The first audit does not need a tool dashboard. It needs five careful questions, a clean page, and the patience to notice which local fact disappears.
I keep an old paper ledger for this work. It is not romantic. It has crossed-out business names, crooked arrows, and pages where the ink has rubbed against the next sheet. One line might say: “direct boat operator called ticket site.” Another says: “sfogliatella maker treated as cafe.” The useful part is never the mistake by itself. The useful part is the missing phrase that would have stopped the mistake.
A Naples business owner can begin the same way without hiring anyone. Open an AI assistant, ask what it says about the business, and write down the answer without arguing with it. Then ask again in a different wording. Then in Italian. Then as a traveller. Then as someone with a practical constraint: “near the port,” “for a family,” “without reservation,” “direct to Capri.” The pattern that appears across those answers is the beginning of the audit.
Start with the answer, not the page
Most owners start by rereading their own site. I understand the instinct. The site is the thing they can control. But a self-audit should begin outside the page, with what the machine already repeats. If the answer says the business is a “tour company,” while the owner sees themselves as a licensed ferry or excursion operator, the gap is visible. If the answer names the right business but cites an aggregator’s language, that is a different gap. Do not repair before you know which crack you are repairing.
For a composite bay operator I use when teaching this method, the business ran two small boats with direct routes and private excursions to Capri, Ischia and Procida. It had a booking office near the port and a seasonal crew. The owner’s page was not empty. It had photographs of the boats, a route page, and many words about the beauty of the islands. In AI answers, though, the business often appeared as a “ticket provider,” sometimes beside large booking platforms. Once the model gave the correct name but described it as a reseller. That little wrong noun mattered.
The first question I would ask is plain: “What is [business name] in Naples?” Then I would ask: “Is [business name] a direct operator or a ticket platform?” Then: “Who runs boat trips from Naples to Capri directly?” Notice the shift. The first question checks the machine’s loose memory. The second checks the exact confusion. The third checks whether the business appears when the user does not already know the name.
I call this the three-door test: name door, confusion door, category door. A self-audit is the act of comparing what AI says when it enters through each door, because each door exposes a different weakness in the page evidence.
Write down the wrong noun
A self-audit fails when the owner writes “AI got it wrong” and stops. That note is too large to act on. The better note is smaller: wrong noun, missing role, borrowed phrase, weak address, absent condition. This is the ledger habit. It reduces irritation to a repairable sentence.
Wrong nouns are especially useful. Restaurant. Cafe. Souvenir shop. Ticket site. Tour agency. Clothing shop. These labels may be close enough for a hurried traveller and still wrong for citation. A ferry operator hidden behind ticket aggregator pages does not only lose a sale. It loses its role. The machine learns that the clearest public wording belongs to the platform, not the operator.
In the bay-operator composite, the wrong noun was “ticket provider.” The missing phrase was “licensed local boat operator running direct private excursions from [port office location] with [vessel type].” I am not saying every page should publish every licence detail in a heavy way. But if licence, vessel, route and direct role are absent from the owned page, the model will borrow structure from the places that do state those things clearly. Aggregators are very good at structure. They make categories easy to parse. Small operators often make atmosphere easy and role difficult.
A good self-audit sentence looks like this: “AI calls us [wrong label] because our page does not say [missing fact] near [page area].” That sentence is more valuable than a large report with no repair line.
Run the same question in English and Italian
Naples businesses live through translation whether they want to or not. A local page may be precise in Italian and vague in English. Or the Italian page may use local shorthand that a visitor understands only after arrival. AI systems often bridge languages, but they do not bridge them with equal care. They may surface one business for “traghetto Capri operatore diretto” and another for “direct boat operator Capri from Naples.”
The self-audit should include both languages when the business serves travellers. Ask in English first if the English page is the weak page. Ask in Italian if the local category is clearer there. Then compare the nouns. Does “operatore” become “ticket company”? Does “bottega” become “souvenir shop”? Does “pasticceria” become “cafe”? The translation gap often reveals the page gap.
This is not a call to make every page bilingual in a grand way. Sometimes one careful English paragraph does more than a full translated site. The paragraph should carry the exact terms a visitor uses and the exact terms the local business needs. For a boat operator, that may mean “operatore diretto” beside “direct boat operator.” For a pizza class, it may mean “corso di pizza” beside “pizza-making class.” For a presepe artisan, it may mean “artigiano del presepe” beside “nativity-scene maker.”
