VeriTruths Demo

AI-Powered Content Analysis

Explore VeriTruths, a prototype tool designed to combat online misinformation. Using AI, it analyzes articles for bias, potential inaccuracies, and source credibility. Interact with the demo below to see how it works.

This interactive demo simulates the VeriTruths analysis on a sample article within an iPhone-sized interface. Click through the different sections ("Full Report", "Profile") to explore the features.

Note: This is a non-functional prototype demonstrating the UI/UX concept. Data shown is illustrative.

How It Works & Technology

VeriTruths aims to provide users with a quick yet insightful analysis of online articles directly within their browser. Here's a breakdown of the process and the technologies involved:

1. Content Extraction & User Interaction

When a user navigates to a supported news article page and clicks the VeriTruths Chrome extension icon, the extension's content script activates. It carefully extracts the main article text, title, author information (if available), and publication source from the page's HTML structure (DOM). This extraction is designed to focus on the core content, ignoring sidebars, ads, and comments for a cleaner analysis.

2. Secure Backend Processing

The extracted content, along with metadata like the article URL, is securely sent to the VeriTruths backend API. This backend is built using Python with the Flask framework, chosen for its simplicity and robustness in handling API requests. User privacy is paramount; identifying information like IP addresses is anonymized using hashing and salting techniques before any potential logging or storage for analytics or caching purposes. The backend currently uses SQLite for simplicity during development, with plans to migrate to PostgreSQL for better scalability and concurrent request handling. Deployment is managed via platforms like Render or Vercel (currently hosted on Netlify for the demo frontend).

3. AI Analysis via APIs (OpenAI & Gemini)

The core analysis happens within the backend by leveraging powerful Large Language Models (LLMs) through their respective APIs, please note that to prevent hallucinations and bias, alternative databases such as Google Factcheck are utilized:

OpenAI GPT-4 / GPT-4o: The extracted article text is sent to the OpenAI API, specifically utilizing models like GPT-4 or the newer GPT-4o. Custom-engineered prompts guide the AI to perform several tasks:
- Bias Detection: Analyze the text for linguistic patterns, framing, source selection, and topic presentation indicative of political, corporate, or other forms of bias. The prompt asks the AI to identify the type and degree of bias (e.g., "Slight Left", "Strong Corporate", "Neutral") and provide specific examples from the text.
- Misinformation Assessment: While direct fact-checking against a live database is complex, the prompt instructs the AI to identify logical fallacies, emotionally charged language, unsourced claims, or statements that contradict widely accepted knowledge within its training data. It flags potential inaccuracies or areas needing further verification rather than making definitive true/false judgments.
- Summarization: Generate a concise summary of the article's key points for the report.
Google Gemini API: Potentially used alongside or as an alternative to OpenAI, the Gemini API can be employed for complementary analysis. For instance, it might be used for cross-referencing information, analyzing author credibility based on broader web context (if author data is provided), or offering alternative perspectives on bias. Using multiple models can improve the robustness of the analysis.

The backend orchestrates these API calls, sending carefully crafted prompts designed to elicit structured JSON responses containing the bias score, justification, potential misinformation flags with quotes, credibility indicators, and the summary. Error handling and rate limiting are implemented to manage API interactions effectively.

4. Frontend Display

The structured JSON response from the backend is sent back to the Chrome extension's frontend (the UI demonstrated above, built with HTML, CSS, and vanilla JavaScript). The JavaScript code parses this data and dynamically updates the UI elements: setting the width and text of the progress bars, populating the author and source details, listing the key insights, displaying the AI summary, and filling out the detailed sections in the "Full Report" view (including the bias scale, misinformation cards, author metrics, etc.). The interface is designed to present complex information in an easily digestible format within a collapsible side panel or popup.

5. Ongoing Development

The project is under continuous development. Future enhancements include refining the AI prompts for greater accuracy, potentially integrating dedicated fact-checking APIs, improving author credibility analysis by cross-referencing publication history, and optimizing the UX/UI based on user feedback. The goal remains to provide a powerful yet user-friendly tool for navigating the modern media landscape. Alternative databases will ensure cross-examination, if an article cannot be confidently analyzed an unknown status will be displayed and feedback is requested.

VeriTruths: Education Version

Beyond combating general misinformation, a specialized version of VeriTruths is envisioned to specifically aid students in their academic research and study processes. This "VeriTruths EDU" pivots the core analysis features towards enhancing academic integrity, critical thinking, and research efficiency.

Watch the demo above to see how VeriTruths EDU could assist students directly within their research workflow.

Student-Focused Features

VeriTruths EDU aims to be an indispensable tool for students navigating the vast landscape of online information for their coursework and research papers. Key proposed features include:

Source Credibility Assessment: Goes beyond basic bias detection to evaluate sources based on academic rigor, publication type (journal, news, blog), author expertise, and citation frequency within academic databases. Helps students quickly gauge if a source is appropriate for academic use.
Citation Assistance: Automatically extracts key information (author, title, publication, date, URL/DOI) needed for citations and potentially formats it into common styles like APA, MLA, or Chicago upon request.
Related Article Discovery: Leverages AI to analyze the current article's topic and suggests related scholarly articles, reputable news sources, or primary documents, helping students broaden their research and find corroborating or opposing viewpoints.
Key Insight Extraction: Identifies and summarizes the core arguments, methodology (if applicable), key data points, and conclusions of an article, allowing students to quickly grasp the essence of a source and determine its relevance to their research questions.
Study Aid Generation: Can create flashcards based on key terms and concepts within an article, generate potential discussion questions, or highlight areas that might be important for exam preparation.

By integrating these features, VeriTruths EDU strives to empower students to conduct more effective, efficient, and critical research, fostering better academic habits and deeper understanding of their subjects.