top of page
Buddy.AI
Mental Health Chatbot
67%
user memory
retention
26%
avg. task completion time
83%
user satisfaction
Duration
3 months
Role
End-to-end Product Design
AI Chatbot for Mental Health
Background
Chatbots are an emerging technology that show potential for mental health care apps to enable effective and practical evidence-based therapies.
They were designed to identify mental health problems, track moods, deliver cognitive behavioral therapy (CBT), and promote positive psychology. Chatbots have been shown to effectively reduce the severity of mental health concerns and provide more convenient access to mental healthcare compared to traditional methods.
There are at least 10 mental health apps on app markets employing chatbots for therapy, such as Wysa, Youper, Woebot, Replika, etc. with over 500,000 downloads.
Improve User Perception of Mental Health Chatbot
Challenge
Despite positive reviews and general openness to interacting with chatbots in mental health settings, according to a study with 215 participants conducted on user perception regarding health chatbots by PubMed Central,
41%
of participants reported hesitancy to discuss health matters with a chatbot and half found chatbots untrustworthy.
This points to problems in user perception and experience interacting with chatbots in a mental health setting, which leads to compromised engagement. How might we improve the user perception of mental health chatbots?
Challenge
Idea
Buddy.AI Mental Health Chatbot
Idea
Onboarding - II
Onboarding - III
Onboarding - I
Onboarding - II
1/5
Following the recent release of ChatGPT, I found myself using it as therapy by confiding and asking for advice from the chatbot. I also tried other apps that use chatbots to provide therapy.
Therefore, I conceptualized Buddy.AI with the aim of exploring and improving user perception of chatbots in a mental health context.
Buddy.AI focuses on the design surrounding a chatbot to provide emotional support for the user through conversation.
I first designed a preliminary version of the app on Figma. The prototype then underwent rigorous testing with a group of six participants chosen in order to improve the user experience.
Process
Journey Overview
1
Research
Survey
Competitive Analysis
Evaluate Insights
2
Define
Personas
Journey Map
3
Design
Wireframes to Prototype
New Features
4
Usability Test
A/B Test
Moderated In-Person Test
Evaluate Feedback
Adjust Features
Research
Research Insights On User Problems
Apathetic Responses, Lack of Transparency & Limited Understanding
Survey
I surveyed 10 people who have used or considered using mental health chatbots to explore and understand their feelings and concerns which may affect their perception of chatbots for mental health.
I’ve discovered that the main user motivation for using mental health chatbots were anonymity when disclosing personal mental health concerns and convenience of access.
Consistent Support
10%
Affordability
20%
Anonymity
40%
Accessibility
30%
“I like how the bot won’t have the type of bias and judgment a human would, but it’s hard for me to trust a bot. It doesn’t have the experience or qualifications of a human.”
- Michael B.
On the other hand, users expressed a lack of trust with chatbots when it comes to receiving advice, either from previous negative interactions and bias or a limited understanding and unfamiliarity.
Overt monetization tactics
Inability to memorize conversation
Inappropriate, apathetic responses
Competitive Analysis
I tested 4 competitors' apps: Woebot, Youper, Wysa & Replika and read both negative and positive user reviews to understand the factors that shape better user perception & experience. This is the summary of what I found.
✅ What contributes to positive chatbot perception
.jpg)
Engaging Visuals
Friendly AI avatar
Constant Presence
⛔ What contributes to negative chatbot perception
Overt Monetization Tactics
Inability To Memorize Conversation
Inappropriate, Apathetic Responses
The competitive analysis has given me a lot of insights on what impacts positive and negative user perception & experience. I realized that there were some issues that could be improved for a better user experience; however it is not through design but rather on the technical/backend side, such as the bot's inability to memorize the conversation. Therefore, I've rounded down the problems where there are opportunities for better design below.
Top Problem Insights
🤖
Cold and impersonal interactions - the user feels a lack of empathy & appropriateness in chatbot response
❓
Lack of transparency on information source & qualification for health advice
🤔
Users also have a limited understanding of AI capabilities
🤑
Overt monetization tactics in conversation - leaves a bad impression of the chatbot
Personas: The Skeptic & The Hurt
Define
From the data I collected, I created two personas based off common user goals and pain points.
