Generating simplified, evidence-based health research insights for health content creators
Health content creators were drowning in research and risking misinformation, so they needed a tool that could streamline research time with verified accuracy before posting content.
Science To People already built a version of their AI platform VeriSci to make complex scientific knowledge comprehensible and accessible, but it had problems.
When our team came on, we found the existing version had confusing navigation, duplicated systems, and unclear labelling throughout, with an already engaged outsourced developer team and a 4 month window to design a new MVP experience.
“Meta … would end its use of fact-checkers and launch a user-based ‘community notes’ system to flag inaccurate or misleading posts. The move has raised concerns among experts … that misinformation about science and health could increase on Meta's platforms.”
“Meta's fact-checking changes raise concerns about spread of science misinformation,” Harvard T.H. Chan School of Public Health, 2025
With investor support, Science To People needed designs for a platform with AI fact-checking and content creation tools that they can take back to their funders. I led the design of this phase and worked directly with product and UX teams while aligning with the client and development teams, ultimately reaching a 75% task-completion rate with an 85% stakeholder satisfaction score that helped secure additional funding before its expected pilot in 2026.
Health and wellness content creators face challenges in research efficiency, misinformation, audience engagement, and workflow integration, making it difficult to produce accurate, up-to-date content.
An AI-powered content creation tool to help health and wellness content creators generate engaging, factual content effortlessly by streamlining research, ensuring accuracy, and adapting to their personal tone of voice.
Features vetted sources from peer-reviewed research for users to have accurate, engaging, and culturally relevant content, reinforcing the company's mission to make complex science accessible in easy-to-understand images.
Features multi-functional layouts for creators to generate and edit desired content that is ready to be shared onto social media platforms, thereby streamlining the content creation process while maintaining creators' personal voice for their audiences.
Features intuitive navigation between the "Research Companion" and "Content Studio" components for creators to transfer their findings into engaging content, boosting user satisfaction and usability while maintaining creator voice and connection with their audiences.
Many great features, many more areas to improve on.
We first assessed the usability of the current version of VeriSci (aka MVP2) to understand what users experience with a designer's perspective. Here's what we learned:
Consistent & clean layout, clear differentiation between UI elements, source inclusion for research, and logical navigation on the home page were some of the strengths that we noticed during the audit.
Redundant information, confusing information architecture, unclear CTAs between work flows, and accessibility had large areas for improvment.
Simultaneously, we conducted a content inventory to understand the depth and breadth of the content in the very limited time we had. With more time, we'd conduct a content audit to assess the quality of the content and how to improve it. Our main takeaway was:
The navigational structure was very confusing due to unclear pathways, duplicated navigational systems, and uncertain labelling and terminology for CTAs.
How do users work and what do they like?
I explored the needs, preferences, and pain points of 3 different content creators and researchers. I then gathered key findings and results to understand their thoughts and frustrations:
Companies provide similar products and services, but none cater towards health content creators.
We then examined 9 AI platforms based on their products and services, including those that are not direct competitors, and evaluated them based on a few criteria (i.e. target users & platforms, onboarding, user flow & navigation, unique features, and customization). Out of the 9, we narrowed down to 3 to focus on:
Addressing constraints with user insights with business goals.
To ensure that we understand and address user needs, we prioritized our impact and efforts and produced product tenets that were critical for the product launch, which infused and balanced the user insights we learned with more business aligned goals.
Piecing the puzzle for visualization.
With the core tenets defined, I translated and organized them to make key information accessible and digestible. We also decided to start exploring some initial ideas of how the user onboarding process could look like as well.
Welcome, Dinah Stevens.
To make our research findings more tangible, I created Dinah, a content creator who is passionate about scientific integrity and promotes science communication through educational content, for a holistic view of the target audience that will keep our design decisions aligned with user needs.
Exploring 2 different options.
By this point, there were 2 different directions we wanted to explore as a team, specifically about how we wanted to present the research and content creation components. After some discussion, our team decided to experiment with both by expanding our roles to explore the two UI layout concepts (aka camps).
Laying the foundation for a cohesive display.
We retained some elements from MVP2's design library and borrowed some components from Shadcn.ui so that both camps can use the same components and features to maintain consistency. I focused on a couple of factors:
- Cohesive color palette of primarily green & purple — emphasizing feelings of reliability and creativity
- Legible and clear typography — ensuring clarity and accessibility for users
- Rounded and softer iconography — reinforcing user comfort and connection
Despite the fact that most office workstation desktops are still in 1080p, content creators are more likely to use more advanced computers and desktops for their work. As such, 1440p would be a reasonable size to work with.
