How AI is Reshaping Travel Planning

How external AI assistants are reorganizing the earliest layers of travel planning
— and what it means for airlines arriving second.

Role

  • Research lead, Customer Strategy & Innovation

  • Designed and led mixed-methods study

Partners

  • Shopping

  • Digital Products

  • Customer Experience leadership

Scope

  • First CX-on-AI research
    at United

  • Informing chatbot and external AI strategy

Summary

Problem

Design

Insight

Impact

56%

of US travelers used AI for
trip planning last year.

74%

of millennials now use AI
for travel planning.

7%

of AI users find AI answers
alone sufficient.

Source: Phocuswright, March 2026.

Summary

Travel planning has always been a multichannel journey, but until recently every channel sat inside a recognizable ecosystem — search, airline, OTA, airport. External AI assistants have introduced something different: a planning layer the airline doesn't own, sitting upstream of every other touchpoint. I led a retrospective diary study of real AI-assisted travel conversations to understand how travelers are actually using ChatGPT, Gemini, and Claude to explore destinations, compare airlines, and frame purchase decisions before reaching an airline's app or site. The research showed that AI is rarely the endpoint of purchase. It's the place where the planning problem itself is being framed — and customers now judge airline channels against the speed and coherence of what AI just did for them. The work reframed the strategic question from whether to build an internal ChatGPT to how to remain authoritative when planning begins somewhere else.

Problem

Multichannel customer experience research has typically assumed that travelers begin inside recognizable ecosystems: search engines, airline sites, OTAs, maps, airport tools, and service channels. That assumption is becoming less stable. External AI assistants are absorbing many of the exploratory and interpretive jobs that used to be distributed across those environments — destination ideation, itinerary drafting, airline comparison, travel prep, and even in-trip troubleshooting.

For airlines, this creates a new upstream problem. By the time travelers enter an airline's app or site, they may already be carrying an AI-generated shortlist, a draft trip structure, and a set of expectations about how planning information should be explained. The challenge is no longer only whether an airline appears in the consideration set. It's whether the airline's channels feel coherent once customers arrive with an AI-mediated understanding of the trip already in hand.

This raised a set of strategic questions:

  • where in the travel journey are customers actually using external AI;

  • what kinds of expectations do those interactions create before customers enter an airline's channels;

  • which AI outputs make travelers more confident, and which still require verification; and

  • should an airline respond primarily through internal AI features, external partnerships, or a hybrid that supports continuity across both?

The broader issue wasn't whether AI is helpful. It was whether travel planning is being reorganized around a new coordination layer — one that now shapes how customers interpret options, compare carriers, and decide when they're ready to commit.

Design

The study used a retrospective artifact-based diary design focused on one real AI-assisted travel episode per participant. Rather than asking abstract questions about whether AI is useful, the research anchored every response to an actual conversation participants had already had while planning a trip, preparing for travel, navigating in transit, or solving a travel-related problem.

The final sample included 38 travelers across a range of travel contexts and planning moments — destination research, itinerary building, flight or hotel comparison, pre-trip logistics, in-trip assistance, disruption management, and advanced workflows like custom GPTs. The study generated 38 chat artifacts, 38 screen-recorded walkthroughs of real AI conversations, and 76 additional video and typed reflections on what participants trusted, what they verified elsewhere, and what they actually did next.

The design choice that mattered most was anchoring the study in real conversations participants had already had. Asking people abstract questions about AI tends to surface beliefs about the technology rather than evidence of behavior. Asking them to walk through a real conversation they'd just had — what they were trying to do, where they switched tools, what they verified, what they trusted — produced something different: a record of how AI is actually being used as a planning layer, not how customers say they use it.

The analysis focused on four recurring dimensions: what participants asked AI to do, the point at which they switched into airline or OTA channels, the kinds of information they still felt they needed to verify, and the presentation formats that increased or reduced confidence. The objective was structural pattern finding — identifying how AI-assisted planning is reorganizing exploration and purchase before the customer reaches the airline's owned ecosystem.

  1. Travelers form intent in external AI before any airline touchpoint, treating it less as a search tool than as the place where planning takes shape.

  2. AI earned trust through legibility — explanation, provenance, continuity — not personalization. Travelers verified elsewhere when stakes turned consequential.

  3. The strategic opportunity isn't to replicate AI inside airline channels. It's to remain authoritative when AI-shaped intent arrives.

Insight

Across episodes, three patterns emerged, each reshaping where airlines now sit in the planning journey.

1. External AI Is Becoming an Upstream Planning Layer

Across the diary episodes, AI most often entered the journey before any airline touchpoint did. In the large majority of focal episodes, participants turned to an external AI assistant before visiting an airline site or app. The prompts were rarely just "find me a flight." More often, participants used AI to translate a vague travel intent into something more structured: what kind of trip made sense, what sequence of destinations was realistic, which airports or neighborhoods were worth considering, and how to balance time, cost, convenience, and interest.

Occasional travelers used AI to reduce setup and uncertainty; more experienced travelers used it to compress work they already knew how to do. In both cases, AI was valued less for first-pass precision than for reducing the cognitive overhead of getting started.

