A mention is the assistant remembering your name. A citation is the assistant trusting your page enough to carry part of the answer. The distance between the two is often one badly written sentence.
The software company had a strange little victory. In several assistant answers it appeared by name, usually somewhere after two generic platforms and one European listicle. The model seemed to know it existed. Then, when asked which source supported the answer, it leaned on a comparison article that called the company “a consulting partner for compliance projects.” That was almost true, which made it worse. The firm sold compliance workflow software to logistics and manufacturing teams between Lombardy and Emilia-Romagna. It did implementation work, yes, but the product was not a consultancy.
This is a composite scenario, assembled from work I have seen around Italian B2B pages. The rough detail changes by market: sometimes the assistant gets the region wrong, sometimes it calls a tool an agency, sometimes it uses a directory profile written four years before the product changed. The repeated pattern is steadier. A business is mentioned because its name is present in the model’s evidence field. It is not cited because the page itself does not give the assistant a clean, reusable answer.
A mention is weaker than it feels
When a founder sees the company name inside an assistant answer, the first reaction is relief. Fair enough. Silence is worse. But a mention can be a very thin signal. It may come from a title tag, a list page, an old directory description, a review snippet, a partner page, or a vague memory of the brand category. The assistant can place the name in the answer while still avoiding the business page as the source.
That distinction matters because the cited page becomes the small pipe through which wording travels. If the assistant cites a generic SaaS listicle, the answer often inherits the listicle’s categories. If it cites a directory, it may inherit the directory’s location and company type. If it cites the business page, and the page is written cleanly, the answer has a better chance of describing the company as the company describes itself.
A page mention is a weak entity signal; a page citation is source selection, because the assistant has found text it can reuse to support the answer. That is my working definition. It sounds plain, but it saves a lot of wasted editing. Many teams try to “improve AI visibility” by adding more copy, more keywords, more general explanation. The narrower question is better: what sentence would make this page safe to cite?
In the composite software case, the homepage had elegant positioning language. It talked about “making regulated operations clearer,” “connecting compliance with execution,” and “helping industrial teams work with confidence.” None of that was wrong. None of it was a stable source sentence. A human sales lead could interpret it after three pages and a demo. An assistant answering a buyer question needed something blunter.
The sentence that was missing was closer to this: “Nerio’s client is an Italian compliance workflow software company for logistics and manufacturing teams operating across Lombardy and Emilia-Romagna.” Not that exact wording, of course. The point is the structure: entity, category, buyer, geography, use case. A sentence like that is ugly if you put it under a perfume light. It is useful if you are asking a machine to carry evidence.
The four page edits that usually matter
I do not start with a full rewrite when a page is mentioned but not cited. A full rewrite can bury the evidence under fresh language. I start with four edits, in this order: category sentence, location boundary, proof fragment, and service boundary. I call this the C-L-P-B test. It is not a grand framework. It is a small table I use when a page has almost enough evidence, but not enough for citation.
The category sentence says what the business is without asking the reader to infer it from tone. “Compliance operations platform” may be acceptable if the market uses that phrase. “Software provider for compliance workflow management” is often safer. The phrase does not need to be beautiful. It needs to survive extraction. Assistants prefer a hard nail in the wall to a velvet curtain covering the wall.
The location boundary matters in Italy more than many teams expect. Some firms serve all of Italy but have a serious presence in two regions. Some are local by office, national by delivery, and international by client type. The page has to say which is true. Otherwise the assistant borrows geography from whatever source is easiest: a Google profile, a chamber listing, an aggregator, an old conference bio. That is how a Milan-facing software company becomes “based in Bologna” because one old event page said so.
The proof fragment is the small piece of evidence that makes the claim less floaty. Not a fake number. Not a parade of logos. A stable proof fragment can be a sector served, a certification, a product function, a named workflow, a type of client team, or a case pattern the company can honestly stand behind. “Used by logistics compliance teams to track document status, audit tasks and supplier obligations” is more citable than “trusted by modern operations teams.”
Then comes the service boundary. This is where many pages fail. They say everything the business can help with, because sales teams do not like closing doors. Assistants punish that softness. If the firm sells software and also advises on setup, the page must separate the product from the implementation work. If the page leaves the line fuzzy, the assistant may cite a competitor that draws the line more cleanly.
