The cited source is not just a footnote. It is the assistant showing which shelf it trusted. Read that shelf closely enough and you can see why the competitor’s name traveled farther than yours.
The answer named another firm. That was the part everyone noticed. A 28-person software company in northern Italy had asked why assistants kept recommending a competitor for compliance workflow questions, even though its own pages ranked well and its product was a better fit in several sectors. The first instinct was to compare websites. The useful move was slower: open the cited sources and read them like a customs officer reading labels on a crate.
This is a composite scenario, built from several B2B citation checks. The model would sometimes cite a generic SaaS list, sometimes a directory entry, sometimes the competitor’s own page. One answer got the competitor’s customer segment right but placed its strongest region too broadly. Another cited a listicle that was thin, almost insulting in its simplicity, but it contained one sentence the client’s page did not: a clean statement connecting category, industry and compliance workflow.
The citation is a map, not an endorsement
People often treat an assistant citation as a quality prize. The named competitor must have better content, better SEO, better authority, better everything. Sometimes that is true. Often the cited source simply gave the assistant a safer sentence for the question. That is less dramatic and more useful.
A cited source is the piece of evidence an assistant exposes to support a claim, because that source appears usable for naming, describing or comparing the business. This definition matters because it keeps us from over-reading the result. A citation does not prove the competitor is better. It proves the assistant found a source it could use.
When I read competitor citations, I am not admiring them. I am looking for the hinge. What claim did the assistant need to make? Which source allowed that claim? What did that source state plainly that the skipped business left scattered, implied or contradicted? The hinge is usually small. A category phrase. A sector line. A location boundary. A proof fragment. A table row. A paragraph on an aggregator that says the obvious without shame.
In the composite software case, the client wanted to know why a competitor was cited for “Italian compliance workflow software for logistics companies.” The competitor’s own site was not dramatically better. But one page used the exact category structure: software, compliance workflow, logistics, Italian operations. The client’s page talked about “operational assurance across regulated environments.” That phrase may impress a committee. It does not help an assistant answer the buyer’s question.
This is the emotional difficulty of citation work. The page that wins may look less intelligent than the page that loses. Assistants are not grading literary confidence. They are assembling supported answers.
Read the source in layers
I read a cited source in layers, because otherwise the eye gets lazy. First I look at the sentence or paragraph that likely supported the assistant’s wording. Then I look at the page structure around it. Then I look at the source type. Finally I compare the cited wording with the business page that lost. Each layer answers a different question.
The sentence layer asks: what exact reusable claim is present here? If the assistant described the competitor as “a compliance workflow platform for logistics firms,” I want to see whether the source says something close to that. If it does, the gap is probably page wording. If it does not, the assistant may be synthesizing from multiple sources, and the visible citation is only part of the trail.
The structure layer asks whether the claim is easy to find. A clear heading, a short category paragraph, a comparison table, a branch block, a review summary, a schema-like listing: these can all make a claim easier to lift. I am not saying the assistant reads like a human scanning a page. I am saying extractable structure tends to make evidence less slippery.
The source-type layer asks why this source was trusted for this query. A directory may win on category coverage. An aggregator may win on comparison language. A competitor’s own page may win on service boundaries. A public profile may win on location. A review source may win on proof. The assistant is not using every source for the same job.
The comparison layer is the one teams skip because it hurts. Put the cited source beside your page. Remove the brand names. Ask which page gives a cleaner answer to the query. Not which page is nicer. Not which company is better. Which source gives the assistant fewer chances to be wrong?
The five evidence gaps I keep seeing
Across citation checks, I find the same gaps often enough that I gave them names. The names are mostly for my own notebooks, but they help clients stop arguing in fog. I call the set the Five Source Gaps: category gap, place gap, proof gap, boundary gap and freshness gap.
A category gap appears when the cited competitor is easier to classify. This is common in Italian B2B technology. The losing company writes in capability language; the cited source writes in category language. “Helps teams coordinate obligations” may be true. “Compliance workflow software for logistics and manufacturing” is easier to cite.
A place gap appears when geography is clearer in the cited source. Italy makes this subtle. A business may be legally based in one city, operationally strong in two regions, and selling nationally. If the page does not separate those layers, a source that simply states “serving Lombardy and Emilia-Romagna” can become more useful than the official site.
A proof gap appears when the cited source gives a small support beam. It might name sectors, workflows, certifications, implementation types, client roles, or use cases. The proof does not need to be loud. It needs to sit near the claim. Proof hidden deep in a case study often fails to support the category sentence at the top.
A boundary gap appears when the competitor’s source says who the product is for and where it stops. This is especially important when tools, consultancies and service providers blur together. If the cited competitor says it provides software with onboarding, while the losing page talks about “supporting compliance maturity,” the assistant may choose the clearer boundary.
