Strategic Simplicity Gets the Most: Evaluating Prompting Techniques in Non-Expert AI-Assisted Fact-Checking

Despite the well-known risks of bias, inaccuracy, and hallucination, large language models (LLMs) are increasingly used by journalists and fact-checkers. However, the potential of prompting strategies to help non-expert users mitigate these risks remains largely unexplored. This study considers seven such techniques in fact-checking tasks under real-world interface constraints. Four publicly accessible LLMs were tested to generate a manually curated dataset of 2520 outputs. A mixed evaluation framework combining human judgements and automated metrics was used to assess task completion, semantic accuracy, and factual consistency. The results show that prompts’ effectiveness depends primarily upon clarity, structure, and task-specific framing rather than complexity. Simpler prompts yield the most consistent and reliable outcomes across models and tasks. By situating prompting techniques within real-world, non-expert scenarios, this study contributes to wider discussions on information design, AI literacy and the practical collaboration between humans and AI in high-stakes fields such as journalism and fact-checking.