Do Hotel Guests Want AI? Where They Still Want a Person

Travel with young kids long enough and you learn which questions can wait for a chat reply. Pool hours can. A fever at 10pm can't. Do hotel guests want AI? For routine questions, many will use it. For the welcome, for problems and for anything that matters to them, they want a person, and they want to see the way to one. So the owner's question is not really yes or no. It is which conversations go where.

If your owner has seen the 89% headline from Mews in 2026 ^[1], the number is real. Mews sells hotel software, and the survey says less than the headline does. This post sorts the published evidence by conversation, so you can say yes to some of it without giving away the moments that make a resort feel like a resort.

If you are still comparing tools, how to judge a hotel AI concierge is the longer read. This one stays with the guest.

Do hotel guests want AI at all?

Yes, for some things. Guests are not against AI. They sort it by what is at stake.

Start with the 89%. In Mews' July 2026 survey of 3,250 travelers in eight countries, 89% wanted "either a balanced mix of AI and human-led hospitality or AI only working behind-the-scenes on repetitive tasks" ^[1]. That is two answers added together, and Mews has not published the split. So it tells you most guests don't want an AI-led stay. It does not tell you how many want AI kept out of sight. In that guest survey, 75% said good hospitality depends on human staff, and 59% of guests believe "AI should take over repetitive tasks, so hoteliers can instead focus on building trust with their guests" ^[1].

French guests point the same way. In a 2025 Ipsos.Digital survey for Mews of 1,000 French adults who had stayed in a hotel in the past year, 77% preferred services handled by a human rather than AI, and 74% favored a hybrid of personal interaction and self-service ^[2]. AI at the Front Desk, the ResortConcierge AI Research report that reads the 2026 surveys side by side sets these beside the rest of the evidence.

There is plenty of routine to hand off. Akia, which sells guest messaging software, classed roughly 80% of inbound guest messages at the average property as repetitive, factual questions in a 30-day sample of 900,000 messages (2026) ^[3]. That counts question types, not questions AI resolved. It is still the pool-hours pile, and it is tall.

Where do guests still want a person instead of AI?

Everywhere Mews looked. In its 2026 guest survey, Mews says "humans are preferred over AI at all nine touchpoints tested," and published figures for five ^[1]. A person led in Mews' figures for bar and restaurant service (72%), the front desk welcome (66%) and check-in and key (58%) ^[1]. Room service orders went 50% human to 23% AI, and even arranging a taxi went 44% to 28% (Mews, 2026) ^[1]. The gap narrows on logistics and widens wherever the moment carries feeling.

In counseling, the first job is rarely the answer. A person who is stressed can't take in an answer until they feel heard. The desk works the same way. A billing surprise, a lost passport, a room still not ready at 5pm: those are person moments. The pool hours are not.

Smaller studies agree. In a 2026 University of South Florida study of 145 people in the southeastern US, 101 of them consumers with recent hotel stays, guests wanted a human concierge for emotionally meaningful requests, like finding a restaurant for an anniversary dinner, and saw AI as useful for extra towels or late checkout ^[4]. Among 214 US travel buyers surveyed by Invoca, a vendor, in May 2026, 54% preferred human help to AI when both were equally available, and 82% called human connection during a high-stakes purchase important or very important ^[5]. Hoteliers draw the same line: in Mews' separate 2026 survey of hoteliers at more than 500 properties, 59% said the front desk welcome and check-in should stay human-led ^[6].

Why do surveys about AI in travel give such different answers?

Because they ask different questions of different people. Skift's State of Travel 2025, as cited by McKinsey in September 2025, found only 2% willing to give an AI tool full autonomy to "make and modify travel bookings without human oversight" ^[7]. Phocuswright, in July 2026, found 40% "would let an AI assistant handle flight and hotel bookings on their behalf" ^[8]. Neither public page states the sample behind its figure. One question removes the human entirely. The other doesn't mention one.

Suggesting and booking split apart even inside one survey. In Expedia Group's March 2026 survey (vendor data, run by YouGov) of more than 5,700 adults in the US, UK and India, 53% were comfortable letting AI suggest travel options, 8% were comfortable booking through an AI platform, and 66% would not trust an AI assistant to buy or book anything on their behalf ^[9].

So read them side by side, and never average them. And 40% is not "almost half of travelers." It is 40% of whoever Phocuswright asked. Five surveys asked some version of this question, and the report charts all five.