What mattered strategically was that airline consideration often entered only after this compression had taken place. By the time participants reached airline or competing carrier channels, they often already had a working theory of the trip: a shortlist of options, a rough itinerary, a bundle of tradeoffs, and a sense of what counted as a good recommendation. AI wasn't just another source in the funnel. It was increasingly the place where the planning problem itself was being framed.

2. Trust Depended on Explanation, Provenance, and Continuity

Participants trusted AI most when it made the structure of the decision clearer. They responded positively when AI explained why one itinerary fit better than another, highlighted tradeoffs, or organized a messy travel problem into a smaller set of plausible choices. AI was most useful when it turned fragmented planning work into a coherent frame for decision-making.

Trust dropped quickly when the output touched volatile or policy-sensitive information without cues about source, recency, or uncertainty. Price, availability, baggage rules, airport-specific procedures, and partner-policy details were the most frequently rechecked items. Only a small minority said they would have acted on the AI output alone for a time-sensitive booking without verifying at least part of it elsewhere. Business-leaning and more frequent travelers were especially intolerant of ambiguity around these details.

Provenance mattered most when AI output and airline channels appeared to conflict. Participants wanted to know what the recommendation was based on, what assumptions it carried, and which parts still required confirmation. "Good AI" wasn't just conversational. It was legible enough to support confident handoff into booking.

3. The Strategic Opportunity Is Hybrid Orchestration

Participants didn't generally treat AI as a booking endpoint. They treated it as a planning and reasoning layer that helped them narrow and interpret options before moving into more authoritative environments. In most episodes, participants verified at least one consequential detail outside AI before acting — and most often, that verification meant returning to an airline channel for fare details, schedule confirmation, baggage rules, seat selection, or final booking confidence.

The switch point wasn't about rejecting AI. It was about understanding when the stakes had changed. AI was strong at narrowing, explaining, and organizing. Airline channels were still treated as the source of record when price, eligibility, timing, and transaction risk became real. Several participants moved back and forth repeatedly, using AI to interpret options and airline sites to confirm what was currently true.Participants didn't generally treat AI as a booking endpoint. They treated it as a planning and reasoning layer that helped them narrow and interpret options before moving into more authoritative environments. In most episodes, participants verified at least one consequential detail outside AI before acting — and most often, that verification meant returning to an airline channel for fare details, schedule confirmation, baggage rules, seat selection, or final booking confidence.

The switch point wasn't about rejecting AI. It was about understanding when the stakes had changed. AI was strong at narrowing, explaining, and organizing. Airline channels were still treated as the source of record when price, eligibility, timing, and transaction risk became real. Several participants moved back and forth repeatedly, using AI to interpret options and airline sites to confirm what was currently true.

The friction wasn't the existence of a switch. It was the lack of continuity across it. Participants often had to reconstruct preferences, restate constraints, or manually translate an AI-generated itinerary into booking terms. When that happened, the airline experience was being compared not just to competing airlines, but to the coherence of the AI planning experience that preceded it. The planning work had already been done; the owned channel often required travelers to do parts of it again.

Only a small minority of participants used advanced workflows like custom GPTs or saved travel assistants. The mainstream behavior was simpler: travelers wanted external AI to help them think, compare, and organize, and then wanted airline systems to validate, personalize, and complete the transaction. The strategic opportunity wasn't "own the entire AI planning journey." It was "remain authoritative when customers arrive with AI-shaped intent."

This points toward a hybrid model. An airline doesn't need to reproduce every external AI behavior inside its own channels. It does need to selectively mirror the strengths of external AI in the moments where decision friction is highest — clearer itinerary logic, side-by-side tradeoff explanation, conversational clarification of fare and policy questions, more intelligible presentation of dynamic travel facts. At the same time, the airline has reason to consider how verified data, itinerary structure, and policy information can be surfaced more accurately in external AI environments.

The clearest opportunities weren't abstract chatbot experiences. They sat at the handoff points: better continuity from AI-assisted exploration into purchase, better explanation of why one itinerary or fare bundle fit a stated need, and clearer cues about what could be trusted as current.

Impact

The work led to, four interlocking outcomes:

  • A new domain of CX research at United. This was the first study at United examining how customers actually engage with AI in travel — what they use it for, what they trust, what they verify elsewhere — establishing AI behavior as a CX research domain alongside the analytical and data-science work happening from other angles.

  • An internal strand: AI chatbot strategy. The behavioral patterns from the study — what makes AI output feel trustworthy, where customers expect provenance, what kinds of explanation increase confidence, where handoff to authoritative channels is needed — are now informing the development and training of United's customer-facing AI chatbot.

  • An external strand: presence in third-party AI. How the airline surfaces verified data, itinerary structure, and policy information inside platforms it doesn't own — like ChatGPT, Gemini, and Claude — is now part of the company's broader AI strategy.

  • A foundational reference at senior levels. The work has been presented to leadership across Digital, Product, and Customer Experience and now serves as the company's behavioral foundation for thinking about AI in travel — not as a single product to build, but as a layered strategy that preserves authoritative truth at commitment, adds AI-style explanation where friction is highest, and explores verified-data pathways for upstream environments.

The broader takeaway, beyond airlines: as customers increasingly form intent through AI before reaching any owned channel, the strategic question is no longer whether to build AI. It's how to remain the place where AI-shaped intent can convert into action.