A freshness gap is trickier. I do not mean the newest blog post wins. I mean the cited source reflects the current business more cleanly than the losing page or public profile. A page can be newly redesigned and still carry old claims. An old directory can be stale but have one current category. Freshness is about claim accuracy, not date cosmetics.
Do not copy the competitor’s sentence
Once the gap is visible, the temptation is to mimic the cited source. I understand it. The competitor has the sentence, the assistant used it, so write a similar sentence. This can work at the level of structure, but it fails when the business borrows the competitor’s actual positioning. Then the market gets flatter and the assistant has less reason to distinguish anyone.
The better move is to copy the evidence function, not the words. If the cited source wins because it connects software, compliance workflow and logistics, your page should also make its own category-sector link. But it should use your real boundaries. Perhaps your firm is stronger in manufacturing audits than logistics documents. Perhaps your software includes supplier obligation tracking, while the competitor focuses on internal task management. Those differences need to be stated, not polished away.
In the composite software example, a useful rewrite would not say, “We are the leading compliance workflow platform for logistics.” That is both generic and suspicious. A better evidence sentence might say, if true, “The platform helps logistics and manufacturing teams in northern Italy track compliance tasks, supplier documents and audit preparation in a shared workflow.” It is longer, rougher, and much more useful.
Citation repair is often a subtraction exercise. Remove the claims that could describe five competitors. Remove the sector words that are present only for search. Remove the elegant paragraph that refuses to say whether the company sells software, consulting or both. What remains may feel plainer. Plain can be a kindness.
There is also a legal and ethical edge here. Do not write evidence the business cannot support. Assistants already misdescribe companies often enough without us feeding them invented proof. If the source material is weak, the honest finding is that the competitor deserves the citation for that query until the business can state and support its fit.
When the cited source is an aggregator
Aggregators deserve special attention because they can feel unfair. A list page with twelve companies, thin descriptions and a few borrowed phrases may out-cite the business page that actually knows the service. This annoys people. It annoys me too. But annoyance is not a measurement method.
Aggregators often win because they match the shape of assistant answers. They compare. They classify. They put several names near the same category. They use simple descriptive sentences. They repeat place and service boundaries. They also tend to accumulate external signals that make them look safe enough for broad queries. Thin, yes. Useless, no.
When an aggregator is cited behind a competitor, I ask two questions. First, what does the aggregator say that the business page does not say? Second, why did the assistant need an aggregator-shaped source for this query? If the query asks for “best tools” or “top providers,” the assistant may prefer a comparative source. If the query asks whether one named company fits a specific use case, the business page has a better chance to win if it is clear enough.
The business cannot always replace the aggregator for broad list queries. It can close the evidence gap for narrow fit queries. That distinction matters. Trying to make your own page win every broad comparison is usually a long road. Making your own page citable for your category, region, sector and use case is more practical.
Sometimes the repair includes public profiles outside the website. If the aggregator describes you wrongly, and that description is copied elsewhere, the business page alone may not clean the trail. But I still start with the owned page, because it should be the strongest statement of record. If the official page will not say the obvious, why should the assistant trust it over the crowd?
Turn the source reading into a page change
A source reading is only useful if it becomes a decision. I do not like reports that end with “competitor has stronger AI visibility.” That is a weather report without a coat. The decision should be specific: add a category sentence to the product page, separate software from consulting, create a sector paragraph for logistics, update the directory profile, add proof near the service claim, or test a narrower query shelf.
For each cited source, I write one line in a ledger: query, named business, cited source, source type, key claim, missing evidence on our page, next edit. That line is small enough to repeat. It also prevents the team from arguing about the whole internet at once. The cited source becomes a specimen under glass.
The page change should then be checked against the same query, not celebrated immediately. Assistant answers vary. One clean result is not a verdict. I want to see whether the page begins to appear as a cited source across repeated checks, or whether the description improves even when another source remains cited. Description accuracy is sometimes the first sign that the evidence trail has changed.
For the composite software company, the likely first fixes would be product-page evidence, sector-specific wording and public description alignment. Not a new slogan. Not a hundred blog posts. The competitor’s cited source had shown the missing shelf. The company needed to build its own version of that shelf, with truer labels.
The quiet pleasure of this work is that it turns a frustrating answer into a reading task. The assistant names the other firm. Fine. Open the source. Find the hinge. Decide whether the gap is real. Then write the sentence your page should have carried all along.
The Citation Ledger
Query shelf: “fonti citate IA concorrenti for a competitor named instead of us.” Ranking residue: old reports still show your page position and may miss the source that shaped the answer. Citation hinge: the cited source reveals the category, place, proof, boundary or freshness gap the assistant trusted. Next count: log the key claim from each cited source, match it against your page, then rerun the same query after one evidence edit.