The same four or five brands always appear in AI answers about your industry — and you're not one of them, even if you're bigger or technically superior. It's not a matter of budget or how many articles you publish: the AI doesn't know you belong to the same competitive market as those brands, so it doesn't consider you when a customer is deciding who to call. Those are exactly the most valuable queries — and you're not in them. Getting into the right cluster is possible, and it requires specific content that teaches the model the relationship.
Across 50 B2B industry queries I tested over the last 6 months, the AI always cites the same 4-5 brands. The other competitors in the industry, even if bigger or better, never show up. Here’s why.
The reason isn’t the quality of the content, it isn’t the domain’s DR, it isn’t even the content budget. It’s that the AI doesn’t know those other brands belong to the same “competitive family”. And if you’re not in the family, I don’t cite you when someone asks for “alternatives to X”.
Let me explain what this means, how to check whether your brand is in the right cluster, and what to do to get into it.
The competitive cluster is an entity, not a list
When you ask ChatGPT or Perplexity “who are the leading architecture and urban planning firms in Puglia”, the model doesn’t run a SERP. It builds a cluster: a group of entities it considers comparable in terms of geography, size, service offered and project type. Then it pulls out 4-5 to cite for you.
This cluster isn’t a static list. It’s a graph of relationships the model learns by observing how brands appear together across the web’s texts: in industry rankings, in comparisons, in comparative reviews, in tenders, in mutual references.
In the world of graph research, unsupervised clustering is an open problem:
“However, important unsupervised problems on graphs, such as graph clustering, have proved more resistant to advances in GNNs.”
Translated: grouping entities into coherent clusters without human labels is hard. The model has to infer on its own who goes with whom. And it does so using the signals it finds in the content.
The operational consequence for you is direct. If your firm never appears alongside the other firms in the cluster within the industry’s content — not cited with them, not compared with them, not listed with them — then as far as the model is concerned you’re not part of that cluster. Period.
Why this comes upstream of everything else
You can have the best entity record on Wikidata, the best Organization schema, the best “About us” page optimized for E-E-A-T applied to AI. If the model doesn’t place you in the right competitive family, you won’t appear in comparative queries.
And comparative queries are the ones that convert. “Best architecture firms in Bari”, “alternatives to [famous firm]”, “urban planning firms in Puglia specialized in urban regeneration”: these are shortlist queries. The customer is deciding who to call.
If your entity exists in the vector space but is far from the cluster of your true competitors, the model treats you as an outlier. It sees you, but it doesn’t include you. That’s worse than not existing: you exist out of context.
Pages that talk only about your own firm, without ever naming the competitive context.
The test you can run in 20 minutes
You don’t need a complicated tool. You need ChatGPT or Perplexity, a list of competitors, and an Excel sheet.
- Write down the 10 competitors you believe to be your direct rivals (the ones you fight for the same tenders, the same clients, the same budgets).
- Open ChatGPT or Perplexity and run 5 real comparative queries from your industry. Examples for an architecture firm in Puglia: “leading architecture firms in Bari”, “urban planning firms in Puglia specialized in urban regeneration”, “Puglia architects working on public construction”, “alternatives to [well-known local firm]”, “best designers for renovated masserie in Puglia”.
- For each query, note which brands get cited.
- Count: how many of your 10 assumed competitors show up? And do you show up?
If the brands cited by the AI are always the same 4-5 and don’t match your list, you have two valuable pieces of information. First: the cluster you think you inhabit isn’t the one the model recognizes. Second: you know exactly who you need to be seen together with to get into the family.
Write 2-3 pages of explicit comparison along the lines of “Our approach to urban regeneration vs [Firm X] vs [Firm Y]”.
The test I ran myself
Across 50 B2B industry queries monitored over the last 6 months — various geographic and vertical combinations, from professional services to manufacturing — the pattern was always the same: 4-5 dominant brands cited in 70-80% of the answers, and a long tail of brands that never show up, even when they objectively have revenues, portfolios and sizes comparable to the top ones.
The deciding factor wasn’t the quality of the website. It was the presence in comparative contexts: industry rankings, “top X in [area]” articles, comparison pages, mentions in panels and roundtables.
An indicative test, not a controlled study: the sample is qualitative, the measure is on aggregate citations. But the pattern is too sharp to be noise. Whoever is never named alongside the cluster leaders is, to the AI, not a competitor of those leaders.
The mistakes I see most often
Talking only about yourself. Pages that talk only about your own firm, without ever naming the competitive context. The model has no way to hook you to a cluster.
Generic comparisons. “We’re different from the other firms” without saying which ones. The AI needs proper names to build the competitive relationship.
Avoiding competitors’ names out of fear. The fear is commercial, not technical. But the cost is invisibility: if you never name your competitors in legitimate contexts (market reviews, industry analyses, reflection posts), the model doesn’t associate you with them.
Thinking link building is enough. Authoritative links help with authority, but they don’t build the competitive graph on their own. You need deliberate textual co-occurrences.
What to do in practice
- Write 2-3 pages of explicit comparison along the lines of “Our approach to urban regeneration vs [Firm X] vs [Firm Y]”. Honest, not disparaging, with clear criteria (methodology, project type, territory served).
- Publish 1-2 market articles per year in which you name the 5-10 players in your cluster. “Overview of architecture firms in Puglia in 2026” works well.
- Request to be included in industry rankings and panels where the other firms in the cluster appear.
- Use Author Entity Recognition to tie your designers as entities to concrete projects: the model associates people with firms with types of work.
- Check periodically with new comparative queries whether you’re entering the clusters you care about.
These actions aren’t magic and don’t produce results in two weeks. But over 3-6 months they start to shift how the model places you in the graph. Real analysis of your positioning in the cluster requires professional tools for monitoring AI answers and semantic analysis of industry content — what I’ve described to you is a first check to understand whether you have a problem, not a complete diagnosis.
Where your cluster ends, your AI visibility ends
The point I want to leave you with: your visibility in AI answers for comparative queries depends on one thing alone — being recognized as a member of the right cluster. Entity graph, schema markup, authority, all the rest comes after. If you’re not in the family, I don’t cite you when someone asks about its members.
In the next articles in this series I dig deeper into how to build relationships between entities in a structured way, how to use Wikidata to explicitly declare competitive relationships, and how to monitor the cluster’s evolution over time.
