AI-Assisted Car Research Tool
CarGurus Discover Tool Redesign
AI has been greatly changed people’s life as well as the way we interact with products. With the boost of AI performance, it also expand the areas of product adoption and design innovation. With the new technology, how can we reshape the experience of online car research experience for auto purchase? Based on the CarGurus Discover Tool context, this project explore how can design benefit product experience and reshape user interaction through design innovation.
Date
2025
Keywords
AI Interaction
Prompt Engineering
Customized GPT
A Complex, Multi-Factor Decision Ripe for AI Assistance
Auto buying involves high financial commitment, long-term ownership implications, and a multitude of options. In 2025, CarGurus integrates conversational AI in its Car Discover feature. With a dialog-styled car research experience, users can input their requirements, needs and budget to get model recommendation and a filtered searching result.
If users don’t know what they need, how can they generate effective prompts for AI system?
Most AI-assisted car research tools—including CarGurus’ current discovery feature, begin by asking users to describe the car they want. This involves two sub challenges:
Prompt Fatigue
Faced with a blank chat interface, users often struggle to translate their goals into specific features—especially if they’re unfamiliar with automotive terminology or options
Unclear or Evolving Needs
Many users haven’t fully defined what matters to them. As a result, their initial prompts may overlook key tradeoffs or reflect unconscious biases.
To better understand the problem space, I conducted five semi-structured interviews to explore:
How users typically begin their car search and articulate their needs
What criteria they use to determine the “best fit” for a vehicle
What was missing or frustrating about their previous car-buying experience
Following the interviews, I invited each participant to interact with the CarGurus Discovery Tool using a think-aloud protocol. This allowed me to capture their internal thought process and reactions as they engaged with the AI-assisted experience. I specially concentrated on the initial process of needs framing.
Based on our interview findings, we observed that users gradually refine their needs through each round of research and interaction with the AI. Maintaining the conversation flow plays a critical role in helping users articulate and deepen their requirements over time. From these insights, we developed two behavioral archetypes that represent our primary user groups.
Observing that contextual cues are powerful triggers—they help users recall and articulate more detailed information. This presents a strong opportunity to design AI interactions that use context as a prompt to unlock deeper user input.
To help user quickly frame their prompt, I used AI generated a list of keywords based on the Features, Advantages and Benefits (FAB) framework to help user better structure their context. With this method, AI can better understand what trigger user made current decision and therefore brings new information and provides different strategies.
Comparing directly generate list result and suggestions, I separated the research process into 3 sections with different purposes with prompt setting. This ensures a continuously conversation with user for more information and make the reasoning process transparent and controllable.
Customized Result:
Engages the user in a natural dialogue to co-create context
Provides interpreted insights
Includes assumption transparency, letting users fine-tune scenarios.
