Master of Deceit: Comparative Analysis of Human and Machine-Generated Deceptive Text

Deception, the intentional act of creating false impressions, has long been studied in human interactions.With the emergence of AI and large language models (LLMs), deception now extends to machine-generated content, raising concerns about distinguishing between human and AI-created content.In this study, we compare deceptive and truthful texts produced by humans and LLMs (GPT-3.5 and GPT-4o) using two datasets; a crowdsourced online Review Dataset and a transcribed interview dataset (MU3D).We replicate the data generation process with LLMs, introducing personas into prompts to examine linguistic differences and potential biases.Using LIWC, we analyze word choice, complexity, and cognitive patterns across human-and LLM-generated deception and truthful texts.Our findings show that LLM-generated deception differs significantly from human deception, exhibiting greater verbosity, formality, and lexical sophistication, while human deception is more socially driven, relying more on social references, interpersonal cues, and natural conversational patterns.Despite improvements in LLMs, context-dependent biases remain embedded in LLM-generated texts, emphasizing the need for stronger bias mitigation strategies and responsible AI deployment.Our study identifies key linguistic markers that differentiate LLM-generated from human deception and highlights the importance of assessing hidden biases and potential risks in AI-generated deceptive text and misinformation.