Apple Augments App Store Ranking With LLMs

Apple researchers published a study, 'Scaling Search Relevance: Augmenting App Store Ranking with LLM-Generated Judgments,' and ran a worldwide A/B test to evaluate LLM-generated textual relevance labels. They fine-tuned a 3-billion-parameter LLM, generated millions of labels, and retrained the App Store ranker. The LLM-augmented model produced a statistically significant +0.24% lift in search-session conversion rate across 89% of storefronts, implying millions more downloads.