Since the client’s product is a web platform application, we collectively agreed to having a stretch, 12 column screen layout to accommodate for the 1440p frame display. However, since the layouts of the combined and separate UI vary greatly, we did not settle on any specified margins or gutters.
We retained DM Sans from the client as the font choice for body text since it has a clean style to share intelligible information. After exploring a few font options for the brand’s header text, we decided on Space Grotesk since it is versatile with variations (i.e. bold, italtics, etc.) and matches the tone and image of the client’s brand: scientific, knowledgable, yet playful and bold.
We mainly considered two options for the icons we’d want to use, namely the Material Symbols & Icons from Google Fonts and the Lucide icon library from the Iconify plugin in Figma. We decided to adopt the Lucide icon library as our iconography due to its roundedness and general softer image to match with the DM Sans font.
We chose to adopt the pre-existing grayscale color scheme from shadcn/ui for our mid-fidelity wireframes as we have yet to take our work to high-fidelity.
Since we were not able to bring the wireframes or prototypes to high-fidelity during this phase of the project, I generated a color style guide for a high-fidelity mockup based on the client’s original branding slides and mood boards created by the design team.
We created two component libraries for the two UI concepts since they each have variations of the same features. In the two component libraries, I differentiated the components between research and content components, which I then categorized specific components under their respective features or functions. I also created a third component library called “Primitives”, which contain components that we kept universal between the two UI concepts.
Creating the first look of a new VeriSci.
Since I was working on the combined UI concept screens, I made design iterations to focus on:
- Clean, navigable layouts for effortless transitions between research & content creation panels
- Intuitive navigation between CTAs and key features on each panel
- Responsive elements to match the working styles of researchers and creators
We also explored some early onboarding wireframes to give the users and client a more cohesive image of VeriSci, despite our limited time towards the end of the project phase.
“Can't we find some way to combine them together?”
We conducted unmoderated, qualitative concept testing via Maze with 8 diverse content creators and tracked heatmaps and post-test interviews with surveys to give us objective data on workflow efficiency and user preferences. Key insights included:
Design A (Separated UI):
- ✅ Better interaction precision (lower misclick rate)
- ❌ Complex, branching navigation paths
- ❌ Multiple failure points throughout journey
- ❌ Slightly higher perceived difficulty
Design B (Combined UI):
- ✅ Clearer, more linear navigation flow
- ✅ Slightly lower perceived difficulty
- ✅ More predictable user journeys
- ❌ Serious interaction precision issues (very high misclick rate)
Leaving notes for future work.
After gathering insights, we annotated our prototypes on areas that users appreciated as well as areas that users found difficult. We provided specific annotations and recommendations in our hand-off file for future iterations and the client developer team.
1Accessibility is key.
Participants preferred having information in one screen rather than navigating through different screens. They also appreciated the visual hierarchy of the screens.
2A more realized workflow can help.
Participants felt the interface & content creation workflow could use improvement. They also wondered about the purpose and relevance of the “Edit in Studio” button to content creation.
Possible suggestions:
- Re-ideate steps to get from research chat to content creation studio
- Show more clarity on what's being edited (e.g. split text into text blocks and include “Edit in Studio” under each block to transport to Studio panel; provide a pop-up with short functionality description when hovering over the CTA button)
3More visuals, better clarity.
Participants didn't realize that a new answer was generated after refined searching.
Possible suggestion:
- Include visual feedback (i.e. loading animation) to show that new answer is being populated underneath
4Inclusive and reliable.
Participants compared the database-like interface to PubMed and Google Scholar, valuing the inclusion of sources and article types. They also appreciated having citations directly below the answer.
1Ease of use, ease of mind.
Participants greatly appreciated how having the Research Chat and Content Studio panels available on the same screen eliminates the need to navigate between pages and enhances their ease of use.
2Make it your own.
Participants valued the ability to add timestamps to scripts, generate different iterations of content, edit directly within the platform, and modify script tone. This gave them better versatility to cater their content towards their specific audiences.
3A feature that helps build success.
Participants expressed enthusiasm for the tool's versatility in content creation and editing, particularly for the “rewrite” and “fact check” features.
4Context always helps.
Participants wanted more context about the red & green boxes, particularly how they work. They also mentioned areas to improve the navigation with page layout & options.
Possible suggestions:
- A drag-and-drop feature for the highlighted sections can provide more interactive context
- Avoid biased/leading choices by removing colors of red & green
- Label the 1st option as “Current Answer” and the 2nd option as “New Answer”
- Have clearer labels for side editing bar (e.g. “rewrite” to “AI rewrite” with newer icon)
