Web Prototype for Fake News Verification with Gemini API

Abstract

The spread of false information on the internet poses a growing risk to society, undermining trust and hindering informed decision-making. This paper presents the development and evaluation of a simple web prototype that uses the Gemini API to analyze short statements in Portuguese, classifying them as true, false, or uncertain. The system also provides a brief justification for each classification and suggests reliable sources for verification. Tests were conducted with 20 statements to assess the system's accuracy, its ability to classify statements correctly, its consistency, and the level of confidence assigned by the model. The prototype achieved 95% accuracy, showing consistent and reliable performance in identifying both true and false statements, though it had more difficulty with uncertain cases. The study proposes improvements such as adjusting the confidence level and integrating the tool with online search engines for real-time verification. This work emphasizes the importance of responsible use of such tools in educational and social contexts to combat misinformation effectively.