Propellic ran a study to see what U.S. travelers actually do
with artificial intelligence (AI), following them as they planned their trips
end-to-end across five stages: choosing a destination, researching hotels,
building a multi-day itinerary, finding day tours and the handoff to booking.
Participants selected and used their own tools, recorded their screens and
talked aloud while they worked.
The full report will publish on September 30. Take a
look at this preview of three of the findings now, because each one moves the
ground travel brands compete on.
AI points travelers to book direct, as they already
preferred
AI told 71% of participants to book directly with the
operator instead of through an agency. Independently, 68% said they preferred
booking direct anyway.
The reasons travelers gave were consistent: no middleman,
easier cancellations and refunds, plus open suspicion that agencies push their
own partner relationships ahead of traveler needs. The minority who favored
agencies cited group pricing and human expertise. When online travel agencies (OTAs) did appear in an answer, they appeared as the booking link rather than as a
brand the traveler had chosen.
Propellic's 2025 AI research found very similar results,
with roughly 80% of observed booking flows in Google AI Mode going
operator-direct against 13% to agencies. That suggests the behavior and the
results aren’t specific to Google AI Mode. It showed up across the tools
travelers chose for themselves.
For operators this is the best distribution news in a
decade, and it only pays out if you have a discoverable presence for AI to
find. We cover what to do with this information in the report, both from the
perspective of an operator and as an OTA.
Trust is built by structure, not by content
Seventy percent of travelers said they trusted AI recommendations more
than traditional search results. Another 25% said they trusted them about the
same, on the grounds that the information was equivalent and simply arrived
faster.
Then we asked why, and the answers came back about format,
not substance. Travelers pointed to a comparison table, a ranked list that
explained its own reasoning, a source link they could click, a follow-up
question showing the tool had understood the brief. Almost nobody pointed to
anything a brand had written.
Several never looked further than their AI chat either. A
typical session ran 2 to 4 prompts, and some took the first list they were
given, treated it as their shortlist and moved on. That puts an unusual amount
of weight on one response.
For anyone who has spent real money on brand voice, this
news stings. Copy that an AI cannot restructure into a table gets paraphrased
into something bland, and bland was the version travelers believed the least.
The fix is unglamorous. Write so a machine can rebuild your page as a table,
and say “why,” not just “what.”
Travelers settle on one AI tool and rarely look at a second
ChatGPT took 47% of all tool use. Gemini took 22% and Google
AI Mode 20%. Google's two surfaces (Gemini and AI Mode) are close behind in
total, just split between them, and Claude, CoPilot and Grok shared what was
left.
Across a full week of planning a real trip, most
participants stayed inside one tool. They didn't open a second AI to check the
first, the way they'd open a second tab in search. Every traveler we watched
who used more than one had ChatGPT among them, either as the tool they started
in or the one they came back to.
Three tools cover the large majority of travelers, so this
is an audit a marketing team can actually run in-house. Put your brand and your
category through ChatGPT, Gemini and Google AI Mode and read what each returns.
They draw on different sources and they answer the same prompt differently.
Then put it on the calendar monthly, so you can see the difference over time.
What else came out of it
Five other findings are in the full report, with the
evidence behind each one and what we'd do about it. There's also
audience-specific direction for hotels, tour operators, destinations and
agencies.
This covers things like how travelers handled the payment
step, what they thought of branding, and some of the biggest aspects of the LLM
responses that drove the purchase decision. Propellic also analyzed how they
responded to ads in LLM results, and how they worked through planning a
multi-day itinerary.
A key takeaway for travel brands
AI is reading your content before your traveler does. That's
the shift. There's a filter sitting between what you publish and who ever sees
it, and it doesn't weigh the same way that search weighed.
You can see it in all three findings. It sends travelers
past the intermediary and straight to the operator. It builds confidence out of
the shape of an answer rather than the words inside it. And it does all of that
inside a single tool, with nothing behind it checking the first answer.
Brand still matters. But the thing deciding what a traveler
sees first is a machine reading for structure, specificity and claims it can
verify, and most travel content wasn't written for that reader. The useful part
is how early this is. Most brands haven't adjusted yet, so the gap is still
open.
Where to get the rest of it...
The
full report will publish on September 30. Take a look at this preview of
three of the findings now, because each one moves the ground travel brands
compete on.