Tripadvisor: AI Chatbot

Background

.

Desk Research

As generative AI becomes part of daily life, helping users navigate it is increasingly crucial.

As of 2024, Generative AI (Gen AI) has become part of the daily routines for approximately half a billion people worldwide. Jared Spataro, Microsoft's Corporate Vice President of AI at Work, emphasized that non-work applications are important. "One of the things we're learning with our customers is the importance of offering clear and practical guidance on how to get started with AI," he says.

The most common are conversational AI.

Various generatvie AI in people’s lives

Meantime, Tripadvisor introduced an AI-powered Travel Itinerary Generator.

In July 2023, the world’s largest travel community Tripadvisor launched Plan with AI, an AI-powered itinerary generator that creates personalized travel plans.

What’s the goal?

Tripadvisor aims to help travelers quickly access key information, make faster decisions, and plan better trips with ease.

Plan with AI

What’s this?

This feature generates day-by-day travel plans based on user preferences such as destination, dates, and activities.

As passionate travelers...

We became curious about how Generative AI could improve trip planning, and decided to explore the feature more.

What benefits do travelers expect from Generative AI in trip planning, and how might we maximize the positive experiences?

Problem Definition

.

Insights from User Survey

We conducted a survey to identify improvements for Plan with AI, examine AI's role in trip planning, and better understand user preferences. Over five days, we gathered 42 respondents. Only one of the 33% of respondents who had used the Tripadvisor app had tried to Plan with AI. However, 56% of those who hadn’t used it expressed interest in using it for future trip planning.

1. Travelers avoid Gen AI for trip planning due to the complexity of adjusting results.

Q: 'Plan with AI' generates travel itineraries based on your preferences. Would you consider using it for your next trip?

Yes

Unsure

Top 3 Reasons of ‘No’ and ‘Unsure’

•

Adjusting results to my needs feels complex

•

Not familiar with AI operation

•

Preferring to plan a trip myself

2. Travelers need AI assistance with route planning, transportation details and expense calculation.

Q: Which parts would you most like assistance with AI for planning travel?

(Measured in people, total: 42; multiple choices allowed)

Analysis: Affinity Map & Empathy Map

Users find AI-generated itineraries impractical as they can’t be modified to fit their preferences.

We collected and categorized over 50 user feedback from the App Store, Google Play Store, and communities to identify areas for improvement in Tripadvisor. Affinity mapping confirmed issues from our user survey that travelers want more detailed travel arrangements between locations and budget considerations. Additionally, we uncovered that users struggle to modify itineraries and want more flexibility to tailor results based on their travel preferences.

“Modifying itinerary is not easy and complicated to use.”

“It would’ve been helpful to customize additional preferences such as budget, dietary restrictions, ...”

“Travel purposes are different. It should focus more on budget management.”

“... It would have been helpful to know whether you can walk to certain places, if public transportation is a good option, or if it’s recommended to rent a car or take taxis.”

Key user quotes from our affinity map in the background

Once we analyzed the Affinity Map to uncover pain points in the current Plan with AI experience, we mapped out users' emotions and behaviors using an Empathy Map for deeper insight. This helped us define our persona based on real user needs.

Me drawing the Empathy map on an online team meeting

The cluster of Empathy Map’s cards

Key user quotes from our affinity map in the background

User Persona and Journey Map

To embody the ideal user of Plan with AI, we created a user persona based on the insights from the problem discovery. By synthesizing the gathered data, our aim is to represent the user's preferences, pain points, and behaviors, allowing for a more effective design for our ideal user.

Name

Jinny

Age

32 years old

Job

Marketer

Eductaion

Communications

Location

Seoul, South Korea

"I'm traveling to London this September, but I don't want to spend too much time planning. I'd like AI to provide itinerary ideas with estimated costs, routes, and tourist recommendations so I can build on them and save time!"

Generating an itinerary

Entering travel preferences to Plan with AI.

Reviewing the itinerary

Evaluating if the generated itinerary suits travel goals.

Modifying the itinerary

Adjusting details to better fit personal needs.

Discarding the itinerary

Abandoning the itinerary after failing modifications.

