Insight

Across episodes, three patterns emerged:

1. Participants used AI for a wide range of planning work, but the prompts shared a common pattern: they weren't searching, they were asking AI to help them think through what kind of trip to take.

2. Participants trusted AI for structuring decisions and explaining tradeoffs but turned to direct sources for anything tied to actual commitment.

3. Participants reported moving repeatedly between AI and other sources during planning, treating AI as one input among several rather than the place where decisions got made.

1. Participants predominantly asked AI to help them think, not search.

The kinds of planning work participants brought to AI varied widely. The chart in the Design section shows the distribution: itinerary building, destination research, flight and hotel comparison, pre-trip logistics, in-trip assistance, disruption management, and advanced workflows. Despite the range of topics, the prompts followed a recognizable pattern.

The conversations rarely opened with a search-style request. Instead, participants asked things like:

"We have a week in May, our daughter is six, my mom is coming with us — where should we actually go?"

"Help me figure out if Lisbon and Madrid in the same trip is too much."

"Is it worth flying into a smaller airport to save the layover, or is that going to add hassle?"

"What's the smartest way to do a four-day trip to Tokyo without losing two days to jet lag?"

The prompts asked AI to help reason through the trip. They didn't ask AI to find a flight or pull up a hotel list. AI's responses came with structure attached — proposed itineraries with day-by-day rationales, lists of tradeoffs to weigh, recommendations explained against the constraints the participant had named.

This pattern aligns with cognitive offloading research. Across other contexts, studies have documented users delegating not just memory and computation to AI, but the formation of frames and beliefs themselves (cf. e.g., the "belief offloading" framing in Hertwig & Engel, 2023). What the travel data showed was the specific shape this takes when the work at hand is planning a trip. Travelers used AI to externalize the part of planning that involves figuring out what you actually want — what trip is worth taking, what tradeoffs are worth making, what counts as a good answer.

2. Participants trusted AI for thinking, not for committing.

Participants described their trust in AI in terms that tracked the kind of work AI was doing for them. The diary reflections returned to a similar set of themes:

"It helped me see what I hadn't thought about."

"It was useful for narrowing things down — I'd been going in circles before."

"It explained things in a way that made me feel like I understood what I was choosing."

The trust came from legibility. Participants described AI as useful when it made the structure of a decision clearer — when it surfaced tradeoffs they hadn't named, organized fragmented planning into a smaller set of choices, or explained why one option might fit better than another. The trust didn't depend on AI being right about specifics. It depended on AI being useful for thinking through the planning problem.

But participants also described where that trust stopped. When the reflections turned to what they did after the AI conversation — what they verified, what they double-checked, where they went next — the pattern shifted:

"I wasn't going to book without checking the actual prices on United."

"For baggage stuff I always go to the airline site. I don't trust anything else with that."

"I'd ask ChatGPT to help me think about it, then I'd go to Google Flights to see what was actually there."

The verification behavior was concentrated around a recognizable set of items: real-time prices and availability, baggage and policy details, airport-specific procedures, and partner or codeshare rules. The pattern was consistent. Participants trusted AI to help them think through the trip and turned to direct sources for anything that had to be true on the day of travel.

This conditional, stakes-sensitive trust is consistent with established research on AI in other high-stakes domains. Studies of AI use in healthcare and financial decision-making show the same pattern: trust holds when the stakes are low or the stakes belong to someone else, and drops when the stakes become personal and consequential [citation: e.g., Hudecek et al. on diagnostic AI trust shifts; Northey et al. on financial AI stakes]. What the travel data showed was where that break-point sits in a planning journey — and what kinds of decisions travelers consistently moved out of AI to make.

What participants wanted from AI was visibility into what it was basing its recommendations on. The strongest trust signals across the reflections involved being able to see why AI had suggested what it suggested, what it was weighing, and what assumptions it carried. The reflections didn't dwell on personalization or warmth. They dwelled on whether AI's reasoning was visible enough that participants could decide for themselves whether to act on it.

3. Participants moved between AI and other sources throughout planning

The reflections described AI as one source among several rather than the place where decisions got made. Participants reported using AI alongside Google Flights, airline sites, OTA platforms, friends, and prior travel knowledge — moving between them as different parts of the planning surfaced different needs.

The pattern recurred across the reflections:

"I'd use ChatGPT to think through the options, then go to Google Flights to see what was actually available."

"AI was good for narrowing things down, but I always ended up on the airline's site to double-check the details."

"I went back and forth a few times — asking AI a question, checking the answer somewhere else, then going back to AI with what I'd found."

"I'd been talking to friends about it for a while, then I asked ChatGPT to help me sort through what they'd said."

The movement wasn't a sign that participants were dissatisfied with AI. The reflections described AI's role as specific and useful but bounded. AI was the tool participants reached for when they needed to think through what to do, organize options, or get an explanation of a tradeoff. They reached for other sources when they needed current information, verification of a specific detail, or the authority of an official channel.

Across the sample, the typical planning episode involved at least two non-AI sources alongside the AI conversation itself. Participants described this as the normal shape of how they were planning trips, not as a workaround.Across the sample, the typical planning episode involved at least two non-AI sources alongside the AI conversation itself. Participants described this as the normal shape of how they were planning trips, not as a workaround.