Hallucination vs interpretation: rethinking accuracy and precision in AI-assisted data extraction for knowledge synthesis