Emerging as a game-changing technology, generative artificial intelligence (GenAI) impacts various sectors, including healthcare, finance, education, and the creative arts. These systems are excellent at producing text, images, audio, and other modalities that resemble those of a human being because sophisticated deep learning architectures like transformer models drive them. Hallucination in AI, particularly in Large Language Models (LLMs), refers to the generation of plausible but inaccurate or fabricated responses. These errors can be categorized as intrinsic hallucinations (internal reasoning errors) and extrinsic hallucinations (outputs lacking justification from incoming data). Hallucinations present in three forms: factual inaccuracies, semantic distortions, and fluency-related discrepancies. Addressing hallucination entails various research methodologies, including prompt engineering like Retrieval-Augmented Generation, which retrieves authoritative information; self-refinement, where models iteratively enhance their responses for precision; model development utilizing context-aware decoding and knowledge graphs; and fine-tuning with factually accurate datasets. Methods like R-Tuning and Faithfulness-Based Loss Functions also help in improving factual alignment and minimizing hallucinations. Hallucinations in content generation diminish the usefulness and legitimacy of AI-generated content. Although prompt engineering has made significant strides, a structured understanding of how prompts affect hallucinations and consistent nomenclature are still lacking. Additionally, current mitigation strategies frequently focus on post-processing or superficial changes to prompts, failing to address the underlying causes of hallucinations that are ingrained in the training data and model design.