Perceived Confidence Scoring for Data Annotation with Zero-Shot LLMs