In a B2B market, a potential customer who asks AI which are the reliable suppliers in your supply chain doesn't see your website — they see the network of partners and companies you're connected to. If those ties aren't declared in a way that models understand, you're invisible even when you've been working for years with recognizable brands. Your real network is worth far more than what AI can currently see. Making it readable is the step that gets you into the answers where they matter.
A tier 2 automotive supplier in Friuli, between Udine and Pordenone, 12 million in revenue, zero AI mentions. When I asked ChatGPT and Perplexity “Italian automotive mold suppliers” the AI never named it — even though it had been working for years with two well-known northern Italian OEMs. Here’s what we changed and what happened after four months.
I’ll tell you the case in full because it’s one of the cleanest I’ve had the chance to follow on this topic: mapping supply chain relationships from the perspective of AI engines. In this series I’ve already explained how AI recognizes entities and how it builds knowledge graphs. Today we get into the piece almost no one takes care of: supplier-customer ties, declared partnerships, the supply chain network. For a Friulian tier 2 — a moldmaker, a precision component supplier for OEMs — it’s often the variable that tips the scales the most.
What it means to be a supply chain entity for an AI engine
When ChatGPT or Perplexity build an answer about a B2B sector, they don’t just read the page of the individual supplier. They read the network. Who works with whom, who supplies whom, which partnerships are publicly declared.
In the research world on this topic, the paper by AlMahri et al. (2024) “Enhancing Supply Chain Visibility with Knowledge Graphs and LLMs” documents a principle that is decisive for our work: supply chain visibility depends on the extent to which supply chain actors publicly share key information about their operational relationships, and this sharing must be built in a form readable by automated systems — not just by humans.
From this follows a direct consequence for anyone who wants to show up in AI answers: if the relationship “Friulian tier 2 X supplies OEM Y” is written nowhere in a machine-readable way — not on your site, not on the customer’s, not on Wikidata, not in a joint case study — for a language model that relationship does not exist. And if it doesn’t exist, your name doesn’t appear when a buyer asks “Italian automotive mold suppliers”.
Why the supply chain comes upstream of everything else
In the previous pieces of this series I insisted on one concept: AI builds semantic proximities in the embedding space. If you’re interested in the technical detail, I explained it in my article on the vector space of embeddings.
Proximity doesn’t come from nowhere. It comes from co-occurrences: names that appear together, repeatedly, in credible contexts. If your brand always appears alongside strong names in your supply chain, you inherit part of their reliability signal. If you appear in isolation, you start from zero every time.
A second academic work, Zheng & Brintrup (2024) “Supply Chain Visibility with Generative AI“, documents that generative models can infer supply chain relationships from scattered public data, but only when that data exists in structured or semi-structured form. From this it follows that for your business the rule is simple: if you want an AI engine to map you alongside the OEMs you work with, you have to make that relationship explicit in at least two or three independent sources. A single claim is not enough.
If your brand is never written next to the name of the district, AI won’t associate you with the territory.
The Friulian tier 2 case: what we measured first
I’ll start from the pre-intervention numbers, so the later comparison is honest. On the Friulian customer — molds and precision components for automotive OEMs — I ran a baseline with 18 queries on ChatGPT, Perplexity and Gemini, varying the phrasings:
- “automotive mold suppliers in Italy”
- “tier 2 moldmakers northern Italy automotive”
- “automotive component mold manufacturers Friuli”
- “who supplies molds to [customer OEM]”
Pre-intervention result: 0 mentions out of 18 queries, across all three engines. Let me state the limit: 18 queries are an indicative sample, not a study. But the signal was clear — zero mentions everywhere, while there were similar tier 2 competitors cited 2-3 times on the same queries.
On the web side the situation was typical: an institutional site with three service pages, a “customers” page with non-clickable OEM logos and no text, no Organization schema, no joint case study published, zero presence on Wikidata. The two OEMs, for their part, didn’t cite the Friulian supplier anywhere public.