The Skeptic
Needs
-
Discreet conversations about general life issues & worries
-
Quality, proven mental health advice
Pain Points
-
Finds mental health chatbots confusing
-
Is skeptical about the reliability of chatbots' advice
Background
-
Has tried chatting to a MH chatbot once on his son's phone
Needs
-
Personalized, daily emotional support, reassurance & guidance
-
Non-invasive, non anxiety-inducing safe space to talk about mental health concerns
Pain Points
-
Finds mental health chatbots distant & apathetic
-
Feels rejected & uncomfortable when presented with paywall
Background
-
Has continuously jumped from one mental health app to another in the hopes of finding one that fits
-
Is depressed, anxious & traumatized
The Hurt
I also drafted the journey map of an user when they talk about a mental health concern to a chatbot to get a grasp of what they think and feel while going through this process. From there, I spotted opportunities for improvements at each step.
Features to Improve User Perception
Design & Iterate
The lack of trust towards chatbots due to the issues above is the root cause of negative user perception. Therefore, it is very important to establish and maintain trust to engage new and existing users alike.
Below are my early wireframes and prototype which have been improved based on the insights collected in the previous step. These improvements are aimed at enhancing positive chatbot perception and minimizing the negative.
Wireframes to Prototypes
wireframes to prototype
wireframes to prototype
1/1
New Features
Feature #1 - Set clear AI expectations & qualification assurances
Letting users know about the limitations & capabilities of the chatbot early on help establishing trust and setting clear expectations. Having this information accessible in chat serves as a reminder for users.
Feature #2 - Ability to customize AI tone
Users can choose from preset conversational tone options. This can be changed anytime and is accessed via "Custom Tone" button.
This feature allows more personalization, empathy & user autonomy for content output.
Feature #3 - Information source transparency & data proof
Users can tap this button to see the source of information or scientific research the chatbot's advice is based on.
This feature helps develop user trust in chatbot advice.
Buddy gifts the user a collectible daily. Once they collect enough of 1 type of collectible they can trade in to try 1 premium feature for a limited amount of time.
This feature proposes a more fun & engaging approach to monetization which less likely affects the chatbot perception.
Feature #4 - Daily check-in with gifts
Adjustments to Improve User Experience
Usability Test
I conducted usability tests with 6 participants and made adjustments according to the insights to improve user experience. I've included some before and after images for comparison.
A/B Test Results
When tasked with finding & tapping on buttons, testers took less time with model B. They also preferred the button placement & avatar size of model B.
2.8s
2.1s
2.5s
1.9s
A
B
Moderated In-Person Test Results
50%
of testers made a mistake in output and wanted an option to stop the chatbot from responding.
Testers also suggested using icons over text for displaying current tone. Therefore, I did that and also included a button the user can press to end the bot's reply.
Before
After
Before
After
Moderated In-Person Test Results
When questioned with the information in the pop-up post-closing, only 1/6 testers can recall what information was in the pop-up (Before).
Therefore, I broke down the content to make it more digestible through including visuals and changing the information architecture. As a result, 5/6 testers can recall the gist of the information in the pop-up (After).
Improved User Experience
Impact
67%
improvement in memory retention among users
83%
user satisfaction (survey) after adjustments
reduction in average task completion time
26%
The improvement time saved for users, memory retention and solid user satisfaction validated that my design was going in the right direction. The user tests I conducted gave insight to areas of improvement as well as areas of strength where the interactions were more seamless.
Constraints & Takeaways
Reflections
Constraint #1 - Participant demographic
The usability testing participants leaned towards young people in their 20s. This skewed the gathered data, as younger participants tend to be more comfortable with mobile applications and more receptive to adopting new technology. To ensure a more representative user base, a diverse range of participants would have been ideal, providing more accurate insights.
Constraint #2 - Technical limitations
As I was researching and finding solutions to user problems & pain points, I realized some of them can only be solved through improving back-end AI data training protocols, which cannot be fixed through UI design. An example of this would be the chatbot’s inability to memorize & draw from previous information provided by the user in chat.
Takeaways - Framing problems and shifting perspectives
Overall, through this project, I learned how to shift perspectives, come up with creative solutions and focus on problems that I can solve. I also learned a lot more about how to frame a case study and expanded my knowledge on the product design processes.
bottom of page