Requirements elicitation is a critical activity in software development, yet traditional methods often fail to capture latent user requirements. As a result, the use of large language models (LLMs) for elicitation has increased. However, standard LLMs lack the ability to accurately represent individual user characteristics. This research hypothesizes that an LLM equipped with detailed user information can better reflect user characteristics and elicit latent requirements. Existing user-representation approaches, such as AI clones, remain costly to construct and maintain. Therefore, this study proposes a lightweight method for constructing questionnaire-based AI clones—referred to in this paper as user clones—and using them for user-requirements elicitation. Unlike comprehensive imitation models, the proposed lightweight approach focuses specifically on capturing the user’s product-related perspectives, thereby reducing construction and update costs by relying on limited questionnaire-based information. The method consists of two stages: (1) constructing of user clones based on user questionnaire responses and (2) eliciting requirements from the constructed user clones. To evaluate its effectiveness, the proposed method was applied to elicit product requirements at Company X, a quality assurance service solutions provider. The results indicated that approximately 90% of the requirements elicited from user clones were agreed upon by actual users, suggesting that the elicited requirements reflected users’ perspectives. These findings indicate that the proposed user-clone method can elicit requirements that align with individual users’ thoughts and may consequently capture latent user requirements.