How ready are your guests for AI: working out where you really stand
Anyone who works in a hotel has already noticed something, even without reading any research: over the last two years, talking to a machine has become normal. Millions of people ask a virtual assistant for advice before buying something, have it summarise a document, have it translate a menu. What looked like a toy for enthusiasts in 2023 is now an everyday habit for a growing share of the population, and the speed at which it happened has no recent precedent.
For a hotelier this is good news, because it means the biggest obstacle to automating service — getting the guest to accept it — is coming down on its own, without anyone having to push. But it also means the right question is no longer whether to move. It is where on the curve you stand, with your guests, in your property.
And this is where the numbers in circulation help very little. This article tries to put things in order: what the available research says, why almost none of it describes your hotel, and how to reach an answer of your own with data you already have.
The curve you find everywhere
Anyone evaluating an automated response system today runs into a ranking almost immediately. A survey from spring 2026 lines up guest acceptance along the stay: concierge-style recommendations on top at 68%, in-room assistant at 66%, post-stay handling at 65%, queue-free check-in at 61%, booking last at 54%. The advice that comes with it is to start from the room and expand from there.
Before building an investment on it, it is worth looking at how it was done. It is 122 people, all in the United States, reached online among those who volunteer for surveys, with no criteria to ensure representativeness. The authors themselves write that the results point in a direction but cannot be extended to the population.
Credit to them for publishing everything: it is precisely this transparency that lets readers judge how much weight the figure carries. Many industry studies do not, and those are the truly unusable ones. The criterion here is simple and applies to anyone who publishes numbers — including the author of this article: research is judged by how it is built, not by who signs it. If the method is sound, a figure produced by a company in the sector holds; if there is no method, it does not hold even when it comes from a disinterested party.
The point that changes everything: it is not where, it is what
There is a university study, published in an academic hospitality journal by researchers from West Virginia University and the University of South Florida, that surveyed 145 people across hospitality professionals and guests with recent stays. It says two things.
The first: people who work in hotels are far more enthusiastic about automation than guests are, because they immediately see the relief on their workload. The second is subtler and more useful. Guests want a person back when the request carries an emotional expectation: the example given is advice on where to have dinner for an anniversary.
Compare this with the earlier ranking, where concierge was the point of highest acceptance. A contradiction? No: the two studies use “concierge” for two different things. One means a quick piece of information, the other a recommendation in which the guest has invested something personal.
And it is the most useful finding of the lot: what decides acceptance is not the stage of the stay, it is the emotional weight of the request. Splitting the work by moments — booking, check-in, room, departure — leads you astray. Splitting it between ordinary requests and requests that touch an expectation leads exactly where you need to go.
It is reassuring news too, if you think about it: it means the technology is not competing with the part of the job that is rewarding. It is competing with the part that eats time.
Where Italy stands (and why lagging behind is an opportunity)
This calls for a serious source, and fortunately there is one. Istat, the Italian national statistics institute, in its Citizens and ICT survey released in April 2026, measured for the first time how many Italians use artificial intelligence tools: 19.9% of those aged 16 to 74 in the previous three months, against a European average of 32.7%. The leading countries, Denmark and Estonia, are at around 47-48%.
The more interesting figure, though, is another one: the reason non-users give for not using it. 59.9% simply say they do not need it, 21.6% that they would not know how, and only 5.5% raise doubts about security or privacy.
Compare that with the United States, where the university study records very high levels of privacy concern, around 76%. They are two different worlds. There, the obstacle is distrust, which is hard work to dismantle. Here, the obstacle is that nobody has yet shown people what it is for — and that is dismantled simply by making the thing work well the first time a guest tries it.
This is why Italy’s lag should not be read as a sentence. An obstacle made of habits falls far faster than one made of suspicion, and those who arrive later skip the whole phase of failed experiments run by others.
Use is also concentrated among the young: 51.2% of 14 to 19-year-olds, 43.1% of 20 to 24-year-olds. These are people who will be booking rooms in a few years, and who will expect to be able to write at any hour and get an answer.
There is no single curve: there is yours
These numbers yield the most practical observation in the whole article, and oddly nobody makes it.
A Danish or Estonian guest has a daily habit of using virtual assistants more than twice that of an Italian guest. A seaside hotel with 70% Italian guests over 50 and a city hotel with 60% Northern European guests under 40 are not at two points on the same curve: they are on two different curves.
Which means that for many Italian properties the right moment has already arrived, because part of their clientele is already above the national average. And it also means that the useful starting figure is not a European average, let alone an American one: it is your arrivals register.
On the business side, the race has already started
On the business front we have something better than a snapshot: three consecutive surveys. Italian companies with at least ten employees using at least one artificial intelligence technology were 5.0% in 2023, 8.2% in 2024, 16.4% in 2025. A doubling in one year, after an almost-doubling the year before.
Two honest caveats. The survey covers companies above ten employees, so it does not capture the thirty-room family-run hotel. And “uses at least one AI technology” also includes data analysis and administrative automation, not just systems that talk to guests.
That said, the direction is unmistakable and the slope is steep. When a technology doubles its reach two years running, the debate over whether to adopt it is already over without our noticing.
Where we will get to: three scenarios, stated for what they are
What follows is not data: it is a reasoned estimate, built on what we know. I set it out with its assumptions in plain view, because a number whose construction cannot be seen is no use for deciding anything.
