Leo AI chatbot
Defining features to build the Brave Browser's AI chatbot from the ground up
Foundational ResearchInterview StudyStakeholder Workshops
Product managers and designers at Brave wanted to know what generative AI features would deliver the most value for users to build the new browser chatbot experience.
Building the wrong features will cost time and revenue and Brave may fall behind its competitors in the AI race.
A generative research study to identify valuable features and pain-points in existing generative AI experiences. Scoping actionable opportunities to build features.
Product Impact: New features to be built based on research
User Impact: Brave users are excited about new AI features
Business Impact: A new feature planned to be monetized
I interviewed the VP of design, the product lead for the desktop browser and a senior product designer who were in charge of the Leo project. I gathered a list of questions and assumptions they had and grouped them into 3 key research questions.
They had questions like:
"What is the intended value we aim to bring?"
- Product lead, Desktop browser
1. How do users use browser-based chatbots and what value do they derive from them?
Rationale: To ensure that Leo is designed to deliver on expectations and provides the same value.
2. What are the user pain-points/missing features in competing AI chatbots?
Rationale: Finding features to build that go beyond what competitors are offering.
3. How do users use browser-based chatbots for writing text and what are the unmet opportunities in this space?
Rationale: Stakeholders believed that having a writing assistant will make the chatbot more appealing to users - we wanted to test this hypothesis.
I interviewed 17 participants who were regular chatbot users.
I screened and recruited participants on HubUX and Userlytics using a screener survey. Participants were a mix of working professionals and students.
Most people we recruited used ChatGPT and few used other chatbots.
I chose a semi-structured interview + contextual inquiry approach.
INSIGHT 💡
Prompt suggestions were helpful during information search.
Users often clicked on the prompt suggestions to dig deeper into topics they were looking for information about and reported that this was an easy to use interaction.
IMPACT 🎯
Existing designs for prompt suggestions were validated.
I validated the existing design concepts which has prompt suggestions.
I recommended emphasizing them when users ask Leo for information from the web.
The design team implemented these recommendations.
.png)
Leo's full page chat with follow-up prompt suggestions
INSIGHT 💡
User used chatbots to generate summaries of web pages.
When using the bing-chat sidebar during web browsing, users asked for article/webpage summaries when there was a lot of text.
The friction could be reduced by giving the summaries upfront.
IMPACT 🎯
Leo was redesigned to show webpage summaries without prompting.
Leo will now automatically summarize webpages with lots of text without needing to ask for it.
.png)
Webpage summarization feature
I organized workshops with design, product and engineering teams to brainstorm new features
The features we came up with during the workshop were added to the product roadmap.
INSIGHT 💡
Users want to dig deeper into specific points in chatbot responses.
When users wanted to find more information about a point within a chatbot response, they would often go do a web search.
Switching into another tab and typing a query takes time.
User flow when users want to dig deeper into chatbot responses
IMPACT 🎯
In the workshop, the team came up with an idea to enable users to follow up on points without leaving Leo by opening a modal.
The feature is now part of the future roadmap for Leo and is being designed.
Chatbots are used extensively to write content like blog posts, social media posts, essays, articles, emails, slogans, marketing text, cover letters. It is an important use case.
However, users faced lots of issues around chatbots not capturing the appropriate tone and context for documents.
User Quotes:
“[ChatGPT] doesnʼt sound like me”
“I would not use words like that”
“you want to express [what] you want and ChatGPT doesnʼt understand the context”
INSIGHT 💡
Users want to personalize their chatbot's writing to sound more like them
Those who frequently use chatbots for writing often had complaints around a lack of personal touch in the tone and context produced by ChatGPT or other LLMs. This was especially true when they used it to write communications like emails, essays and social media posts.
I recommended adding this as a new feature for Leo.
IMPACT 🎯
From the workshop, the AI team prioritized personalization as a key feature in Leo's next update.
📣 Roadmap announcement: Sep 2023
Product Impact - Helped define product vision
4
Existing concepts validated
3
New features on the roadmap
1
Feature to be monetized
User Impact - Users are excited about upcoming features
Personal Impact - The team enjoyed collaborating with me and I with them
"Tarun was a pleasure to work with. He displayed sufficient research knowledge at each stage of the research project such as data collection and insights synthesis, which contributed to the success of the research project. Tarun was a team player and collaborator, gaining a deep understanding of the research project background and objectives by working with stakeholders. Overall, Tarun was a valuable addition to our team, I’m excited about his future in UX research."
- April Yang, Sr. UX Researcher