Signal Who Belongs Together? Topical and Social Structure in Bluesky Starter Packs

Summary

Researchers Sima Adleyba and Onur Varol examine whether Bluesky's human-curated 'starter packs,' collections of accounts and feeds meant to help new users discover communities, capture meaningful topical and social structure. Studying 2,718 active starter packs covering roughly 144,000 accounts, they combine pack metadata, follow graphs, post histories and inferred demographic attributes, sorting packs into 17 topical categories. They find that pack members align more closely with their own pack's members and description than with accounts in other same-topic packs, indicating specificity beyond broad topical categories. The study also finds semantic resemblance between curators and pack members, while demographic resemblance varies by attribute and topic. Network analysis shows higher follow reciprocity within packs and between members and curators than across packs on the same topic, though a member's followers typically follow only a small share of other pack members, limiting audience coverage. The authors conclude that human curation captures coherent communities and could complement algorithmic recommendation systems for discovery.