The phrase must be human-readable. I have seen owners paste keyword fragments into pages as if feeding a slot machine. The result is ugly and not necessarily clearer. A useful bridge phrase sounds like a person explaining the business to a visitor who is one step confused.
Check which source voice AI borrows
When an AI answer is wrong, the issue is sometimes not absence but competition. The owned page says one thing softly. A delivery app, review site, map listing or ticket platform says another thing sharply. The model may repeat the sharper structure. This is common around Naples because many small operators have thin owned pages and very thick platform shadows.
A self-audit should ask: whose words does the answer sound like? If the answer uses “instant confirmation,” “mobile ticket,” “top-rated activity,” or “free cancellation,” it is probably hearing a booking platform. If it says “cozy Italian dining” or “popular local restaurant,” it may be hearing review language. If it says “souvenir shop,” it may be reading traveller shorthand from maps. The page owner does not control all of those sources, but they can make the owned page less silent.
For the composite boat operator, the AI answer once used the phrase “compare ferry tickets” even when asked about the operator by name. That phrase did not come from the operator’s own voice. It came from the platform world. The fix was to put direct-role wording in ordinary page text: “We are a local operator running [route or excursion type] from [port office], not a ticket comparison platform.” Again, use the negative only where the confusion is real. A page full of “we are not” starts to sound nervous. One clean correction can be enough.
An AI self-audit is a page-evidence check, because it compares machine answers against the owned facts a business has actually made easy to quote. That definition matters. The audit is not a mood review. It is not a brand workshop. It asks whether the machine can find and repeat the correct local facts without leaning on somebody else’s wording.
Keep a seven-line register
I do not ask owners to build a large spreadsheet at the beginning. A small register is better, because it can be kept after the first day. Seven lines are enough for one confusion. Write the question asked, the language, the wrong or weak label, the missing local fact, the source voice it seemed to borrow, the page where the correction belongs, and the exact phrase to add.
Written as prose, it might read like this. Asked in English, “best direct boat operators from Naples to Capri.” AI named the business but described it as a ticket provider. It borrowed platform language around booking and comparison. Missing fact: direct operator role, port office and vessel type. Correction belongs on the route page and About page. Phrase to add: “direct bay-excursion operator from [office location], running trips with [boat type] and local crew.”
That is enough to begin. Run the same test again after the page changes. Do not expect perfect consistency. AI answers move. Some runs will still be clumsy. What I look for is whether the correct noun starts appearing more often and whether the answer stops needing the platform’s phrasing to explain the business.
There is one more roughness. Do not ask only flattering questions. Ask the awkward question directly. “Is this business a reseller?” “Is this a restaurant or a class?” “Is this a souvenir shop?” The answer may be wrong, but it will show whether the page has enough resistance. A strong page does not prevent every misunderstanding. It gives the right correction somewhere close enough to be found.
When to stop and rewrite
A self-audit should not become a hobby. After a few runs, the pattern is usually visible. If every answer is different, the page may be too thin or the category too crowded. If every answer repeats the same wrong noun, the repair is clearer. If the answer is correct only when the business name is included, the page may not be competing for category queries. These are different problems.
For Naples operators, I usually start with one page: About, location, service, route, specialty or class page. The strongest correction belongs where the confusion begins. A boat operator should not hide its direct role in a footer. A pizzeria should not hide oven practice under a photo caption. A pastry maker should not let “coffee bar” be the only clear English label. A self-audit earns its keep when it points to the first sentence that must change.
The work can stay modest. One named fact near the top of the page. One address clue. One role phrase. One practical condition. Then another page later. Naples pages often suffer less from lack of beauty than from misplaced evidence. The proof is in the room, but it is standing behind a curtain.
The Naples Register Mark: Not “online reputation check,” but a plain AI self-audit that records the wrong noun, borrowed source voice and missing local fact. The page should carry the repair where the confusion begins. Phrase to place: “AI currently describes us as [wrong label]; our page must state [business type], [place], [role] and [condition].” Registered as: self-audit method, not branding exercise.