Skip to content
AI Visibility Observatory · Italian tourism

Italy/Method

Method

How we measure, in order. Every choice here is a design decision, not a result: results are on the other pages.

01

In short

Every month we ask the same set of 2,837 travel questions to AI engines with web search: ChatGPT, Perplexity, Copilot, Grok and Gemini.

Each question is asked several times every month on every engine, and its answers become a single value: so model randomness is not mistaken for change.

Every question comes from a real Google search: together, their keywords add up to 65,680,440 searches a year.

For every answer we look at two things: who is named (destinations, hotels) and who is the source (the links cited), split into Tourism promotion sites, Hotel, OTA and Other.

Every month rankings are compared with the previous month's, on shared questions only: the ± next to each number is that change.

02

Questions

The set covers seven categories: sea, mountains, lakes, villages, cities, snow, and the region in general (“what to do in Puglia”).

Two levels: comparing destinations (“where to go to the sea in Puglia”) and inside a destination (“best hotels in Monopoli”). Questions cover the whole trip (where to go, what to do, where to stay) and, for hotel questions, traveller types: families, couples, business, pets, budget, luxury. Phases are used to build a balanced set; on the site results add them up.

Every question is justified by a keyword with at least 1,000 searches a year in its language (Italian in Italy, English worldwide), measured on Google Ads: volume of the exact keyword, sum of the last 12 months. It is the volume of that phrase (“where to stay costa smeralda”), not of the place name, which is searched far more. The keyword and its volume are under each question.

No questions without searches. A question enters the set only if people search for it. That is why not every region appears in every category: where the base has no places of the category, or searches stay below the threshold, the region is left out of the ranking and the page says so. The same goes for hotels: a place where few people search for somewhere to stay has few hotel questions, and the page says how many searches are behind them.

Under «The questions», on every page, the total Google searches behind those questions, each keyword counted once. It says how many people search for those things: not how many visits the place gets, nor how many times we asked the AI engines.

Destinations come from official sources: ISTAT tourist nights, Ministry lists, OpenSkiMap ski areas, OpenStreetMap parks, Most Beautiful Villages of Italy. Areas (parks, ski areas) are asked as “where to stay near X”.

The base is completed by the answers. Places AI proposes in one month that no official list has (Lake Bilancino, a minor ski area) become candidates for the following month, if at least two engines propose them in one answer out of four and they pass the same search threshold. A wave's set never changes while the wave is running.

03

Engines

Answers are collected from the engines' real interfaces, as a user sees them, with web search on where possible: ChatGPT and Perplexity with search, Copilot in search mode, Grok in Expert mode.

Country: Italy for Italian questions; for English ones, Italy, except Gemini, which uses the United States because with Italy it answers in Italian.

The model each engine declares is stored with every answer: a model change shows in the data.

04

Waves

A wave is the month's survey: the same set of questions, asked several times on every engine. Every question is worth several answers, never just one.

Every answer is kept in full: if the way we count it changes, it is re-read without querying the engines again.

A month's wave comes out on the first day of the following month: July's carries the date of 1 August.

05

Sources: tourism promotion sites, hotels, OTAs, other

A source is classified by its domain only, never by the text. Rules, in order: hand-written registry; OTA brands; municipality websites; the site of a hotel named in the same answer; everything else is Other.

In Italy, in the categories and in the regions, who tells the destination is measured without the hotel questions (“what are the best hotels in…”): whoever asks for a hotel necessarily gets hotel sites and OTAs, and those questions would inflate their shares everywhere. On a place page those are exactly the questions that count, because there the comparison is between the hotel's own site and the OTAs.

Tourism promotion sites: those run by public bodies of the territory (regions, municipalities, park authorities and their agencies). A portal with an institutional-sounding name run by a private company is Other, as are national promotion portals: the registry records the check.

OTA: online agencies and metasearch. Tripadvisor and Google Hotels are OTAs. Google cards cited by Gemini are OTA “Google Hotels” when they belong to a property, Other “Google Maps” otherwise.

