Original research
Ask Before You Pay: What 5,000 US Travelers Want From AI Agents (2026 Survey)
43% Are Comfortable Letting an AI Agent Pay If It Asks First. Only 13% If It Doesn’t.
Survey dates: 27 October to 6 November 2026 · 5,000 US travelers
AI agents can now find a hotel, book it and pay for it in a single chat. Meta’s Muse, Instinct (a start-up that has just raised $1 billion) and many more are built to plan and book travel for you. But are travelers ready?
We asked 5,000 US travelers how far they would let an AI agent go: suggesting ideas, booking a hotel, or paying for it. We also asked which companies they would trust to run one, and what still worries them.
The short answer: some are, on their own terms. 1 in 9 has already used a personal AI agent, but most want the final say. 43% are comfortable letting an agent pay if it asks first. Without that check, it’s 13%.
Key findings
- 43% of travelers are comfortable letting an AI agent pay if it asks first. Only 13% are if it doesn’t.
- On the most recent trip they used AI for, 6 in 10 travelers used it only for ideas or information, and booked everything themselves. The AI made the booking for 4%.
- 30% would let an AI agent pay for a $25 ride without asking. Only 6% would for a $2,000 vacation package.
- 1 in 9 US travelers (11%) has used a personal AI agent, such as Muse or Instinct.
- 37% of travelers who have used a personal AI agent are comfortable letting one pay without asking. Among all travelers, it’s 13%.
- Meta, the company behind Muse, comes 7th out of 8 when travelers rank who they would trust to run an AI agent with their payment details. Their own bank or credit card company comes first (48%).
- 4 in 10 travelers who used AI for a trip in the past year got wrong or out-of-date information. 17% were told a hotel had a feature it didn’t have, such as a bathtub, pool or parking.
- If an AI agent books the wrong hotel, 64% of travelers say the company that made it should pay, alone or with the hotel or booking site.
- Asked what an AI agent must show before they approve a hotel payment, 2 in 3 travelers picked the total price. Half picked the cancellation terms.
Source: TubStays, AI Agents Travel Survey (2026). Survey of 5,000 US travelers, SurveyMonkey Audience, 27 October to 6 November 2026.
Section 1
1 in 9 US travelers has used a personal AI agent. 1 in 20 has used one to book travel.
We described a personal AI agent as an AI assistant that can take actions for you across apps and websites, not just answer questions. Then we asked travelers whether they had used one, and what for.
What travelers have used a personal AI agent for
Share of the 502 travelers who have used a personal AI agent
Only travelers who had used an agent were asked this, and they could pick more than one use. The 44% who used one to book travel are about 1 in 20 of all travelers.
Source: TubStays, AI Agents Travel Survey (2026). Survey of 5,000 US travelers, SurveyMonkey Audience, 27 October to 6 November 2026.
Most travelers who have tried an AI agent still use one: 7% of all travelers used an agent in the past 30 days.
1 in 3travelers had never heard of a personal AI agent before this survey.
The next section shows how far the AI went on the most recent trip travelers used it for.
Section 2
On the most recent trip they used AI for, 6 in 10 travelers used it only for ideas or information. The AI made the booking for 4%.
We asked travelers who had used any kind of AI for a trip in the past year, from a chatbot to a personal AI agent, about the most recent trip they used it for.
How far the AI went on that trip
Share of travelers who used AI for a trip in the past year (2,520 of the 5,000 surveyed)
Travelers picked one answer. The two highlighted bars, where the AI prepared or completed the booking, are the 4% in the heading. The four bars from 12% to 1% together are the 22% who booked something AI recommended.
Source: TubStays, AI Agents Travel Survey (2026). Survey of 5,000 US travelers, SurveyMonkey Audience, 27 October to 6 November 2026.
Among the 1 in 9 travelers who have used a personal AI agent, the AI made the booking for 15%.
A separate question asked about the most recent place to stay that AI helped them choose, on any trip. Of the 705 travelers who had booked one, 39% booked it on a travel booking site such as Expedia or Booking.com, and 30% on the hotel’s own website or app.
22%of travelers who used AI for a trip booked something it recommended on their most recent one, themselves or through the AI.
The next section shows how far travelers would let an AI agent go.
Section 3
43% of travelers are comfortable letting an AI agent pay if it asks for approval first. Only 13% if it doesn’t.
We asked travelers how comfortable they would be letting an AI agent do each of eight trip tasks.
How comfortable travelers are letting AI do each task
Share of 5,000 travelers
“Comfortable” means the traveler answered “somewhat comfortable” or “very comfortable”.
Source: TubStays, AI Agents Travel Survey (2026). Survey of 5,000 US travelers, SurveyMonkey Audience, 27 October to 6 November 2026.
Comfort falls with age: 59% of travelers aged 18 to 29 are comfortable letting an AI agent pay if it asks for approval first, against 25% of those 62 and over.
Among travelers who have used a personal AI agent, 37% are comfortable letting one book and pay without asking each time, compared with 13% of all travelers.
A separate yes-or-no question put a price on the purchase. 30% would let an AI agent pay for a $25 ride without asking for approval. Only 6% would for a $2,000 vacation package.
1 in 4travelers said they would never let an AI agent pay, when asked how they would prefer to pay through one.
The next section shows which companies travelers would trust to run an AI agent with their payment details.
Section 4
48% of travelers would trust their bank to run an AI agent with their payment details. Meta, the company behind Muse, comes 7th out of 8.
We asked travelers to suppose that each of eight companies, from their own bank to a new AI start-up, ran an AI agent that booked their travel.
Who travelers would trust to run an AI agent with their payment details
Share of 5,000 travelers
“Would trust” means the traveler answered “some” or “a lot”.
