Quantifying Cognitive Bias Induction in LLM-Generated Content