The rapid integration of foundation models like GPT-4o, Gemini, LLaMA, Qwen and DeepSeek-R1 into real-world applications has revolutionised AI systems, providing impressive performance and capabilities across domains [1, 2]. Foundation models now underpin language translation services, virtual assistants, content recommendation engines, and increasingly, reasoning-oriented systems such as ChatGPT Search. Unlike traditional search engines, these LLM-mediated systems contextualise, synthesise, and interpret information beyond surface-level retrieval—influencing millions of users worldwide. However, they often function as opaque systems (black boxes), making it hard for users and developers to understand their decision-making processes [3, 4]. This lack of transparency undermines trust, hinders explainability, and raises ethical concerns in human-AI collaboration. This paper examines a possible theoretical framework for enhancing AI explainability by applying René Descartes’s method of doubt—a philosophical approach characterised by systematic scepticism and the critical examination of beliefs until only undeniable truths remain [5]. We introduce Cartesian Methodical Doubt (CMD) as a design-theoretic cognitive framework that guides reasoning and decision-making, while also functioning as metacognitive scaffolding—supporting reflection on those reasoning processes to enable transparency, explanation, and alignment in human-AI interaction. Exploring this perspective, heuristic principles—pragmatic, rule-of-thumb methods relevant to both human reasoning and AI communication [6, 7] are considered to highlight possible parallels between human and machine reasoning. The framework is examined in relation to human-centered design and co-design methodologies, considering its potential role in improving AI interpretability, transparency, and reasoning while fostering further investigation into its theoretical implications for user trust, human-AI interaction, and decision-making under uncertainty.