AI Travel Chat Bot

A cross-site AI experience designed to inspire trip ideas, answer travel questions, and seamlessly capture recommendations within a user’s plan.

Role:

Lead Designer

Duration:

6 months + project

Contribution:

Visual direction, User Testing & UX

Platform:

Web + iOS

Extensive research was conducted to assess the likelihood of users adopting this product, including surveys, concept validation, and usability testing. Insights from this work revealed a strong user need for quick, easy-to-access information at their fingertips—especially during active trip planning—while also highlighting friction in the existing experience. These findings directly informed a comprehensive design revamp, prompting stakeholders to prioritize the tool not only as a response to evolving traveler needs, but also as a critical step in keeping pace with emerging AI industry standards.

The AI Travel Assistant launched in late 2024, just ahead of Tripadvisor’s 2025 brand refresh, requiring a full re-evaluation of existing colors, components, and interaction patterns to align with the new design system. The assistant is a cross-site AI chat experience designed to help travelers ideate trips, get answers to planning questions, and save personalized recommendations directly into their trip plans—where decisions actually happen.

My role focused on defining the core user problems, shaping the end-to-end experience, and partnering closely with product, engineering, and brand to ensure the assistant felt intuitive, cohesive, and genuinely helpful throughout the planning journey.

Hello! 👋 I’m Ollie, your AI travel companion! Some questions you can ask me:

What is the best time to visit Honolulu?

Recommend a romantic restaurant in Honolulu

What are the best beaches in Honolulu?

Research

From pre-launch to post-launch, we conducted ongoing mixed-methods research—including surveys, concept testing, usability testing, and unmoderated sessions—to measure engagement, diagnose drop-offs, and validate improvements across devices and page types. Insights directly informed changes to prompting, chat flow, and entry points to improve completion and overall usefulness.

Finding 1
Respondents have mixed reactions about the AI assistant
39% said they were satisfied with the AI assistant. The top reasons include:
- Responses were helpful
- Recommendations were relevant
- Responses were easy to understand
Finding 2
When the AITA answers well, it can feel like a game-changer.
Participants were really impressed by how rigorous and sensible the AITA’s responses were, especially when it:
- Provided unique suggestions instead of recycling recommendations.
- Suggested other suitable options when more parameters are added.
- Included a rationale explaining the next best recommendation.
Finding 3
The AI assistant was easy to use for most respondents
Two-thirds of respondents said it was easy or very easy to use.

16% said it was difficult to use largely because the AI assistant did not work or did not answer questions adequately.

Before

After

Further Improvements

Problem: The FAB drove significantly lower engagement than in-page entry points.

The problem:

The introduction of a persistent entry point (FAB) on Hotel Review pages caused a significant, negative impact to hotel click revenue during the test, suggesting this type of entry point might be a bit aggressive for a page that is focused on booking/conversion.

The data:

On destination pages where the in-line entry point is visible, 70% of queries are initiated via prompts, while this makes up only 40% of queries via the FAB entry point.
Out of all the dynamic prompt interactions, 25% happen in in-chat entry (FAB) and 75% happen in in-page entry (existing entry point on Destination pages).
Entry point click through was much higher in the control (6%) compared to the test (0.24%). The same is true to message send rate with 40% of users successfully submitting a message in the control compared to only 7.4% in the test

Hypothesis:

Context Matters: The entry is close to room selection or about sections, aligning with moments users are most likely to have questions (e.g., after scanning rooms or reading high-level info).
AI Visibility Make it clear to the users that it’s AI who’s answering their questions and not the tour company/customer service.
Persistent & Thematic Placement: In all cases, the AI ask box is styled similarly to other help/FAQ modules, making usage intuitive for support—not transaction

Results:

Stronger Entry Point Performance

Contextual, in-page prompts drove +75% of all dynamic prompt interactions, outperforming the FAB as the primary entry mechanism.
Entry point click-through rate increased from 0.24% (FAB) to ~6% (in-page prompts), aligning with control performance.

Higher Chat Activation & Completion

Message send rate improved from 7.4% in the FAB test to ~40% when using contextual entry points, indicating clearer intent and reduced friction.
Users entering chat via in-page prompts were significantly more likely to ask follow-up questions and complete a conversation.