If your brand shares its name with another company, a professional or a place, the AI might be using your content to build someone else's reputation — and the clients looking for you find the wrong information. This is not a hypothetical scenario: just 15 minutes of testing across three AI engines is enough to find out whether it is happening to you right now. Making your profile unmistakable is a precise intervention that ensures every piece of content you produce works for you, not for the namesake.
Open ChatGPT right now and ask: “notary for transferring LLC shares in Lecce”. Then ask Perplexity the same thing. Then Gemini. Look at which names come up, which URLs are cited, and above all check whether the name of the firm is the same across the three engines — or whether one of the engines suggests a namesake, perhaps a firm with the same surname but in Brindisi, or a lawyer who happens to share a name with a notary.
This exercise is the visible part of a problem that, in my articles, I call identity dissolution: the AI has linked your brand to the wrong entity, and from that moment on every piece of information you produce ends up feeding someone else’s reputation. Here I explain why it happens, how to verify it in 15 minutes, and what to change on your site so you are recognized as a unique entity.
What “ambiguous entity” means for an AI engine
Before a model generates an answer about you, it has to do something seemingly trivial: decide which “Rossi Firm” you are. In the field of research on knowledge grounding for LLMs, this step has a precise name.
“This task is known as Entity Linking (EL), and it is a fundamental step in the transformation of unstructured text into structured knowledge.”
Translated: the engine takes an ambiguous string (“Rossi Firm”, “Notary Bianchi”, “Amato Dairy”) and links it to a unique entry in a structured knowledge base. If this link succeeds, the information about you accumulates under your entry. If it goes wrong, it accumulates under a namesake, and as far as the engine is concerned you become invisible.
The practical consequence is simple: the first filter for showing up in AI answers is not content quality, it is the unique identifiability of the subject. If the AI does not know who you are, the rest does not matter.
Why it is the first step, not the last
In the previous articles in this series I described how AI engines read text — they break it into tokens and project it into a vector space where nearby concepts end up close together. Disambiguation sits upstream of even these steps: if the wrong entity enters the vector, the entire rest of the pipeline works on the wrong subject.
In the field of research on entity linking applied to LLMs, this step is described as a foundational piece of information extraction.
“Entity disambiguation (ED), a crucial task in information extraction (IE), involves link- ing text fragments representing real-world entities to entries in structured knowledge bases like encyclopedias or dictionaries.”
It follows that if your site offers no clear signals for linking to the right entry — or worse, offers signals that lead to an entry other than yours — no subsequent optimization will save you. The AI will keep talking about someone else.
If another firm with your surname can copy and paste your About page without any changes, you are not disambiguated.
The reverse engineering test you can run in 15 minutes
You do not need a professional tool to find out whether you have a disambiguation problem. You need patience and a spreadsheet.
- Choose 5 queries that a realistic client of yours would make. For a notary firm in Lecce: “notary share transfer Lecce”, “property gift deed Salento”, “notary first-home mortgage Puglia”, and so on.
- Run each query on three engines: ChatGPT (with web search enabled), Perplexity, Gemini.
- Note in a table: the name of the firm cited, the source URL, the city the engine associates with the name.
- Cross-reference the data. If the same name shows up across multiple engines but with different cities or specializations, you have a disambiguation problem underway.
- Open Wikidata and search for your brand. If there is no entry, or if there is one but it is sparse, that is a strong indicator: you do not yet have a consolidated identity in the public knowledge graph.
Binary decision threshold: if across 15 answers (5 queries × 3 engines) you find even just 2-3 mix-ups with namesakes, you are in the red zone. That is not a normal margin.
Rewrite your About page with a unique description: what you do exactly, in what field, in what area, since when, with which specializations.
The test I ran myself
I tried this exercise on 12 queries of the type “best [profession] in [mid-sized Italian city]” — notaries, accountants, dental practices, architects. When I asked the same query to ChatGPT and Perplexity, out of 12 queries 7 returned a set of sources with one feature in common: the sources cited were not the firm’s official site, but industry directories (local yellow pages, professional registers, review portals). In the other 5 queries the official site did appear, but only when the firm’s name was already rare in itself (uncommon surnames, compound names).
Test limitations: small sample, queries I built myself, results dependent on the day of the query. It is an indicative observation, not a study. But the pattern is clear: when the name is ambiguous, the AI leans on the third party that has already done the disambiguation work for it — and that third party is a directory, not you.
Ding et al. compare different entity disambiguation systems across six benchmarks: performance varies considerably depending on the context. The consequence for your business is that AI engines are not infallible at disambiguation, and if you do not give them strong signals they stay with the statistically most probable choice — which is almost never your small business.
The mistakes I see most often
When I work on clients’ sites, four patterns recur in an almost identical way.
- Generic About page. “We are a professional firm with decades of experience.” Zero elements that distinguish you from another 2,000 Italian firms using the same phrase. To the AI, you are indistinguishable from your namesakes.
- Zero sameAs. The site does not link to LinkedIn profiles, Wikidata, the official professional register, the Google Business Profile. Without these bridges, the engine does not know that “Amato Notary Firm on LinkedIn” and “Amato Notary Firm on the website” are the same entity.
- No unique identifier. VAT number, tax code, professional register membership number, LEI code for those in finance: these are the strong handholds a graph uses so it does not confuse you. If you do not put them in the schema, the engine stays with the first and last name.
- Ambiguous office name. “Office in Lecce” with no street, no postal code, no coordinates. If there are other firms in Italy with the same surname in Salento, the engine has no way to tell you apart.
What to fix on the site (priority order)
- Rewrite your About page with a unique description: what you do exactly, in what field, in what area, since when, with which specializations. If another firm with your surname can copy and paste your About page without any changes, you are not disambiguated.
- Add the sameAs block in the Organization or ProfessionalService schema. Link all official profiles: company LinkedIn, Facebook page, Google Business Profile, the professional register if public, any Wikidata entry.
- Insert unique identifiers in the schema: `”identifier”` with the VAT number, the professional register code, the LEI code where applicable. These are the hooks a knowledge graph uses to assign you a dedicated entry.
- Verify the result with Google’s Rich Results Test: paste the homepage URL, check whether the Organization schema appears clean, whether sameAs has all the links, whether identifier is present.
- Open Google Business Profile: verify that name, address and phone number match exactly those on the site. Even minor discrepancies (Via vs V., S.r.l. vs SRL) feed the confusion.
These steps are an honest entry level. The real analysis of how Google’s knowledge graph and the AI engines see you requires professional tools and months of work, but if these four blocks are in order you are already in the minority of Italian small businesses with a readable identity.
The thread that holds it all together
Showing up in AI answers is not a matter of having written “optimized” articles: first and foremost it is a matter of being a recognizable entity. If the engine does not know which “Amato Firm” you are, every piece of content you produce works to the advantage of the ambiguity, not yours. Disambiguation is the move that makes everything else usable — E-E-A-T, author recognition, content structure. Without disambiguation upstream, you are building reputation for a namesake.
In the next articles in this series I will go into the detail of the operational building blocks: how to construct your brand’s entry in Google’s graph, how to map the relationships between your entity and the others in your sector, how to maintain all of this over time with a periodic audit.
