sophiehjlee

HOME
PORTFOLIO
ABOUT
- BIOGRAPHY
- CLIENT
- HISTORY
CONTACT

Real-Time AI FactChecker 2.0

We scroll, we consume, we assume. Misinformation isn’t just about false facts, it’s the invisible overload that quietly shapes what we believe.
To explore this, I designed a conceptual Chrome extension that makes fact-checking feel like a natural part of browsing, not an extra step.
Think of it like spellcheck, but for truth.

Version 1.0 used the Google Fact Check Tools API to surface verified claims and show where public fact-checks already existed. But the truth isn’t always binary. So, in Version 2.0, I reimagined the system not just as a tool but as a creative co-pilot, powered by Google Gemini, that helps users reflect on whether something feels credible or not. This isn’t a finished product. It’s a working prototype and research-driven design system, asking not just “What’s true?” but “How do we feel truth?”

Type

• 2024 Thesis @NYU ITP

• Web & Mobile UX/UI

• Predictive AI Fact-Checking

• Solo case study

Duration

• Version 1.0:

Nov 2023 - Apr 2024 (6 Months)

• Version 2.0:

Feb 2025 - Mar 2025 (3 Weeks)

What I Did

Product & Interaction design

Created a conversational AI-assisted browsing experience
Designed AI-driven user experience
Designed seamless interactive experience with generative AI
Conducted usability testing and iteration

Mentor's Testimonial

Yuliya Parshina-Kottas

Graphics and Multimedia Editor at The New York Times

"The way she's tackled real-time fact-checking here is really impressive.She's taken something that could feel complicated and made it surprisingly user-friendly, almost second nature, which is no easy task. The fact that she's built a tool that encourages people to think critically without feeling bogged down—that’s a huge win. It shows she really gets what today’s users need."

Sawad Brooks

Sr. Director of Product Design at Olo

"She's done an incredible job here. Every step in her prototype shows real thoughtfulness, making fact-checking feel straightforward rather than overwhelming. The way she's focused on keeping each feature purposeful and essential is exactly what builds strong user habits over time. It’s clear she's committed to truly valuable and easy to integrate into people’s everyday lives."

Problem

Where does our anxiety come from?

My story started with the anxious feeling of scrolling through endless news, posts, and updates, unsure of what’s true. The more I looked into it, the more I realized that I wasn’t alone; many people want to fact-check but feel it’s inconvenient or overwhelming.

Despite the availability of fact-checking tools, users rarely take the extra step to verify facts. Why?

The challenge isn't just misinformation, it's the unnoticed overload shaping our perception.

Solution

Not just a fact, a full picture:

Real-time AI Fact-Checker Chrome extension, powered by Google Gemini

This solution began as a simple real-time fact-checker using the Google Fact Check API, 1.0 version, designed to surface quick, in-context verification.

But truth isn't always binary (like true or false). In the improved 2.0 version, I reimagined the experience using Google Gemini's API, Fact Checker AI, not just as a backend tool, but as a creative collaborator.

Key improved features in 2.0 version:

01. A credibility score based on Gemini’s real-time analysis
02. A conversational summary explaining the reasoning
03. Related high-credibility articles for deeper understanding

Research

1) Data analysis of misinformation patterns

Using Python, I analyzed misinformation datasets to uncover recurring linguistic patterns and the psychological impact of misleading content.

2) Contextual UX & cognitive friction review

Reviewed academic research, industry reports, and cognitive science studies on misinformation processing and the effectiveness of existing AI-assisted fact-checking tools.

3) User behavioral analysis & interviews

Conducted in-depth user interviews to map fact-checking behaviors, cognitive biases, and the role of AI in influencing information trust.

4) Observational research on digital habits

Observed natural browsing behaviors, identifying unconscious interactions with misinformation and how AI interventions could seamlessly integrate into daily routines.

Big takeaways

✔️ People value fact-checking, but when overwhelmed with too much information, they tend to ignore it.

✔️ Fact-checking is most effective when it’s built into the browsing experience,

not something users have to seek out separately.

✔️ Instead of a simple “true” or “false,” showing a confidence scale helps users trust AI judgment more.

The Goal

How can we integrate fact-checking into everyday browsing, empowering users to critically evaluate information without the stress of uncertainty?

Ideation

Designing for the way we actually browse

We don’t stop to fact-check, not because we don’t care, but because it feels like a detour.
So I asked, "What if truth showed up right where we needed it?"

💡Creating Chrome Extension!

Instant Access: Fact-checks appear in real time, without ever leaving the page.
In-flow Interaction: Designed to fit naturally into the way we scroll, highlight, and explore.
Low-Effort, High-Trust: One click, and users get credible, AI-powered insight.

