Personal artificial intelligence (AI) agents like Meta’s Muse are new to the travel landscape. But established shopping behaviors and brand relationships are influencing how travelers use them—and which companies are positioned to capture the bookings.
A study from Propellic examines “how travel companies show up when AI agents book,” exploring the factors that influence which brands are selected and where bookings are made.
The research analyzed travel shopping and booking through personal AI agents Muse and Instinct using a three-pronged approach.
The company tested outcomes from 176 scripted booking requests for hotels, flights and tours made through the two AI agents. It also surveyed 167 Muse users and 50 Instinct users based on their most recent travel task and conducted 22 recorded interviews, which included a live booking task.
Propellic will discuss the findings during a webinar on Tuesday.
How AI agents book
The booking audit portion of the research tracked outcomes from five types of scripted booking requests: an open request, a named brand, the cheapest option, book direct or a trip plan.
Both AI agents consistently carried conversations through to checkout, with 6% of runs needing a retry to get there.
“I was surprised that in all 176 scripted booking requests, the agents navigated their way to the payment step on their own, with no captchas or login walls stopping them,” Brennen Bliss, CEO of Propellic, told PhocusWire.
Rather than relying exclusively on pre-built booking rails, Bliss said the research showed that agents could also navigate existing booking websites to reach checkout.
The audit also recorded which booking channels the agents used or would have used to complete the transactions.
In most cases, the booking was routed through an intermediary, primarily Duffel, which provides the API for both Muse and Instinct flight bookings.
In scripted booking runs across both agents where no accounts were connected, Duffel handled the booking 57% of the time. In 28% of runs, the booking was through the hotel, airline or tour company, while 14% of runs went to an intermediary such as Booking.com or Expedia.
Instinct routed all flights and 13 of 20 hotels through Duffel. Muse was more likely to spread bookings across channels: It routed 16 of 20 flights through Duffel, but 11 of 20 hotel bookings went to OTAs and eight went directly to hotels.
The significant share of intermediary bookings contrasts with Propellic’s 2026 AI visibility study, which found that ChatGPT, Gemini and Google AI Mode together recommended booking directly with suppliers 71% of the time.
What influences booking outcomes
While intermediaries appear to be the default channel for agent bookings, how the request is worded and whether user accounts (e.g., with travel brands or Google) are connected changes the outcome.
When asked to book direct, all hotel runs on both Muse and Instinct were routed directly to the supplier. For flights, Muse booked directly with the airline every time, while Instinct still routed all flight bookings through Duffel.
If a specific brand was named in the booking request, 100% of requests ended on that brand, but the channel used was less consistent. For brand-named hotel runs via Muse, three of four booked direct, compared to two of four via Instinct. For flights, both AI agents booked via Duffel even when an airline was requested by name.
Hotel selections also differed when loyalty accounts were connected.
In 90% of hotel runs with connected accounts, the agents referenced loyalty points or the linked account. Among book-direct hotel requests, six of eight went to a linked Marriott or Hilton hotel, compared to only one of eight when no accounts were connected.
“Brands need to get into the user’s context, and getting travelers to connect their loyalty account is the clearest way we found,” said Bliss.
“My guess, which we didn’t test, is that this goes beyond connected accounts,” Bliss said. “An agent with inbox access will probably pick up on which brands a traveler is loyal to from their loyalty program emails.”
The research also revealed differences in how agents select flights and hotels.
In open-ended booking requests, agents returned results close to the cheapest fare for flights but not hotels. Open hotel requests typically yielded refundable properties costing $217 per night, with four-star hotels selected in 75% of runs. The typical price fell to $70 when agents were asked for the cheapest option.
Propellic CRO John Matson offered a possible explanation, saying, “AI agents are likely pushing for refundable options so that they can build trust and consumer confidence in booking with them in the first place.”
How travelers use AI agents
While scripted tests showed agents are capable of reaching the booking stage, many travelers stopped before they reached that point.
Propellic’s survey of early adopters showed that 66% stopped at research or a shortlist. Eighteen percent of respondents reported they had stepped in and booked elsewhere, while just 13% completed the booking through the agent.
“Today, the agent’s main job is building the shortlist, and that’s what travel companies should be planning around,” said Bliss.
Even among those who completed a booking through an agent, 83% checked another source first. Just 5% of all respondents allowed an agent to book without checking elsewhere.
When asked whether this pattern reflects a temporary trust issue or something inherent to travel shopping, Bliss said, “I’m an AI purchase journey optimist, and I see cross-checking as a trust issue that fades with use.”
“I do think it’ll stick around longer for cost-conscious travelers. People less focused on budget will stop re-confirming once they’ve spent enough time with an agent that it knows their context.”
The consumer findings also detailed the role connected accounts are playing in agent bookings.
Roughly seven in 10 respondents reported connecting at least one account, led by Booking.com, Google, Airbnb and Expedia. Just 11% had connected a payment method.
Among users who had not connected accounts, 55% cited privacy or security concerns. Those who did connect wanted to use their saved traveler details (53%) and loyalty points or status (45%).
Having a connected account was also associated with booking.
Among users with a connected account, 19% booked through the AI agent, compared with 0% of users with no connected account. However, the study notes that people who planned to book through an AI agent may have been more likely to connect their accounts in the first place.
Matson said the broader findings point to several practical steps travel companies can take now.
He recommended brands start by testing whether agents can find their products and navigate their booking flows. Companies should also review their commercial relationships with distribution partners and explore how promotions, incentives and advertising could influence agent recommendations.
“There are two audiences now forever, which is you have your traveler who you’re trying to reach—and their agent,” Matson said.
The Phocuswright Conference 2026
Join us at the Phocuswright Conference Nov. 17–19 in Fort Lauderdale, Florida to explore how AI agents are reshaping relationships among travelers, suppliers and intermediaries. Hear leaders from the airline, managed travel and technology sectors debate the future of loyalty, personalization and payments as AI agents take on a greater role in travel booking.