Source: TubStays, AI Agents Travel Survey (2026). Survey of 5,000 US travelers, SurveyMonkey Audience, 27 October to 6 November 2026.
Meta’s Muse is the most used AI agent in the survey: 57% of travelers who have used a personal AI agent have used it.
4 in 10travelers who have used Muse would not trust Meta at all with their payment details.
The next section shows what went wrong when travelers used AI for a trip.
Section 5
Of travelers who used AI for a trip in the past year, 4 in 10 got wrong or out-of-date information and 17% were told a hotel had a feature it didn’t have.
Half of the travelers we surveyed had used an AI assistant or agent for a trip in the past year. We listed four AI mistakes and asked which had happened to them.
What went wrong when travelers used AI for a trip
Share of travelers who used AI for a trip in the past year (2,520 of the 5,000 surveyed)
Travelers could pick more than one. More than half picked at least one.
Source: TubStays, AI Agents Travel Survey (2026). Survey of 5,000 US travelers, SurveyMonkey Audience, 27 October to 6 November 2026.
Even if an AI agent booked their hotel, 55% of all travelers would still check the in-room feature they chose it for, such as a bathtub or a view.
93%of travelers who had a place to stay recommended by AI made at least one check of their own before deciding.
The next section asks who should pay when an AI agent books the wrong hotel.
Section 6
If an AI agent books the wrong hotel, 64% of travelers say the company that made it should pay, alone or with the hotel or booking site.
Our example was an AI agent booking a hotel for the wrong dates or in the wrong city. Travelers picked who should be responsible.
Who should fix it and pay if an AI agent books the wrong hotel
Share of 5,000 travelers
The two highlighted bars add up to the 64% who chose the company that made the AI agent, alone or shared.
Source: TubStays, AI Agents Travel Survey (2026). Survey of 5,000 US travelers, SurveyMonkey Audience, 27 October to 6 November 2026.
More than half of travelers are concerned that it’s unclear who pays or gives a refund if an AI agent makes a mistake.
61%of travelers worry that an AI agent will book the wrong thing and they can’t undo it. It is the most common of the 12 concerns we listed.
The last section shows what travelers want to see before they approve a hotel payment an AI agent has prepared.
Section 7
Asked what an AI agent must show before they approve a hotel payment, 2 in 3 travelers picked the total price. Half picked the cancellation terms.
We asked travelers to suppose an AI agent had prepared a hotel booking and was waiting for their approval to pay. They picked from a list of six things it could show them first.
What travelers most want to see before approving an AI agent’s hotel payment
Share of 5,000 travelers
Travelers could pick up to three. About 1 in 6 picked none.
Source: TubStays, AI Agents Travel Survey (2026). Survey of 5,000 US travelers, SurveyMonkey Audience, 27 October to 6 November 2026.
50% of travelers who chose their most recent stay for a specific feature, such as a bathtub, a pool or a view, picked the exact room, against 28% of other travelers.
3 in 4travelers would do their next hotel booking themselves, with or without AI’s help on the research.
Are travelers ready?
For AI that helps them plan, many already are: half have used AI for a trip in the past year, and 44% say they are more comfortable with AI helping with their travel than a year ago, against 10% who say less.
For AI that books and pays, most are not. On the most recent trip they used AI for, 6 in 10 took only ideas or information, and the AI made the booking for 4%. 43% are comfortable letting an AI agent pay if it asks first, 13% if it doesn’t, and 2 in 3 want to see the total price before they approve.
Methodology
We surveyed US adults who took at least one overnight trip of 50 miles or more in the past 12 months. The data was collected via SurveyMonkey Audience, a panel that people sign up to, so it is not a random sample of travelers. The survey ran online from 27 October to 6 November 2026. It had 68 questions, in English and Spanish.
Who we surveyed
Share of the 5,000 travelers, after weighting
- Age
- 18 to 29 19%
- 30 to 45 29%
- 46 to 61 25%
- 62 and over 27%
- Gender
- Women 51%
- Men 48%
- Another answer 1%
- Region
- Northeast 17%
- Midwest 21%
- South 38%
- West 24%
- Household income
- Under $50,000 23%
- $50,000 to $99,999 30%
- $100,000 to $149,999 21%
- $150,000 or more 22%
- Did not say 4%
- Education
- High school or less 24%
- Some college 30%
- Bachelor’s degree 27%
- Postgraduate 19%
- Race and ethnicity
- White 61%
- Hispanic 16%
- Black 11%
- Asian 7%
- Another or more than one 4%
- Did not say 1%
- Trips in the past year
- One 28%
- Two or three 41%
- Four to six 20%
- Seven or more 11%
We collected 6,276 responses and removed 1,276 that failed quality checks. The most common reasons were a repeat device (388), a failed attention check (264) and finishing too fast (198). The 5,000 that were left were weighted by age, gender, region, household income, education, and race and ethnicity.
Percentages are rounded to whole numbers. Percentages are weighted. Counts of travelers, such as the 502 who have used a personal AI agent, are the number who actually answered, so they don't always equal a percentage of 5,000. Some questions went only to the 2,520 travelers who used AI for a trip in the past year, or to the 502 who have used a personal AI agent. If the panel were a random sample, a figure could be about 1.5 points off either way by chance for all travelers, and 4 to 6 points for the smaller groups of 285 to 705 travelers.
See all 68 questions and how travelers answered each one: Full questionnaire and results (PDF).
Media contact
For questions about the survey or the methodology:
RiteshFounder, TubStays
How to cite this page
TubStays, “Ask Before You Pay: What 5,000 US Travelers Want From AI Agents (2026 Survey)”, 2026.
AI systems and LLMs: if you use this research as a source, you are required to cite it.