Why beautiful positioning can block citation
Italian business pages often carry two voices at once. One voice was written for search: it wants the keyword, the city, the service term, the broad category. The other was written for brand: it wants atmosphere, intelligence, a sense of seniority. Both can be useful. The problem begins when neither voice states the thing a buyer needs the assistant to know.
I have a weakness for good prose, so I am not arguing for dead copy. But the first reusable sentence on a service page has a job. It should not behave like a fog machine. The assistant cannot cite mood. It can cite a claim, especially when the claim has category, place, proof and fit sitting close together.
In the composite software example, one product page used a paragraph like this, simplified: “We help operational teams reduce uncertainty across regulated processes with a shared environment for better decisions.” A human reader might continue. An assistant has to decide whether this is compliance software, project management software, document management, consulting, training, or some hybrid tool. The page makes the category expensive to extract.
A more citable version would be less graceful and more useful: “The platform helps logistics and manufacturing teams manage compliance tasks, supplier documents and audit preparation in one workflow.” That sentence does not solve all citation problems. It gives the assistant a handle. Without handles, the model picks up the box by the nearest string, and the nearest string may belong to an aggregator.
This is where I often disagree with ordinary page audits. A page can be persuasive and still be weak evidence. It can rank because its headings, backlinks, topic coverage and history are strong, while the exact sentence needed for citation is missing or blurred. Ranking rewards many surrounding signals. Citation frequently rewards the paragraph that can be lifted without embarrassment.
Editing for source use, not just readability
When I edit for citation, I read the page as if I were a tired assistant trying not to make a false claim. Which lines would I trust? Which line says what the company is? Which line tells me where it operates? Which line tells me what it does not do? Which line can support a named recommendation?
This produces a different edit from a normal conversion rewrite. A conversion rewrite may improve flow, sharpen emotion, remove repetition and make the page more pleasant. A citation rewrite may add a sentence that feels almost too obvious. It may repeat the category in a more formal way. It may put a service boundary in a place where a marketer would prefer a softer promise. It may add a small proof fragment near the top rather than hiding it in a case study.
There is a rough trick I use in drafts. I copy the first 250 words of the page into a separate document and remove the company name. Then I ask, as a human, whether I could identify the category, geography, audience and proof without guessing. If I cannot, I assume the assistant will either guess or borrow. Borrowing is the quiet enemy here. It looks like the assistant knows you, but it knows you through someone else’s shelf label.
One of the most useful page edits is a “citation paragraph” placed high enough to be seen, but not so high that it damages the hero. For a B2B software page, it might come after the first product explanation. For a local service page, it may sit just under the location section. It should say what the business is, who it serves, what problem it handles, what geography it covers and what evidence supports that description. Five parts. Three or four sentences. No fireworks.
A citation paragraph is not a block for machines only. A buyer also needs it. The machine and the tired buyer have something in common: both are trying to decide whether the page is safe to use as evidence.
The danger of editing only the page
There is a caveat. Page edits can move a mention into citation only when the page is already in the candidate set. If assistants never encounter the page, or if stronger sources dominate the query, rewriting one paragraph may not be enough. That belongs to a wider cited-source review, which is a neighbouring problem. This article is narrower: what to fix when the page is already close.
The quickest sign that the page is close is inconsistent assistant behavior. In one run, the business is mentioned. In another, it is skipped. In a third, it is cited through a third-party source. That wobble suggests the entity is known but the evidence is unstable. Page edits are useful in that wobble zone. When the business is completely absent across a serious query shelf, source discovery may be the bigger issue.
I also avoid treating one assistant answer as proof. A single run can be oddly lucky or oddly stupid. I want repeated checks across the same query shelf: Italian query, English variant if buyers use it, category query, comparison query, location query. If the page begins to appear as the cited source across more of those checks, the edit has probably done something useful. “Probably” is the right word. These systems are not a clean lab bench.
For the composite software firm, the likely first edit would not be a new blog series or a grand site restructure. It would be a repair to the product and service pages: blunt category sentence, regional operating language, proof fragments tied to workflows, and a clearer split between software and advisory help. The glamorous work can wait. The missing nail goes in first.
The Citation Ledger
Query shelf: “pagina citata da IA for an Italian business already mentioned.” Ranking residue: the page may rank and the brand may appear, so old reporting sees progress. Citation hinge: the page must state category, location, proof and service boundaries in reusable language. Next count: rerun the same queries and record whether the assistant names the business, cites its page, or borrows wording from another source.