“Leveraging information seeking behaviour to design human-centric chatbots for the vulnerable.”
In moments of critical need, whether navigating a diagnosis, caring for a loved one, or facing major life changes, the human desire for control and reliable information is paramount. Technological tools like AI-driven chatbots can now effectively bridge information gaps, empowering marginalised individuals and communities by providing access to credible information, timely resources, preventive measures, and crucial health information. Fortunately, over the past few years, the public health sector in the global south has embraced this technology to enhance knowledge dissemination, empower communities, reach the last mile and contribute to overall well-being.
While the use of AI is inevitable, one cannot emphasise enough the importance of a human-centric approach while designing chatbots that truly meet user needs and foster a stronger connection between users and technology. By integrating empathy and behavioural science to gain a deep understanding of user behaviour, the chatbots can be designed to cater to emotional mental models of satiating curiosity and intuitive information-seeking habits of the users.
At TinkerLabs this year, we got the opportunity to use our Human-Centered Design X Behavioural Science approach to design 2 chatbots in the space of ASRH and TB with UNFPA and Dure Technologies as our partners.
For both projects, we used our methodology to understand & deeply analyse behaviours related to
- Where and how are users currently seeking information
- What are the current fears/barriers to getting the information
- What are the inhibitions regarding the use of technology
- What are the scenarios, mental models and emotional need states while seeking information
This user deep-dive led us to some design directions that made our solutions more inclusive, intuitive, and responsive to the tailored needs of the users. Here are some examples:
- Users tend not to go out of their comfort zone to satiate their curiosity
While adopting a new behaviour, humans prefer to take the least resistant path, something that requires the least cognitive effort and is easy to understand, process and navigate. This can be explained through the BE principles like the “status-quo bias” and “cognitive ease”.
In our work with adolescents seeking SRHR information for Just Ask Chatbot, for example, we observed that adolescents approach their friends, family, and young relatives and sometimes Google for information related to SRHR. Even though they are not satisfied with the responses they get, they do not go out of their way to seek more reliable and credible knowledge. We further recognized that downloading a new app or accessing an unfamiliar website is a barrier to access as it requires putting in extra effort that they are not used to. Therefore, we leveraged existing platforms that are familiar and readily available to the users, like WhatsApp. Users are not just comfortable with WhatsApp, but it feels personal, friendlier and more transparent that they can easily access the chatbot without drawing any attention towards themselves.
2. Users feel shy to ask specific questions or find it difficult to articulate nuanced questions
The friction cost of articulating the question is a big hindrance to receiving exact, correct and relevant answers. Users evaluate effort vs output — “After putting in all the effort to ask the question, will the answer be worth it?”
Knowing this, we ensured that the users have an alternate option to get information without the need to articulate queries themselves, but, at the same time, be able to get to the answer within minimal clicks (less than 3 clicks). In addition to a user-initiated conversation flow, we built a structured flow that allows the user to explore topics of interest on their own.
However, during testing our SRHR chatbot, we observed that the users feel confused about what information they should expect within a certain topic — they wonder which topic is the most relevant to choose from for them to get the answer that they are looking for quickly. To mitigate this confusion, we ensured that the user gets clarity on topics in the form of FAQs and finds a question that seems relevant to them.
3. Users do not know how much information there is to seek
After getting the specific information they were seeking, users tend not to ask follow-up questions — sometimes they do find it difficult to articulate or feel shy, but more often, they do not know what else there is to know. To address this, we designed our conversation flow in a way that after each piece of information is shared, there is a set of related questions for the user to explore.
Curiosity is a trail of hidden questions, where users have visibility to the very first question and the door to the next question opens up only after getting an answer to the previous one. The related questions feature got validated during user testing, where users expressed that they didn’t know that they could have a question of this kind but were happy that it got addressed immediately! So there must be a trail of breadcrumbs, encouraging the user to know more.
4. Users feel overwhelmed with TMI
One of the key barriers to seeking health information online is the “information overload” and “decision fatigue” it causes when users receive an excessive amount of data, making it challenging to process and comprehend relevant information.
It is crucial to deliver information in a simplified, broken-down, and non-overwhelming way, sharing the right information at the right time with minimal clicks.
We addressed this in the following ways:
- Personalise the conversational flow as per the user: When designing the TB chatbot, we created a structured information architecture completely based on the trail of questions gauging the type of user and where they lie in the TB journey (know very little about TB, are worried about what a positive diagnosis means, what does life look like after a diagnosis, how to maximise the treatment phase, when to know the treatment has ended). Through this, we were able to deep dive into their specific needs, questioning mindset and the intent behind those questions, which enabled us to design concise and relevant answers.
- Make it a 2-way conversation: We designed the flows by breaking long pieces of information into consumable bits, restricting responses to 2–3 small messages and nudging users to click and find more related and relevant information, rather than bombard them with monologues.
- Leverage visuals to simplify and aid content
5. Users have unique fears about privacy that can be easily overlooked
Another prevalent barrier to seeking health information is the fear of others knowing about it — whether it is a teenager fearing their parents while seeking information about wet dreams or a young wife fearing the in-laws while trying to find ways of contraception. Potential loss of privacy outweighs the potential gain of the information (described as “Loss aversion”) and this becomes a barrier to the use of technology. We learned that while addressing generic data privacy concerns and ensuring transparency and credibility, it is also essential to design features that build a feeling of safety by understanding the nuances of user realities.
For instance, when designing the SRHR chatbot, we understood that adolescents and youth may not have access to a personal device, yet may still want privacy. Hence, we designed a bot-prompted “clear chat” feature for users to clear their chat history and prioritised only user-initiated chats, so users have full control over their interactions.
We gathered a plethora of such user insights from our on-field research during the discovery and testing phase. By paying attention to the smallest of the behaviours, the context in which they occur and trying to understand why these behaviours occur, we were able to arrive at design features that are unique to the specific users and foster a deeper connection with the chatbot. These seemingly small features remove hidden barriers, elevate the user experience and build trust.
