Cracking the code to the AI travel planner

How Tripadvisor’s AI Trip Builder leverages user insights and unique data to deliver fast, more relevant and customizable travel planning experiences.

By Alex Specht, Tripadvisor

Over the past few years, countless trip itinerary builder products have been brought to market, including Tripadvisor’s AI trip builder launched in July 2023.

Slight nuances aside, each of these products have historically operated in effectively the same way — they ask travelers to provide travel preferences via a chat UX or quiz, use AI to suggest recommendations that match these inputs, and format the response into a day-by-day itinerary. After a year in the market, it became apparent that while our product was being heavily used, it wasn’t solving the traveler’s problems in the right way. And so we broke with convention to find a solution.

In August 2024, we released a new and improved AI trip builder on our web platform which has since doubled the rate at which travelers save the resulting recommendations and improved surveyed customer satisfaction by 10%. So, how did we do it?

We focused on two key insights that were uncovered during user research: (1) travelers felt the recommendations lacked relevancy, and (2) travelers were in their early stages of trip planning and sought greater flexibility. These findings led us to rethink both the model we use to generate our recommendations and the user experience of our AI trip builder.

Let’s start with the model…

One of Tripadvisor’s primary differentiators is our user-generated content — namely, our traveler reviews. However, with the original version of the AI trip builder we did not use these reviews in the formation of our recommendations. Instead, we leveraged ChatGPT to identify recommendations that matched user inputs, and then ranked and selected these based on traveler ratings.

This resulted in us missing unique opportunities to identify highly relevant recommendations that ChatGPT wouldn’t know on its own. So, we fixed the problem. We developed our own recommender model that directly analyzes our millions of English-language traveler reviews to identify the most relevant matches to a traveler’s inputs. In doing so, we eliminated the middle man (Large Language Models) in the formation of our recommendations, thereby also reducing our latency from ~40 seconds to about ~6.5 seconds on average. This significant shift leans into what makes Tripadvisor unique by leveraging our community to drive greater relevancy, as proven by a 30% increase in the perceived quality of our recommendations. With these changes, we have been successful in delivering a better, faster and even further differentiated product.

Now for the user experience…

Most AI travel planners have designed their products based on the assumption that travelers want all of the work done for them — from what to do to when to do it.

But we learned through user research that our travelers really enjoy planning their trips. They want to spend time discovering things to do and places to eat, and they’re interested in receiving recommendations and guidance to help make the most of their vacation.

With this in mind, we redesigned how we display our recommendations to give the traveler greater control and customization. Specifically, we moved away from the standard format of a day-by-day itinerary and reorganized the recommendations into uniquely formed categories that are relevant to the destination and the traveler’s interests. This resulted in an easy-to-scan set of recommendations with the flexibility to save what you’re interested in and leave the rest behind. After refining your list, you can choose to have AI organize it into a daily plan, or just save the list to return to later. This added level of discovery and flexibility meets the traveler where they are in the planning journey and ushers them more gently through their unique planning process.

Before

After

The takeaway…

With these two updates, we have observed significant improvements to our key metrics and customer satisfaction — in other words, we can call it a success!

We weren’t afraid to challenge our previous assumptions or the status quo. We were intentional in leaning into what we do well. And last, but certainly not least — we listened to our customers.

To see the experience for yourself, visit or try it out in the Tripadvisor app.