Your brand has been cited a hundred times — but always on its own, never in the same sentence as the leaders of your industry. The AI knows you exist, but it doesn't know where to place you: you're a name without a context. AI models learn by association, and if you appear twenty times alongside the reference brands of your market, that association is worth more than a hundred isolated mentions. The difference between being recommended and not being recommended often comes down entirely to this. Getting into the "top X" articles of your industry is the most direct path.
Your brand mentioned 100 times on its own carries less weight than your brand mentioned 20 times alongside 3 well-known competitors in your industry. AI learns by co-occurrence, not by volume.
Keep that in mind while I explain what happens when a model reads thousands of articles in your industry and builds a mental map of who does what. It doesn’t matter how many times you appear: what matters is who you appear with, and how often those “who” are already names the model recognizes as authoritative.
In this series of articles on Digital PR for AI I’ve told you that citations on the web are the fuel with which models learn who you are. Today I’ll show you the next step: it’s not enough to be cited, you need to be cited in the same sentence, in the same paragraph, in the same list as the players the AI already considers references in your industry.
What a co-mention really is for a generative model
When you read an article in a trade magazine and come across the sentence “among the Italian art glassworks producing contemporary blown glass, Venini, Barovier and the Piedmontese [name] stand out”, to you it’s a trivial list. To an AI model it’s a learning event.
In the world of research on knowledge graphs and knowledge embeddings, the documented principle is clear: entities that appear close together in training texts are represented with vectors that are close together in semantic space. It follows that textual proximity to authoritative entities transfers authority. Not because there’s an explicit rule, but because the very geometry of the representations imposes it: you are defined by your neighbors.
Translated into practice: if your brand appears 100 times in articles where there’s no other well-known name in the industry, the model registers you as an isolated entity, hard to place. If you appear 20 times in articles where Venini, Seguso, Barovier sit next to you, the model learns that you belong to the same conceptual family. And when a user asks “best producers of Italian artistic blown glass”, you’re a candidate for the answer.
This dynamic is the same one I described to you when I talked about tokenization and backlinks as a citation proxy: the model doesn’t “see” your brand the way you see it, it sees it as a position in a network of relationships.
Think of it this way: if a friend tells you about a new restaurant in Alessandria and describes it on its own, you have nothing to grab onto to understand it. If they describe it by saying “it’s like that other place but with a different angle”, now you have a coordinate. The AI does the same thing, at a scale of billions of sentences.
Why this sits upstream of everything else in your PR
You can write press releases, get mentions in local outlets, get interviewed on niche podcasts. If you appear on your own in all of that content, you’re feeding the model a version of yourself disconnected from the competitive context.
The model doesn’t know on its own that you’re a blown glass producer. It infers it from the company it finds you in. Without co-mentions with recognized industry entities, you remain a floating entity: the model knows you exist, it doesn’t know where you exist on the market map.
It’s the same mechanism I described when talking about author entity recognition and entering Google’s Knowledge Graph: you need a relational anchor, not just presence.
There’s another reason this sits upstream: AI answers often take a “list” format. When a user asks “best X” or “main Y”, the model produces a list of 5-10 names. To get into that list, you must already have been seen by the model inside similar lists during training or during real-time retrieval. If you’ve never been in a list, it’s structurally much harder for you to be generated in one.
Getting an article entirely about you feels like a win, but if the piece doesn’t name at least 2-3 players in your market, the model has no relational anchors.
The A/B test I ran on two art glassworks
Let me tell you about a comparison I ran last week. I took two Italian art glassworks of similar size — let’s call them Glassworks A and Glassworks B — with comparable web citation volume (around 40-50 indexed mentions for each over the last 12 months, a number estimated with search operators, not an official metric).
Structural difference:
- Glassworks A appeared mostly in monographic articles: pieces dedicated solely to it, on the local chamber of commerce website, in Italian design blogs.
- Glassworks B appeared in list-based and comparative articles: “10 contemporary Italian art glassworks”, “between Murano and the rest of Italy”, pieces where Venini, Seguso, Salviati sat next to its name.
I then ran 15 queries on ChatGPT, Perplexity and Gemini (3 engines × 5 queries each), all variations of: “best Italian producers of contemporary blown glass”, “alternatives to Venini for art glass”, “Italian art glassworks outside of Murano”.
Result: Glassworks B was cited in response to 11 out of 15 queries. Glassworks A in 2 out of 15. With (roughly) the same total mention volume, the co-occurrence pattern produced a difference in visibility in AI answers of about 5 times.
Stated limitations: small sample, 15 queries are not a study, the mention estimate is approximate. But the pattern is consistent with the principle I laid out above, and I see it repeat across clients in different industries.
When you find a list in your industry where you’re not present, the operational takeaway is this: contact whoever wrote it with concrete data that justifies your inclusion (production figures, awards, collaborations with artists or institutions).
The mistakes I see most often
When I look at my clients’ PR strategy, these are the recurring patterns that destroy co-mention potential:
- Refusing comparisons. The client doesn’t want to be named alongside competitors because they “fear a direct comparison”. From an AI standpoint, on the contrary, that’s exactly what places them in the right category.
- Asking for monographic pieces. Getting an article entirely about you feels like a win, but if the piece doesn’t name at least 2-3 players in your market, the model has no relational anchors.
- Getting named only in disconnected local contexts. The press coverage from the provincial fair is fine, but if the only names next to yours are institutional bodies and no industry competitor, you’re feeding a wrong categorization.
- Not monitoring the industry’s list-based articles. “Top 10”, “best X”, “alternatives to Y” pieces are the densest opportunities for co-mention. If they come out every year and you’re never in the list, the AI learns your structural absence.
What you can concretely do this week
A 3-step audit, doable with free tools:
- Map the existing lists. On Google, search “best [your industry]”, “top [industry] Italy”, “alternatives to [industry leader]”. Save the first 20 results. Check whether you’re there.
- Check your historical co-mentions. Open Google Search Console, look at the queries that bring you traffic and cross-reference them with the content where you appear. Are they monographic or comparative pieces?
- Ask the AI engines. Open ChatGPT, Perplexity, Gemini and ask “who are the main Italian producers of [your category]”. If the answer contains a list of 5-8 names and you’re not in it, you have a documented co-mention gap.
Simple decision threshold: if you appear in fewer than 1 list-based article out of 10 for your category queries, the problem isn’t PR volume, it’s the type. You need to shift budget from monographs to comparatives.
This is an honest entry-level check: a serious analysis of your brand’s co-occurrence map requires professional entity analysis tools and logs of AI answers over time. But as a first step it already gives you a clear directional indication.
When you find a list in your industry where you’re not present, the operational takeaway is this: contact whoever wrote it with concrete data that justifies your inclusion (production figures, awards, collaborations with artists or institutions). Don’t ask to be added as a courtesy: offer the element that makes the piece more complete.
Where it fits into your AI visibility strategy
Co-mention with competitors is the bridge between the citations you earn and the categorization the model assigns to you. Without this bridge, even well-executed PR remains scattered traffic that doesn’t translate into mentions in AI answers.
In the following articles in this series I’ll show you how to actively build these co-occurrences through pitches targeted at industry journalists, how to read the weight of implicit citations, and how to leverage events and speaking to generate high-quality co-mentions, a topic I’ve already introduced when talking about speaking authority.
The common thread stays the same: showing up in AI answers isn’t a question of how many times you’re named, it’s a question of which digital neighborhood you’re named in.
