Subjective product QA (PQA) aims to answer users’ questions regarding their experiences, feelings, and applicability of the product. Unlike factoid questions which answers are specific phrases or text spans, subjective PQA typically involves multiple viewpoints in various perspectives. Existing methods mainly rely on fixed templates to extract some spans. They are unable to integrate these diverse and even contradictory viewpoints, leading to incomplete and biased answers. To address these issues, we propose a new adaptive template-guided model with a three-step design. In detail, we first adopt a multi-turn dialogue to clarify users’ implicit and ambiguous needs in the question. Next, we extract subjective and objective question-related content from a wide range of product reviews and specifications. The content undergoes fine-grained sentiment analysis to capture the viewpoint statistical distributions. Considering the answers in PQA often require rigor, we construct a template corpus to enhance the answer’s controllability. Based on the question and external prior knowledge, we retrieve the most suitable template. It can be used to guide the LLMs to adaptively select and organizes subjective and objective clues. That can produce logical, coherence and fluency answers, better aligning with users’ needs. Experiments on the classic benchmark show the effectiveness of our model.