a boundary. “International” becomes “inside the EU.” “Businesses” becomes “French SMBs.” “Administration” becomes “VAT filings and export-documentation records.” Boundaries make a page less grand but more citable.
And sometimes the fix is removing a keyword. If the phrase has appeared six times, the seventh mention is not helping. It is only taking space from the sentence that should have been there.
A page that repeats the query may earn attention; a page that defines the work earns a safer citation.
The human buyer still matters
There is a small fear I hear from business owners when I recommend these rewrites. They worry the page will become too plain. French business copy often likes a certain polish: phrases about proximity, confidence, expertise, responsiveness. Some of that language belongs on the page. Buyers do not make decisions from extracted sentences alone.
But the factual line does not insult the buyer. It respects their time. A founder trying to understand whether an accounting firm can help with EU trade does not need six versions of “personalised support.” They need to know whether the firm handles the paperwork they actually have on the desk.
The same sentence can serve both audiences. The human sees clarity. The answer engine sees a quotable claim. The company gets described with less guesswork.
That is the shift I care about. Not more copy. Not a new keyword list. A better sentence.
The Lift Note
Query: “page service trop optimisée.” Liftable sentence: “A keyword-stuffed service page signals topic without defining the business action, service boundary or evidence.” Missing proof: a factual sentence naming the entity, service, buyer situation and limit. Rewrite instruction: keep the main keyword once, then replace repeated promises with one service sentence, one proof hinge and one boundary sentence.