AI · 2h ago
Study: 92% of AI prompts fail to reach 'good' quality
A PromptEval analysis of over 1,000 real prompts found an average score of 52 out of 100, with only 8% reaching 'good' (75+) and fewer than 1% 'excellent'. Defining the output format was the strongest predictor, adding 27 points on average. Robustness was the weakest dimension in 9 out of 10 prompts.
Meridian48 take
The data quantifies what many developers suspect: most prompts are mediocre, but small structural fixes yield big gains, making this a practical guide for improving AI interactions.
prompt-engineeringai-quality