AI knows you exist, but the number it gives your clients is wrong — or the street is the one from when you were in another office, or your name is written three different ways depending on the source. The client tries to reach you, fails, and moves on to the competitor. This is not a rare problem: most Italian SMEs have their contact details scattered across contradictory versions on the internet. Fixing everything for good takes a week of work, no more.
AI knows you exist but sends people to the wrong number. It happens when your name/address/phone are scattered across 5 different versions on the web.
It’s one of the most annoying symptoms I see with professional firms: ChatGPT or Perplexity cite your firm in response to “best accountant Salerno Amalfi coast”, but the address has two different street numbers, the area code is wrong, and the VAT number leads to an associated firm you shut down in 2021. The client looks for you, can’t find you, closes the window. You’ll never know.
In this piece I’ll explain why consistency of name, address and phone (NAP, Name-Address-Phone) is one of the least sexy but most decisive levers for coming out clean in AI answers, and how to check it in a single morning without enterprise tools.
What AI sees when it searches “accounting firm Salerno”
Generative models don’t “read” your Google listing the way a human does. They take signals from dozens of sources (Google Business Profile, Pagine Gialle, Bing Places, Apple Maps, category directories, professional association registers, company LinkedIn pages, citations on local sites) and try to stitch together a single entity: “Studio Rossi — accountants — Salerno”.
In the world of Knowledge Graph research, the mechanism behind this work is called entity resolution: the system compares identity attributes (legal name, address, phone, VAT number) and decides whether two records are the same thing or two different things. There are no LLM-specific papers on NAP consistency for Italian professional firms, but the adjacent principle is rock solid and you’ll find it in any entity linking work from the last ten years: the more attributes match character by character, the higher the confidence that the entity is a single one.
For your firm, this has a very practical consequence: if on Pagine Gialle you are “Studio Rossi & Associati — Via Roma 14” and on Google Business you are “Rossi Commercialisti Salerno — Via Roma 14/A”, the system can quite easily think you are two different firms. Neither one, at that point, has enough weight to be cited with confidence.
Why NAP comes upstream of everything else in this series
In previous articles I told you how AI engines build the representation of your brand starting from the meaning of tokens and placing it in a vector space where “accounting firm Salerno” and “tax consulting Amalfi coast” sit close together.
That semantic work, however, rests on a layer below: the entity must exist as a unique object. If the NAP is fragmented, the system can’t even bring into focus who you are, let alone give you authority in your sector.
That’s why in my work on the entity pillar this is almost always the first check, even before talking about schema markup or author entity recognition. The wall has to go up straight, otherwise everything you lean on it ends up tilting.
A single different postal code between two directories is enough to make the system doubt that the “Rossi Commercialisti” on Via Porto is the same as the one on Via del Porto.
The test you can run in 30 minutes
You need an Excel sheet with one column for each NAP attribute and one row for each source. The goal is to see the discrepancies with your own eyes.
Fill it in like this:
- Legal name: “Studio Associato Rossi & Partners” — trade name: “Rossi Commercialisti”. Pick one and use it everywhere, always written the same way.
- Address: street, street number, postal code, city, province. “Via Porto 14” ≠ “V. Porto 14” ≠ “Via Porto, 14/A” for an automated system.
- Phone: same format everywhere. +39 089 123456 ≠ 089/123456 ≠ 089 12 34 56.
The sources to check for a professional firm in Campania are these, in the order of importance I see working best:
- Google Business Profile — it’s the one most AI models cross-reference most often
- Bing Places — feeds Copilot and part of ChatGPT
- Apple Maps Business Connect
- Pagine Gialle and Pagine Bianche
- Register of the Order of Accountants of Salerno
- LinkedIn company page
- Industry directories (e.g. Commercialista Telematico, FiscoOggi listings)
- Spontaneous citations on sites like Confindustria Salerno, CCIAA, local blogs
For each row, copy and paste verbatim what you find. Don’t correct as you copy: the goal is to photograph the inconsistency, not to hide it.
The decision threshold is binary: either the cells in a column are all identical character by character, or there’s work to do. There’s no “almost equal” in entity resolution.
Choose a canonical version of the NAP and write it down in an internal Google Doc.
What I observed across 25+ professional firms over 4 months
Over the last four months I kept a group of 25+ Italian professional firms under observation — accountants, labor consultants, law firms — monitoring both the citations they received on ChatGPT and Perplexity and the state of their NAP on Google Business, Bing Places and Pagine Gialle.
The longitudinal pattern that emerged is fairly clear, and I’m sharing it with the usual caveats (small sample, observation over a short horizon, sectors homogeneous with one another, this is not an academic study):
- Firms with an identical NAP across at least 4 main sources received AI citations with a markedly higher frequency than those with a NAP fragmented across 3+ versions
- The most recurring problem was not the lack of presence, but the phone in different formats: 17 firms out of 25 had at least two phone formats in circulation
- In 6 cases out of 25 the AI model cited the “old” entity (the associated firm before a split) even when everything had already been updated on Google Business — a sign that historical citations on industry directories still carried weight
- After NAP cleanup work on 8 firms I followed directly, the correct citations started to stabilize within 6-10 weeks, not before
I’ll say it again: it’s a field observation, not a controlled experiment. But the pattern is consistent with the principle of entity resolution: the more the signals converge, the more the system recognizes you as a single, reliable entity.
The most common mistakes in this field
There are four patterns I keep finding when I open the folder of a new firm:
Double company name. The associated firm has one name, the web domain another, the Google listing a third. To the owner they’re the same thing. To AI they’re three disconnected entities.
Owner’s phone instead of the switchboard. Mr. Rossi’s mobile ends up on Google Business, the landline on Pagine Gialle, the reception number on LinkedIn. Three numbers, three possible entities.
Post-move address updated only halfway. The firm moved from the Vietri hamlet to central Salerno. Google Business is updated, Pagine Bianche isn’t, the Order’s register isn’t either. The system sees two addresses and doesn’t understand which is the right one.
Postal code wrong by one digit. It seems trivial. A single different postal code between two directories is enough to make the system doubt that the “Rossi Commercialisti” on Via Porto is the same as the one on Via del Porto.
What to do concretely this week
Three actions, in order:
- Choose a canonical version of the NAP and write it down in an internal Google Doc. Legal name + trade name, address in a standard format, phone with international prefix. From today, wherever the firm appears, you copy from there.
- Update the 8 main sources I listed above, starting from Google Business Profile and ending with the industry directories. Count on about 2-3 hours of work for 8 listings, if you have all the credentials.
- Compare with the 3-5 competing firms AI cites when you ask ChatGPT or Perplexity “best accountant Salerno” or “tax consulting Amalfi coast”. Open them, check their NAP, see if they converge. That’s your operational benchmark.
This is an honest entry-level audit: it serves to unlock the most frequent case. The full analysis of entity, knowledge graph and authority in your sector requires professional AI citation monitoring tools and a few months of work. But starting from the NAP is the fastest way to eliminate the noise at the root.
Where this work leads
Once the NAP is clean, the system stops seeing two or three versions of you and starts consolidating a single representation. At that point the next steps make sense: getting your name recognized with the E-E-A-T criteria, working on LocalBusiness schema markup, building citations consistent with the industry directories. These are all pieces that strengthen your visibility in AI answers, but they only work if the entity underneath is a single one.
In the next articles in this series I’ll explain how to build a LocalBusiness listing in JSON-LD schema without hurting yourself, why citations (mentions of your brand with a consistent NAP on other sites) weigh more than backlinks for local AI, and what happens when your firm has multiple offices: managing multi-location is a chapter of its own.
