My experience building my first chatbot using voiceflow
Link to the chatbot: https://sites.google.com/view/worklink-scripted/home
About the Project
I created WorkLink as part of a university project for my Chatbots and Text Analytics course. The project focused on designing, building, and evaluating two different types of chatbots: a scripted chatbot and a generative AI chatbot.
The idea behind WorkLink came from a real issue I noticed while working on the project. Workplace rights and tax information in Australia is available online, but it is often spread across different official websites such as the ATO, Fair Work Ombudsman, and Services Australia. For someone who is new to working in Australia, this can feel confusing and time-consuming.
I wanted WorkLink to act as a simple starting point for people who need general workplace and tax information. The chatbot was designed with young workers, international students, recent migrants, and casual employees in mind, because these groups may not always know where to look first or what terms to search for.
WorkLink does not replace official advice. Instead, it helps users understand basic information and points them toward the right official sources.
The Two Chatbot Versions
For this project, I built two versions of WorkLink in Voiceflow.
The first version was WorkLink Scripted. This chatbot used a structured menu-based design, where users could choose from buttons and follow pre-written conversation paths. I designed this version to be clear, consistent, and easy to verify. Because the answers were manually written, I was able to include checked source links and control exactly what information was shown.
The second version was WorkLink GenAI. This version allowed users to type questions in their own words instead of choosing from menus. It used a Knowledge Base to generate responses based on the workplace and tax information included in the project. I designed this version to feel more flexible and conversational, especially for users who may not know which category their question fits into.
Building both versions helped me understand how different chatbot designs can create very different user experiences, even when they cover the same topic.
What WorkLink Covers
WorkLink provides general information about common workplace and tax-related topics in Australia, including:
- Tax File Numbers
- casual work
- sick leave
- payslips
- superannuation
- employment status
- income reporting
The chatbot was built to keep the information general and safe. It does not ask for sensitive personal information and does not provide legal, financial, tax, visa, or personal advice.
How I Tested the Chatbots
To compare the two chatbot versions, I conducted a small user evaluation with 30 participants. Each participant tested both WorkLink Scripted and WorkLink GenAI using the same set of workplace-related tasks.
The tasks included a simple question about Tax File Numbers, a casual-work scenario involving sick leave and superannuation, and a broader new-job scenario involving tax, payslips, superannuation, and reporting income.
After testing both bots, participants completed a survey about their experience. The survey looked at things such as task completion, ease of use, trust, source checking, satisfaction, and how well each chatbot handled simple and more detailed questions.
I also analysed the written feedback from participants to understand what they liked, what frustrated them, and which chatbot they preferred.
What I Found
One of the main findings was that both chatbots worked well for helping users complete the tasks. However, participants reacted to the two versions in different ways.
The GenAI chatbot was generally seen as more natural and flexible. Many participants liked being able to type their question in their own words instead of following a menu. It was especially useful for broader or more detailed questions where users needed information from more than one topic.
The scripted chatbot, on the other hand, was often seen as more trustworthy. Because it used structured answers and direct source links, participants found it easier to check where the information came from. This was important because the chatbot dealt with topics such as pay, tax, superannuation, and workplace rights.
For me, the most interesting part of the project was seeing this trade-off clearly. A chatbot can be smooth and easy to use, but that does not automatically mean users will trust it more. When the information is important, people also want to know that the answer is reliable and can be checked.
What I Learned
This project helped me understand that chatbot design is not only about making something conversational. It is also about choosing the right structure for the type of information being shared.
For simple or high-risk information, a scripted design can be useful because it gives more control over the wording and source links. For broader questions, a GenAI chatbot can be useful because it gives users more freedom to ask questions naturally.
The project also showed me that a hybrid approach may work best. A future version of WorkLink could use GenAI for natural-language questions, while still using scripted responses and verified links for important information that needs to be accurate and easy to check.
Skills I Used
Through this project, I developed practical experience in chatbot design, Voiceflow development, generative AI setup, Knowledge Base design, survey design, user testing, data analysis, and text analytics.
It also gave me a better understanding of how user experience, trust, and source transparency all matter when building chatbots for public-information support.
Louis Angelus 