We start from three anchors. The speed already observed among businesses, which is high. The distance from Europe, about thirteen points, which says how much catching up lies ahead. And a natural ceiling: if acceptance depends on the emotional weight of the request, then the maximum achievable is not 100% of contacts, it is the share of ordinary requests the hotel receives.
One limit to declare: on the guest side we have a single survey, the 2025 one, because the question was not asked before. We know where we are, not how fast we are moving. The projection therefore proceeds by analogy with what we see among businesses and in the countries ahead of us.
On this basis, looking at how many guests accept resolving an ordinary request with an automated system without asking for a person:
Cautious scenario. Early implementations work poorly and burn trust. Over the next eighteen to twenty-four months, acceptance among Italian guests stays around a quarter; at three to five years it stops below a third.
Central scenario, the most likely. Italy reaches today’s European average two to three years late, driven by generational turnover and by habits formed elsewhere. Over the next eighteen to twenty-four months it gets to a good third of Italian guests, and already today to considerably more among Northern European ones; at three to five years it passes half on ordinary requests.
Fast scenario. Talking to an assistant becomes the normal way to search and book, and acceptance grows not because guests change their minds but because the door they come in through changes. Eighteen to twenty-four months to reach around half, three to five years to pass two thirds.
Confidence differs between the two horizons. Over eighteen to twenty-four months the range is narrow, because none of the variables can move too far in that time. Over three to five years it is a scenario exercise, useful for steering a decision but not verifiable.
And the point that matters more than the numbers remains, true in all three cases: what grows is the volume of the ordinary that technology absorbs, not the nature of the extraordinary. Whichever scenario comes true, the part of the job that needs a person stays there.
How to build your own curve in a month
The method is worth more than the estimate, and it is done with data you already own.
First, look at where your guests come from. From the arrivals register you get the breakdown by nationality and age band. Weighting the shares by the AI usage rates of the respective countries — published every year and freely available — gives you a starting estimate that concerns your clientele, not an average that does not resemble you.
Second, split the requests you receive. One month of notes on calls and messages, separating the ordinary ones — opening hours, availability, parking, breakfast, confirmations — from those in which the guest is looking for advice or reassurance. That ratio is your real ceiling, and you will not find it in any market research because it depends on what kind of hotel you are.
Third, count how many you lose. How many of those requests today go unanswered, or get an answer too late to be useful. It is the number that tells you whether your problem is guest acceptance or the property’s capacity to respond — and the second is usually solved first, with less technology and a faster return.
Conclusions
Artificial intelligence is entering everyday use at a speed our sector has never seen before, and the historic obstacle to automating service — resistance from those on the receiving end — is coming down on its own as we speak.
Italy starts further behind than almost all of Europe, but for a reason that works in favour of those who move now: it is not distrust, it is a lack of occasions to understand what it is for. Resistance of this kind falls quickly, and it falls first where someone offers an experience that works.
The figures circulating on guest acceptance should be used for what they are: indications of direction, almost all measured on populations that do not resemble yours. The direction is clear and points entirely the same way. The exact point where you stand, on the other hand, is written in your arrivals register and your switchboard — and it takes a month of attention to find it.
Frequently asked questions
How many Italians use artificial intelligence today? According to the Istat Citizens and ICT survey for 2025, 19.9% of people aged 16 to 74 used generative AI tools in the previous three months, against a European average of 32.7%. Among 14 to 19-year-olds the share rises to 51.2%.
Why do Italians use AI less? Is it a privacy issue? No, or only marginally. Among non-users, 59.9% say they do not need it and 21.6% that they do not know how to use it; only 5.5% cite security or data protection concerns. It is an important difference from the United States, where distrust weighs much more.
How many Italian companies use artificial intelligence technologies? According to the Istat Enterprises and ICT survey, 16.4% of companies with at least ten employees in 2025, against 8.2% in 2024 and 5.0% in 2023. The figure does not cover companies under ten employees, a threshold below which many accommodation businesses fall.
Is guest acceptance the same for every hotel? No. AI usage rates vary widely by country — from 19.9% in Italy to 47-48% in Denmark and Estonia — and by age. A property with many young Northern European guests is much further ahead than one with mature Italian guests.
What determines whether a guest accepts talking to an automated system? More than the stage of the stay, what counts is the emotional weight of the request. Ordinary informational questions meet high acceptance; those in which the guest seeks personal advice or reassurance preferably stay human.
How should industry research on technology adoption be assessed? By how it is built: how many people, in which country, chosen how, and whether it measures actual behaviour or stated intentions. A study sponsored by a company in the sector that publishes its method is more usable than one that does not publish it at all.
A note on sources. Data on the Italian population come from the Istat survey “Cittadini e ICT – Anno 2025” (Citizens and ICT – 2025), published on 22 April 2026, and the Eurostat European comparison reported there; data on businesses from the Istat release “Imprese e ICT – Anno 2025” (Enterprises and ICT – 2025). The university reference is the study “Investigating Customer Service and Engagement Levels in the Lodging Industry: High-Touch to High-Tech Conversational AI”, published in the Journal of Hospitality and Tourism Technology by Ajay Aluri and Amelia Szczesny (West Virginia University) and Luana Nanu (University of South Florida), conducted on 145 participants in the south-eastern United States. The acceptance ranking by touchpoint comes from an April 2026 survey of 122 US respondents with a non-representative volunteer sample, sponsored by a voice technology vendor; for both, the sample size is given because it is what determines their usability. The scenarios presented are the author’s own elaborations built on the elements stated in the text and are not survey data.