Four questions to ask before you repeat an AI survey:

• What exactly was asked? Skift's "full autonomy" question (2025) got 2% ^[7]. Phocuswright's "handle bookings on their behalf" (2026) got 40% ^[8].
• Who was asked, how many, and where? Some public pages don't say.
• Were answers combined? Mews' 89% (2026) adds two options together ^[1].
• Who paid for it, and what do they sell? Label vendor data as vendor data.

What do guests need before they will use an AI concierge?

Two things. First, a visible way to a person from the first message, not only after the AI fails. In Five9 data reported by CX Dive (2026, vendor, cross-industry), 55% of consumers trust AI customer service with a clear human contact option, against about a quarter without ^[10]. Expect guests to use it: about 27% of engaged callers to HelloShift's hotel AI voice agent wanted a person, the top request (2026, vendor) ^[11]. That is the system working. ResortConcierge AI works in text today; voice is planned.

The handoff should carry the conversation, since 83% of consumers say they have to repeat themselves at least sometimes after a transfer (Five9, 2026, vendor) ^[12]. ResortConcierge's escalation queue shows guest, room, reason and sentiment, so the next person starts with context.

Second, right answers. Over 60% of 1,029 US travelers cited accuracy, the top barrier to AI travel planning tools (Cornell Center for Hospitality Research, 2026) ^[13]. Those are planning tools, not a resort's assistant, but a Texas A&M study of hotel booking chatbots found inaccuracy drew the strongest negative response (2026) ^[14]. The business owns the answer: a British Columbia tribunal ordered Air Canada to pay $812.02 over its chatbot's wrong fare advice (Moffatt v. Air Canada, 2024) ^[15]. So keep one source current. ResortConcierge answers from your own property knowledge base, which your team writes and edits.

Then say plainly, at the start, that it is AI. What the disclosure rules require, and wording that works covers the detail. This is general information, not legal advice.

Is AI paying off for hotels that already use it?

Not measurably, for most. In a 2026 study of 113 hotel chains by h2c, a consultancy, 91% already use AI, but just 13% report measurable return on investment (ROI), and only 28% have a company-wide AI strategy led by senior leadership ^[16]. The study drew 122 responses, wasn't weighted by chain size and was sponsored by vendors. In Mews' 2026 hotelier survey (vendor), 41% of hoteliers had no formal AI policy ^[6].

I don't read that as a reason to wait. I read it as a reason to start narrow, where guests already accept AI and the result is easy to count. Five numbers an owner will accept covers what to count, including the baseline before launch.

One more h2c number belongs in the owner meeting. Asked about autonomous AI in guest interactions, the chains in h2c's 2026 study named unclear liability in the event of AI errors as their top concern (56%), with data privacy and security close behind (54%) ^[16]. That worry points straight at the keep-with-a-person column below.

Which guest conversations should a resort give to AI, and which should stay human?

Sort them by what is at stake, not by what the software can do. Three columns, and one rule above them all: any guest who asks for a person gets one.

Be honest about the middle column. In Mews' 2026 survey, guests leaned human at every touchpoint tested, taxis included ^[1]. The middle is where AI does the work and a person stays visible, not where guests voted for AI. Booking changes sit there for the reason Expedia's respondents split suggesting from booking ^[9]: AI collects the details, and a person confirms.

In the keep-with-a-person column, AI's job is to notice and step aside. The questions an AI concierge should decline get their own post.

Noticing matters most with the guest who is unhappy but never says so. ResortConcierge flags conversations where a guest sounds unhappy, so a manager can step in before the guest has to ask.

Take a family arriving at a hypothetical 150 room beach resort. The pre-arrival message goes out in the background. The parking question and the 9pm pool hours question go to AI, with a person one tap away. The 10pm question about a pharmacy for a feverish child goes straight to a person, and so does the surprise charge at checkout. Five conversations, three columns, one stay.

Start with the middle column on a few topics, read real conversations every week, and widen only when the answers hold up. Starting with a person reading every reply is one way to run those first weeks.

If your owner wants one sentence, I'd give them this: AI for the questions with one right answer, and a person for the moments guests will remember. This week, sort your resort's 20 most common guest questions into the three columns. Then run the one-hour readiness check in the report and see how AI tools answer for your resort today.