A company fine-tuned a pre-trained language model to create a customer service chatbot. They used a large dataset of complete, unedited chat transcripts between human agents and customers. After training, they observed that while the model can generate human-like conversational text, it often fails to directly and concisely answer specific user questions (e.g., 'What are your shipping fees?'). Analyze the likely reason for the model's failure to follow instructions effectively and describe the key characteristic of the data that should have been used instead.