A report analyzing over 1,000 prompts used with large language models has revealed that only 8% achieved a "good" score (75 or higher). The most significant factor in prompt quality, contributing an average of 27 points, is clearly defining the output format. Robustness emerged as the weakest dimension across most prompts, with 9 out of 10 failing to handle ambiguous or unexpected input effectively. Prompts that are too short or rely on vague descriptors like "engaging" tend to perform poorly, while fields with higher stakes, such as healthcare, produce more carefully constructed prompts. AI
IMPACT Highlights critical areas for improving LLM prompt engineering, suggesting a focus on output formatting and robustness for better results.
RANK_REASON Analysis of prompt quality based on a dataset of scored prompts. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