"I hope AI suggests a solid plan based on my preferences."

"Looks alright! Let’s see if this itinerary matches my travel goals."

"I just need to tweak a few things to make it perfect."

"This isn’t working. I’ll just plan it myself."

Let’s refine itinerary details to better fit users’ travel goals and make modifications smoother and easier for an efficient trip planning experience.

Ideation

.

2x2 Prioritization Matrix

Prioritizing Ideas by Effort and Impact

We brainstormed concrete redesign ideas and prioritized them using a 2x2 prioritization matrix. Each team member rated the ideas on a scale of 1 to 5 based on Effort and Impact, and we plotted the average scores on the matrix to visualize their priority. We focused on ideas in the "Do Now" quadrant while discussing mid-range ideas further to determine their priority. This collaborative approach ensured alignment on our redesign goals and helped us make strategic design decisions.

Priority Matrix

Priorities of Feature Ideas

Click on a priority below to check out the corresponding ideas.

Q1. Do Now

C. Comprehensive transportation details

E. AI-generated itinerary feedback

I. Tutorial of Generative AI feature

Q2. Do Next

A. Modification of Itinerary Details before saving it

B. Budget calculation & management

L. Day-by-day map view with travel routes

Q3. Do Later

F. Seamless booking integration

G. Expansion of preference-based hashtags for recommendations

H. Direct booking links for transport options

Q4. Don’t Do

D. Centralized booking information page

J. Multiple itinerary generation

K. Distance-based filter

Considering the project timeline, we decided to focus on solutions prioritized in both the Do Now (Q1) and Do Next (Q2). These were grouped into two feature areas: the generated itinerary, enhanced with detailed travel information, and the AI chatbot, designed for intuitive itinerary modifications.

New
AI Chatbot Feature

Tripadvisor’s Generated Itinerary

Redesign: Generated Itinerary

Comprehensive Transportation Details (Q1)
Display relevant transport options and estimated travel times.
Day-by-day Map View with Travel Routes (Q2)
Visualize daily travel routes on an interactive map for better planning.
Budget Calculation & Management (Q2)
Provide estimated costs and expense tracking to help users stay within budget.

New Feature: Edit with AI

Modification of Itinerary Details (Q2)
Allow users to adjust AI-generated plans before finalizing them.
Generated Itinerary Feedback (Q1)
Let users rate and report satisfaction with AI-suggested plans for continuous improvement.
Tutorial of Generative AI Feature (Q1)
Help first-time users navigate AI-generated trip planning with an onboarding guide.

Solution Development: Generated Itinerary

Rapid Ideation to Final Concept

Before starting high-fidelity work, we rapidly sketched the prioritized ideas. We then evaluated, voted, and discussed each sketch to choose the final concept. This helped us stay aligned and move forward with a clear redesign direction.

How We Decided Solution Sketch in FigJam

Define key focus areas for the redesign.

Create rapid sketches and evaluate each idea.

Vote individually and decide Solution Sketch.

Solution Development: Edit with AI

Laying the Groundwork for Chatbot Interaction

To support diverse user needs with a flexible editing experience, we chose a chatbot interface for the Edit with AI feature. Our goal was to reduce decision fatigue and enable users to adjust their itineraries efficiently. To shape this interaction, we first analyzed the types of support users seek from AI throughout the editing process.

Interaction Goals Based on User Expectations

Step

Starting

Editing

Applying

User’s
Mindset

"I’m not sure what to edit or where to begin."

"I know what to fix. I just need to do it efficiently."

"I want to review and finalize my changes."

What Users Expect from the Chatbot

To reduce cognitive load and guide the next steps

To generate suggestions that match what I’m trying to do

To offer a final check with simple edit feature

Defined Role of the Chatbot

Clear, structured guidance

Contextual suggestions

Editable and reviewable outputs

Based on our analysis, we explored existing AI chatbot references that aligned with our defined interaction goals. Throughout the process, we found that offering options helps reduce users’ decision fatigue by minimizing the need for text input. This led us to design the Edit with AI chatbot with tap-based guided options.