Hotel: the official site of a property or a chain.

Other: guides, media, blogs, social, private portals, non-tourism institutions. The most cited domains get a label and a subtype.

06

Named hotels and destinations

Hotels: a language model reads every answer and lists the properties it names. A name only counts if it appears verbatim in the text; names of towns, areas or lakes mistaken for hotels are discarded, and so are the items of a numbered list (“Hotel 1”, “Hotel X”).

Places: on comparison questions we count the places the answer proposes (lakes when lakes are asked for, ski resorts when the question is where to ski), not places named only as a reference ("a few kilometres from Florence"). Every place is matched against a gazetteer, so the same place under two names counts once.

When the way we read answers improves, past waves are re-read too.

07

Metrics

Cell = one question on one engine, with all its answers in the wave. For every entity (source type, tourism promotion site, hotel, place, region): presence = answers where it appears ÷ answers in the cell; first = answers where it is the first source or first named ÷ answers in the cell.

Every published number is the plain average of the cells in the selection: every question weighs the same, with no search-volume weights.

Change (±) = this month's presence minus the previous month's, in points, computed on questions present in both months only. Within the spread between the month's answers it is stable: the number stays, with no green or red.

08

How solid the numbers are

It is not a poll. We do not pick a sample: we ask every travel question about Italy that passes 1,000 Google searches a year, and no other. Every question is a real search.

Every question is asked several times every month on every engine, and comes back unchanged the month after: we follow it over time. No number on the site comes from a single answer.

Quality before quantity. Where people search little, questions are few: in many places just one, the one people actually ask. A real question followed over time is worth more than ten questions nobody asks.

AI engines change model and rules often. If answers stay the same from one month to the next, that is a finding; if they change, that is a finding too: it is exactly what the observatory records. A change within the spread between the same month's answers counts as stable: the number stays, without green or red.

The ± is computed only on questions present in both months: a new question does not move the comparison with the previous month.

The questions are public: the CSV at the bottom of the page lists each one's ID, text, keyword and yearly Google searches.

09

Rankings and movement

Sorted by presence: how many answers name an item. At equal presence, first decides: how many answers name it first. Two items named in every answer are not tied if the AI always puts one of them ahead. Only when this is equal too do they share a rank. No minimum threshold.

Each category has its own questions, and so its own ranking. Among a region's cities the question is "which cities to visit"; for the region in general it is "what to see", "what to do", "where to go for a weekend", "an itinerary". For a weekend the AI often proposes villages and valleys too, not just the main city. The same place can be first in one ranking and third in another: it is not an error, the questions are different.

The path: a category in Italy gives the ranking of regions (only those where the category exists), with the most cited source in each; a region gives the ranking of places and how much tourism promotion sites cover it; a place gives the ranking of hotels AI recommends and who tells its story, from the hotel's own site to the OTAs.

Tourism promotion sites and hotels have separate rankings. OTAs and other sources stay as a comparison.

Next to the figure, the rank movement versus the previous month: an up arrow for a climb (with how many places), a down arrow for a drop, a dash for no change, a dot for a newcomer.

10

Limits

An AI answer changes with the user, history, location and day: we measure an anonymous user, from a declared country, every month.

A domain outside the registry and the rules stays Other: rarely cited tourism promotion sites can end up there until they are added to the registry.

The hotel extractor can be wrong: discarded mentions stay in the database with their reason.

A destination named negatively (“avoid X in August”) still counts as a mention.

Search volumes are Google Ads estimates of Google searches, not of AI platforms: they say how much a question weighs, not how often it is asked to an AI.

11

Versions

Every wave records the version of the question set, the extractor and the source registry. A published wave never changes silently: if we find an error, we fix it and say so in a note.

Questions

All the questions in the wave: ID, language, category, place, text, keyword and yearly Google searches.

Download the questions (CSV)