The Right and Wrong Way to Use AI Agents to Find Cheaper Flights

In the wake of United Airlines’ recent ban on AI agents booking flights for customers, here’s what travelers should consider when using AI to find the best possible airfares.
United Airlines planes ascending into a blue sky, with clouds below the aircraft

United Airlines is taking a stand against AI agents filling in as booking agents.

Photo by Hieu/Unsplash

Last week, United Airlines announced that it was blocking AI agents from booking flights on behalf of customers. Some in the travel industry were outraged. Some were delighted.

Many travelers were simply baffled.

Here’s the short version: United is prohibiting artificial intelligence from doing something you maybe (probably?) shouldn’t be asking AI to do anyway.

Using AI agents to find flights

AI “agents” (not to be confused with travel agents) are digital tools that can take action on your behalf. They can autonomously add calendar events from your emails, spin up a weekly meal plan and shopping list, and make restaurant reservations while you sleep.

They also can, with sufficient directions and permissions, buy your airline tickets.

There are good reasons not to allow AI agents to purchase your flight. As the disclaimer at the bottom of nearly every AI chatbot warns, “AI can make mistakes.” It can miss a lower fare, forget to include your carry-on, or send you to LAN instead of LAX. A grocery cart with random sauerkraut in it is one thing; a nonrefundable ticket to Lansing Capital Region International Airport in Michigan instead of Los Angeles is another.

“There are a lot of understandable concerns right now regarding AI, starting with its capability to go rogue,” says Henry Harteveldt, president and airline industry analyst at Atmosphere Research Group.

Most travelers use AI tools primarily for information gathering and travel planning, he says. Once they find what they want, they book the itinerary themselves.

Monitoring airfare drops

Still, travelers who have had luck with AI flight agents point to one clear perk: Some tools continuously monitor booked routes for price drops and alert users when a fare falls, making it easy to cancel and rebook at the lower rate for a refund or airline credit.

While agents can certainly be programmed to watch routes for price drops, standard flight-search tools like Expedia’s or Kayak’s already do that.

Presumably, you’d use an AI agent to avoid the tedium of comparing fares across airlines that have different add-on charges and ideally save money. Yet, in practice, the tools often deliver neither.

Testing AI agents on flight searches

To put agents to the test, I asked Google’s Gemini agent—available to Google Pro subscribers ($20 a month)—to find a nonstop flight from any Washington, D.C.–area airport to Cancún with specific dates and bag rules. It clicked through dozens of Google Flights screens like a 1995 software demo, dawdling at drop-down menus as if stroking its chin.

It took 21 minutes to surface a $767 JetBlue flight. It stopped at the traveler information page, where I provided information in the chat that it used to complete the form fields in the same halting manner. It stopped to ask if I wanted to proceed before pulling the trigger. This wasn’t faster and easier than manual booking; it was slower and harder.

Proponents note that agents can work in the background, making slow operations painless—though fares can expire while a bot cogitates, automated searches can tie up seats for other fliers, and AI’s characteristic stumbles can cause booking errors, according to a growing collection of industry reports.

An AI agent has to stop to acquire necessary additional information or permissions from a human, which not only takes time but also means the AI agent typically can’t work fully autonomously in the background anyway.

I then tried Google’s AI Mode—the familiar question-and-answer conversational search—using the same fare parameters I’d used for the agent. In less than 10 seconds, it returned a flight costing $138 less and provided a link for the precise fare. I’d have to book the flight myself.

Next, I tried Meta’s Muse agent, a mobile app with a clean interface. In three minutes, it surfaced the same lower fare Google’s AI Mode found and chattily summarized the trade-offs: one option saved money with a tight layover; another maximized beach time. It asked if I was ready to book and, if so, requested my passport, contact, and frequent-flier details, explaining I’d pay via Stripe, the third-party payment processor Muse uses to handle checkout without exposing card numbers directly to the AI.

Who should complete the flight booking? The human or AI?

The final step of enabling AI to complete the booking is where Muse—or any AI agent—differs from searching for flights in Google’s AI Mode. It’s also what gives some airlines indigestion, as they lose opportunities to upsell the traveler on better seats and services, Harteveldt says. Airlines also risk losing the primary relationship with you as a customer.

Other companies are building Muse-style agents that pull live fares and execute bookings through direct connections to airline systems, as opposed to those from OpenAI, Anthropic, and Perplexity, which send clunky browser-clicking bots onto the same sites you’d use.

The companies building these slimmer, faster agents include Expedia, AI travel planner Mindtrip, and airlines that are launching their own tools on ChatGPT. Meanwhile, carriers such as Delta Air Lines and Southwest Airlines allow approved AI agents to assist travelers with finding or booking flights, provided travelers sign in to finish checkout directly with the airline.

All that tech investment runs headlong into a basic consumer reality: Travelers are happy to let AI shop, not buy. While the International Air Transport Association found in its 2026 Global Passenger Survey, which is based on responses from more than 100,000 travelers, that nearly half of air passengers have used AI to help plan trips and search for flights, a recent Skift research survey showed 2 percent of leisure travelers are willing to let an autonomous agent book a ticket on their behalf.

Still, Harteveldt expects that hesitation to fade.

“It’s not a matter of if AI is going to be used for travel-related transactions,” Harteveldt says. “It’s a matter of when. . . . I expect that within three years, we will see the use of AI tools become quite commonplace for flight booking.”

Whether airlines will allow it is another question: More carriers could follow United’s lead and block unauthorized screen-scraping bots, pushing travelers toward airline-approved AI partners or carriers’ own built-in AI tools instead.

Until then, the best approach is to let AI handle the window shopping—and keep a human finger on the “buy” button.

Craig Stoltz, a former travel editor of the Washington Post, writes about travel, food, and drink for Garden & Gun, Fodor’s, Frommer’s, GoWorld Travel, Virginia Living, Wine Traveler, and many other publications.

Enjoyed this article? Save us as a preferred source

FROM OUR PARTNERS
Sign up for our newsletter
Join more than a million of the world’s best travelers. Subscribe to the Daily Wander newsletter.
MORE FROM AFAR