You get two hundred mentions a year, but almost all of them come from the same three or four sites in the same circuit: as far as the AI is concerned, you're repeating the same weak signal two hundred times, not building authority. Five independent sources — a newspaper, a forum, a university, an industry directory, a social channel — are worth more than a hundred mentions from the same cluster. You're investing time and budget in a strategy that looks productive but doesn't move the needle that counts. Mapping the gaps takes thirty minutes.
I analyzed two brands with identical total mention counts (~200/year). One gets mentions from 15 different sources, the other from 3. The first is cited by AI 4x more often. Source diversity matters more than volume.
It’s a pattern I see recurring when I compare those who show up in the answers from ChatGPT, Perplexity and Gemini with those who stay out. Raw mention volume isn’t enough. What moves visibility is from how many different worlds those mentions come: news outlets, academic papers, vertical forums, industry directories, social mentions, independent blogs. The Citation Diversity Score measures exactly this.
In this article I explain why AI weighs citation diversity more than volume, how to map your sources in 30 minutes and what to do if you discover that 90% of your mentions come from the same type of source.
What the Citation Diversity Score measures for an AI model
The raw count is considered a poor indicator. You need a model that looks at the structure of the citation network, not just its number.
“A literature evaluation model centered on structural diversity is proposed, offering a new perspective on revealing the multidimensional value of literature within academic dissemination networks.”
Translated: instead of counting how many times a piece of content is cited, you look at the structural diversity of the network of those who cite it. Content cited by independent clusters of sources is worth far more than content cited a hundred times within the same cluster.
The operational consequence for your brand is direct. If Perplexity or Claude have to decide which sources to use to answer a query that concerns you, their training has learned to weigh the variety of citations as a reliability signal. A brand with 200 mentions all from guest posts on the same network is worth less than one with 50 mentions spread across a business outlet, a professional directory, an industry forum and a cited paper.
Why structural diversity matters more than volume
The mechanism isn’t an SEO invention. It comes from studies on social and academic networks that have already demonstrated the same thing in completely different contexts.
“Their research, based on data from the Facebook social network, demonstrated that the number of relatively independent groups within an individual local network, or the structural diversity of the neighborhood, is strongly correlated with the breadth and depth of influence the individual experiences in major decision-making processes.”
In practice: the influence of a node (person, brand, content) depends on how many groups that are independent of one another touch it. Ten friends who all know each other carry less influence than five people belonging to five distinct worlds.
For your business this means something specific. When an AI decides whether to cite your brand in answering “best administrative law firms in Basilicata,” it doesn’t just count how many times your name appears across the web. It counts from how many different types of source it comes. If all the mentions come from generalist directories, the signal is weak. If instead they come from a law journal, a forum of jurists, a ruling cited in an academic commentary and the Bar Council’s portal, the signal strengthens in a non-linear way.
I’ve discussed this in more general terms in the articles on backlinks as a citation proxy and on the weight of implicit references: source diversity is the next level of that reasoning.
A brand that buys 50 guest posts on the same network of SEO blogs has volume, but zero diversity.
The limit of pure counting
The paper warns about exactly this point.
“Secondly, these indicators are weak in capturing disciplinary diversity and academic innovation, which may lead scholars to focus excessively on citation counts while overlooking the long-term value and originality of their academic contributions.”
Translated into your context: optimizing for the number of mentions is shortsighted. A brand that buys 50 guest posts on the same network of SEO blogs has volume, but zero diversity. AI, which learned to recognize network patterns during training, tends to discard that kind of signal or weigh it lightly. It’s not a carved-in-stone rule, it’s a statistical tendency you can see in the models’ behavior.
Identify the 2-3 types of source you’re missing entirely (an industry forum? a specialist outlet? an academic presence?).
The observation I made about law firms
Over the last ten months I followed law firms spread across six Italian regions (Lombardy, Veneto, Emilia-Romagna, Lazio, Campania, Basilicata), comparing their visibility in the answers from ChatGPT, Perplexity and Gemini for queries like “administrative law firm
The pattern that emerged is clear, even though the sample remains small and should be taken as indicative, not as a study. The firms that were cited by AI had on average 8-12 different source types in their mention profile: a legal outlet, the Bar Council, legal directories, specialist forums, commentaries on rulings, conference participation, local interviews, scientific publications in a journal. Those invisible to AI often had 2-3 source types, almost always generalist directories and social pages.
One law firm in Potenza specialized in administrative law stuck with me. Web mention volume comparable to a Naples-based competitor, but a very different distribution: the Basilicata firm had citations from regional legal journals, commentaries on rulings of the Basilicata Regional Administrative Court, talks at conferences of the University of Basilicata, presence in administrative law directories. The Naples firm had almost only generalist directories and a couple of guest posts. On AI queries related to administrative law in southern Italy, ChatGPT and Perplexity cited the first, not the second.
An indicative test, not a controlled experiment. But the pattern repeats often enough to deserve attention.
The test you can run yourself in 30 minutes
You need to map your mentions by source type, not by number. Here’s an entry-level audit, aware that real analysis requires professional brand monitoring tools.
- Open Google Search Console and look at the “Links” report to understand who actually links to you.
- Run a Google search for “brand name” (in quotes) and scroll through the first 10 pages of results.
- Classify each mention into one of these categories: media/outlets, industry directories, forums/communities, academia/papers, social mentions, independent blogs, institutional sites, promotional guest posts.
- Count how many different types appear in your profile, not how many total mentions.
The binary threshold I use as a first indicator: if fewer than 4 different source types appear in your profile, you’re below the diversity threshold that AI tends to reward. If there are 5-7, you’re in the mid zone. Above 8, you’re in the strong zone. These are indicative thresholds derived from observation, not a proven formula.
To round out the picture you can also run the same queries for your sector in ChatGPT, Perplexity and Gemini and see which source types the AI actually cites for your 3-5 competitors. That’s your reference map.
The mistakes I see most often
When I run this audit for clients, four patterns recur.
Guest-post monoculture. 90% of mentions come from a network of SEO-oriented blogs. High volume, zero diversity. AI weighs it lightly because the network pattern is too homogeneous.
Generalist directories only. The brand is on Yellow Pages, Europages, Yelp and two or three vertical directories, but is absent from outlets, forums and academic environments. It’s the classic profile of someone who delegated PR to “list me everywhere.”
Everything on the founder’s socials. Lots of mentions, but all inside LinkedIn or Instagram. However active the founder is, it remains a single source type. You need to cross the boundary of social media.
Total absence from the academic or technical world. No citations in papers, commentaries on rulings, industry research, cited whitepapers. For sectors like legal, medical, engineering, the absence of this level is a weak signal.
Here’s what you can do concretely
If the audit reveals a poorly diversified profile, the goal for the next 6-12 months is to add source types, not volume within the types you already cover.
- Identify the 2-3 types of source you’re missing entirely (an industry forum? a specialist outlet? an academic presence?).
- For each missing type, find 3-5 concrete targets and build relationships, not a single one-off appearance.
- If you’re a professional firm, seek citations in commentaries on rulings, vertical journals, publications from professional bodies and associations.
- If you’re a B2B company, seek presence in industry papers, regional Confindustria research, vertical podcasts, practitioner forums.
- Don’t abandon the source types where you’re already strong, but stop adding volume there: the marginal return is low.
Every new type of source you add moves AI visibility more than ten extra mentions in a type you already cover. This is the point that research on structural diversity documents well.