Guided Option-Based Chatbot References

Structured guidance

Airbnb

Contextual suggestions

Ada

Editable and
reviewable result

Bing

Since users can receive different outcomes depending on the options they select, we first mapped out a user flow to anticipate various interaction paths. This gave us a clear roadmap of the screens and allowed us to structure interactions accordingly.

User Flow Chart of “Edit with AI”

The chart above illustrates a partial view of the full user flow.

Delivery

.

Final Design: UI for Digital Prototype

1. Plan with AI: Generating an Itinerary with Expanded Flexibility for Preferences

Group Size

Travel Preferences

By expanding the input fields to include group size and pets option, we enabled the AI to generate more tailored suggestions for like places to stay.

The original feature lacked options for travel preferences. To offer more tailored itineraries, we added the input fields of budget, food, and accommodation types.

To be

As is

To be

As is

2. The Itinerary: Editable and More Detailed Travel Information

Edit Features

Routes

Map

Getting Around

Budget

To support flexible itinerary editing, we added edit features to the generated plan. Users can now modify trip info directly from the main page, which redirects to the chatbot. A shortcut to ‘Edit with AI’ was also added to the navigation.

To address the lack of budget considerations, we added a Budget Breakdown section with categories like Food, Activities, and Transportation. It’s auto-filled based on the itinerary, but users can also adjust or add expenses for flexibility.

The original itinerary lacked information about how to travel between locations. We added transport modes, estimated time, and cost details to help users compare options and choose the best way to get around.

We added travel routes to the map and repositioned it to stay sticky at the top, allowing users to view plans and routes at a glance as they scroll through days. The map can also be flexibly folded or expanded using the drag indicator.

The original map showed all suggested places at once, overwhelming users and making it hard to plan routes. We added filters for day-by-day, suggested stays, and all places, helping users navigate the map according to their planning needs.

To be

As is

3. Edit with AI: Option-Based Chatbot

Starting

Things to Do

Place to Stay

New Itinerary

Tripadvisor didn’t support AI-powered itinerary editing, requiring users to revise their plans manually. To address this, we designed “Edit with AI,” a new chatbot feature that lets users easily modify their generated travel plans with AI support.

Users receive recommendations for things to do, select the places they like, and assign them to the itinerary. The AI then updates the itinerary accordingly and suggests follow-up actions, smoothly guiding users through the next steps.

Users receive stay recommendations, select what they like, and assign them to the itinerary. The AI then updates the itinerary and suggests next steps, naturally guiding users toward booking, aligning both user needs and business goals.

This feature helps users who aren’t fully satisfied with the initial itinerary by offering a tailored refresh. Instead of starting from scratch, they can adjust key preferences to receive a new, improved itinerary that better fits their needs.

To be

As is

Usability Test

Real users tested the redesigned AI-powered itinerary generator with new editing features.

To assess the impact of our design improvements to Tripadvisor’s Plan with AI, we conducted a usability test. Five participants took part in hour-long sessions that included interviews, task-based evaluations, and a System Usability Scale (SUS) survey to gather user feedback.

Testers completed the tasks following the given scenario.

During the session, participants were given specific tasks while we observed their behavior and assessed task completion ease. If a participant struggled or failed to complete a task, we followed up with an interview at the end of the session to understand the challenges they faced.

Task Completion Assessment

Criteria are Easily Done, Slightly Struggled, Significantly Struggled, and Failed.

New chatbot and travel details made trip planning easier and more efficient for users.

The redesigned Plan with AI and its new chatbot feature achieved an average SUS score of 82.5, rated "Good." While all testers failed to return to the chatbot’s first step after editing, and some noted unclear UI elements, overall feedback was positive and aligned with our design goals. Testers appreciated the added travel routes, transport info, and budget breakdown. They also found the option-based chatbot more intuitive than typing, which made trip planning easier and faster.

User Feedback

Positive

Negative

"Tapping options was much easier than typing everything myself."

"The flow was smooth. I just followed the options without thinking hard."

"Seeing routes with place info helped me visualize the trip."

"I wasn’t sure where this button would take me since the icon felt vague."

"The preference list was dense and hard to scan quickly."

"It felt like deep conversation wasn’t supported. It might be good to show that it is."

Conclusion