Rewrite the partner page: from a logo gallery to a page with narrative text that explicitly declares sectors, years of collaboration, geographies.
What we changed in four months
The intervention was to explicitly build the relationship graph so it would be readable by an AI model. No magic, patient work on very concrete pieces:
- Partner page rewritten: from a mute logo gallery to a page with text that said “since 2017 we produce molds for [OEM A], since 2019 precision components for [OEM B]”, with years, declared indicative volumes and dedicated sections.
- Organization schema and relationships: adding schema markup that declared the sectors served and the geographic clusters of belonging (Friulian mechanical district). Verifiable with the Google Rich Results Test.
- Three joint case studies: two published on the supplier’s site (with written permission from the customers), one picked up by the customer itself in an industry news piece.
- Wikidata entry created for the supplier, with “industry” and “location” properties, linked from the site’s pages.
- Interview in two industry outlets in which the OEM names were cited naturally.
None of these actions are exotic. It’s supply chain hygiene done with how an AI model will read the network in mind.
What happened after four months
Same battery of 18 queries, same engines, same method. Post-intervention result: 7 mentions out of 18 total queries, with the supplier’s name present 3 times on ChatGPT, 3 times on Perplexity, 1 time on Gemini. In two cases on Perplexity it appeared with a direct citation of the published case study.
Intellectual honesty: I can’t 100% isolate the effect of the entity-relationship work from other factors. In the same months the customer published three technical articles on its own site. But the pattern is clear: the AI mentions arrive precisely on the queries where the relationship with the OEMs is declared. Generic queries on “Italian molds” without a supply chain tie still remain poorly covered.
I’ll state the entry-level check right away: 18 queries are a first step, real analysis requires professional tools that track AI citations across large volumes and continuously over time.
The mistakes I see most often on Italian supply chains
Working with various tier 2 and tier 3 manufacturers, the patterns I find are always the same:
- A mute “our customers” page: only logos, no text, no context. To an AI those logos are invisible.
- NDAs used as an excuse to say nothing: often the customer would allow citing the relationship in a generic way (“we’ve been working for X years with a German tier 1 automotive group”). No one asks them.
- Case studies written only by the end customer: the supplier never picks them up, never links them, never comments on them on its own site. The signal is lost.
- Zero presence on the district name: the Friulian mechanical district, the Piedmontese coachbuilding hub, the Lombard moldmaker cluster are strong geographic entities. If your brand is never written next to the name of the district, AI won’t associate you with the territory.
What to do concretely, in order of priority
If you run an Italian B2B supply chain company, this is the minimum path:
- Open the Google Rich Results Test on your homepage and check whether you have a valid Organization schema. If it’s missing, that’s the first thing to fix.
- Rewrite the partner page: from a logo gallery to a page with narrative text that explicitly declares sectors, years of collaboration, geographies.
- Ask 2-3 strategic customers for written permission to publish a joint case study. It doesn’t need to be long: 600 words with indicative numbers are enough.
- Create a basic Wikidata entry for your brand, with industry and location, and link it from the site.
- Compare your positioning with 3-5 supply chain competitors that AI already cites in your sector. Look at what they have that you don’t in terms of declared partnerships.
The thread with visibility in AI answers
Mapping supply chain relationships is not a magic factor. It won’t get you cited by ChatGPT tomorrow if today your site has three pages and no case study. But it’s the lever that, for a B2B supply chain business, moves positioning the most in the knowledge graph of AI engines.
The reason is simple: your visibility in AI answers isn’t built only on your pages, it’s built on the network of entities you’re connected to. For a tier 2 manufacturer, partners are the most underestimated capital.
In the next articles of this series I dig into two related topics: how to declare temporal relationships between entities (timelines, histories, partnership evolutions) and how co-citations between brands in the same sector influence AI answers. If you haven’t read them yet, I’d also start from E-E-A-T for AI and from Author Entity Recognition, because the principle of relationships applies to the people behind the brand too.
