This study aims to examine how widely publicized generative AI capability milestones are associated with shifts in participation and answer-side knowledge curation in a large public Q&A repository, with attention to how repository value is sustained through verification, refinement and endorsement.
Using a four-year panel of 4.2 million Stack Overflow posts (Jan 2021–Nov 2024), the author marks the initial ChatGPT release, GPT-4 and enterprise/vision rollouts as sequential capability phases. Day-level and post-level regressions estimate how platform usage, answer volume per question and community-evaluated answer quality are associated with these phases. Heterogeneity tests compare experienced versus novice users.
Public knowledge seeking contracts after the initial milestone and remains lower across subsequent phases, as reflected in declines in daily posting and post-level views. Conditional on participation, answer-side curation outcomes are higher on balance in the post period, including score and acceptance, with differences that vary across capability phases. Collaboration measures exhibit phase dependence, and experience-based heterogeneity indicates that post-period curation and evaluation patterns are not evenly distributed across novice and experienced contributors.
For knowledge managers, the results motivate designs that preserve peer verification and corrective refinement by directing attention to questions that still benefit from public scrutiny, rewarding stabilization work and supporting transparent attribution practices to maintain reusable organizational memory.
The study extends prior short-window evidence by using a multi-year panel that captures capability-phase differences and experience-contingent patterns that are difficult to detect in early post-launch snapshots. They position Social Learning Theory as an interpretive lens, alongside other mechanisms, to organize how public answer-side curation may be reallocated when private generative AI alternatives become widely accessible.
