The speedy spread of disinformation on social media, news blogs, and online newspapers makes it extremely challenging to determine credible sources. Conventional machine learning approaches, especially supervised classifiers, are generally ineffective in coping with the dynamic nature of disinformation, which now often entails satire, irony, and interdisciplinary content. This paper proposes a novel method employing large language models (LLMs) in a weakly supervised setup for real-time disinformation detection, by measuring source quality and credibility signals, such as factual incivility, completeness of content, reference utilization, and neutrality. Our model judges content from different authenticity of online textual content across diverse domains of news, social media and blogs without over-reliance on human annotation. This approach fills current shortcomings in automatic detection of misinformation, which involve the ability to automatically handle English language and divergent data integration. Cross-validation using multiple datasets of different domains indicates the efficacy of the model to detect and tag potentially misleading material, thus optimizing resources for monitoring misinformation in real-time.