Jennifer
Lo☻ong




design
play
art
photo
writing






about
CV
linkedIn



Portfolio under construction 🚧

CredBot

Inundation of misleading, false, and now unlabeled gAI content on the internet makes it very difficult to discern trustworthy sites from not-as-much ones. Particularly with politicized or sensitive subjects such as current news or medical information, users have to contend with not only self-vigilance but then proactively maintain verification processes. 

We created CredBot to automatically check site credibility, alert users of detected risk, and educate users using research-backed safety signals. 


Programs
Role
Time
Affiliation Figma \ Qualtrics \ R \ OpenAI API 
UI/UX Designer + Researcher in team of 3 
Sept-Dec 2023
CredLab, Professor Eni Mustafaraj, sponsored by NSF
Wellesley HCI Lab, Professor Orit Shaer 

Deliverables 1. Deployable chatbot Chrome extension 
2. Academic paper evaluating methods to deliver credibility signals 
3. Presentation at the 2025 Amherst AI in Liberal Arts Conference


I. Background

Our objectives were to: 
1) Evaluate whether AI can consistently interpret digital content in accordance with our credibility signals
2) Deliver signal evaluations in an educational manner to improve user web literacy 








II. Artifact Design  
 
Designing the chatbot as a browser extension, how we would present the scores as qualitative instead of quantitative 

We designed a chatbot so users could conversationally acquire more knowledge as they wanted, instead of being handed a block of text as one-and-done lectures. 




III. User Study

I developed our user study to evaluate the educational efficacy of our signals and our chatbot’s usability. 

Pre-Interview Questionnaire
In reference to both our set list of signals and previous literature, I created a set of nominal and open-ended questions measuring a participant’s web safety through web safety term familiarization and internet habits. 

Usability Testing  
I designed our user study for 2 phases: pre- and post-CredBot interaction. Users were given 2 fixed sets of 4 websites, with CredBot’s credibility scores calibrated by researchers to ensure consistency. The sites were chosen to test scenarios where the user searches for medical or political information. 

I created a Figma prototype of a Google Search Engine Results (SERP) page. SERP site rankings were randomized for each participant. 

Users were first permitted to explore a given site before answering a questionnaire evaluating their perceived credibility, before visiting a different site with CredBot activated. This was done so we could gauge how well the credibility signals and CredBot replaced the need for users to conduct external research. 


IV. Data Analysis  

Our pilot study involved 20 participants ages 18-24, skewing towards female and those with CS/CS-adjacent knowledge. 

Analysis of user interviews and survey results showed a general increase in web literacy following interactions with CredBot. A thematic analysis revealed users similarly expected 2 functions out of our tool: education and assistance. 


V.  Discussion, Future Work

Behavioral observation and typed conversations between participants and CredBot showed 2 subsequent user expectations after initial credibility evaluations: 1) definition, or what each signal meant contextually and 2) justification, why specific signals are relevant to a site.