1) Observation & User Testing

I conducted user testing to measure how people engage with fact-checking in different scenarios.

✅ Without fact-checking tool:

Users rarely fact-checked content when the process required extra steps.

✅ With fact-checking tool:

When fact-checking was visually embedded in their workflow, users were 3× more likely to verify information. Frictionless access led to higher engagement and increased trust in AI-assisted verification.

2) Paper Prototype

Work flow V1

(2024 Prototype)

Built with Google Fact Check Tools API + HTML/CSS/JS

Used Google’s official Fact Check API to retrieve verified claim data in response to user input. Displayed results with light UI feedback to show whether a claim had existing fact-checked records.

→ This version focused on surfacing externally verified facts.

Work flow V2

(2025 Version)

Built with Google Gemini API + custom filtering logic + HTML/CSS/JS

Used Gemini to analyze linguistic tone, modality, and clarity in real time. Applied a custom rule-based layer to assess credibility levels and reflect that through color, motion, and ambient visual feedback.

While not a fully coded product, this prototype simulates how a real-time fact-checking AI could work behind the scenes. This is the flow I mapped, focusing on the trust journey, latency tension, and non-binary feedback structure

Iterations & Improvements

Enhanced Feature 1.

Interactive trigger: Highlight to trust

I observed users naturally highlighting text while reading by integrating fact-checking into this subconscious gesture.

Enhanced Feature 2.

AI that doesn't just answer, it explains

The earliest version offered binary results, "true or false." But users wanted more, not just what to believe, but why it matters. In this evolved design, Google Gemini becomes more than a backend engine. It steps forward as a guide, offering layered, conversational insights that feel more like dialogue than verdict.

Confidence score: A credibility rating based on real-time AI reasoning.
Conversational summary: AI-generated details that read more like a conversation than a judgment.
Live source links: Direct access to high-credibility sources for deeper, self-guided exploration.

Enhanced Feature 3.

Control the experience, Customizable UI

In early tests, nearly half of users said the persistent UI elements made them feel visually overwhelmed, like the interface was trying too hard to be helpful. So I stepped back and asked: What does helpful actually feel like? The answer was "choice". Now, with a simple toggle, users can turn floating fact-check prompts on or off. It's not just about customization

Enhanced Feature 4.

Visual cues that make curiosity, not just conclusions.

✔️ Used responsive elements:

Color-reactive sentences, animated uncertainty layers, generative motion for credibility shifts

Originally, the system gave binary labels, true or false. But trust isn’t black and white, and neither is truth. So, I created a dynamic credibility scale, designed with nuance in mind. Instead of handing users an answer, it gives them a signal, a gentle prompt to pause, think, and explore further.

Final Design Solution

Design System

Interface as emotional guide

UI that feels, not just functions. I designed the interface to act as an emotional companion, not just a display. I designed the full UI system from scratch, with a focus on emotional clarity and frictionless interaction. Every element from typography to transition rhythm was crafted to build trust and reduce the sense of judgment, especially in moments of uncertainty.

Rather than emphasizing authority or analytics, the interface responds gently and visually, using color, motion, and pacing to reflect the tone of each sentence in real time.

1) Adaptive UX for cross-platform consistency

AI-driven fact-checking adjusts to different screen environments without disrupting user flow.

2) Inclusive & accessible design

Applied WCAG standards to optimize readability, interaction clarity, and AI-assisted accessibility.

What I Learned

This project started with digital anxiety, the quiet kind that creeps in when we’re overwhelmed but unsure why. In designing with AI, I wasn’t just trying to fix misinformation. I wanted to restore confidence, not in machines, but in ourselves. If we can make trust effortless, maybe we can rebuild it, one scroll at a time.

As I wrap up, I see immense potential for growth-enhancing functionality, refining UX, and deepening AI’s role in fostering critical thinking. But this is just the beginning. Addressing digital anxiety is crucial, and through this project, I’ve aimed to make fact-checking effortless and intuitive. Much like our daily check-ins, ‘How are we feeling today?’ I hope this project sparks conversations about our digital confidence. By integrating real-time fact-checking into our browsing habits, we can navigate the digital world with greater clarity and trust.

🌟 Future Direction

Emotionally tuned fact-check with NotebookLM

What if fact-checking wasn’t just about what’s true, but how the truth is said? In the next version, I imagine a conversational interface powered by Google Gemini and organized with NotebookLM, where the same fact is expressed through different emotional tones.

Neutral and factual
Aggressive or polarizing
Empathetic and calming

Users wouldn’t just learn what's correct, they'd feel how tone shapes trust, and realize that belief is not purely logical, but deeply emotional. The goal isn’t to decide what’s right for them, it's to let them experience the invisible power of language.

ChatGPT_Image_Apr_26,_2025,_05_59_22_PM.png