The Double-edged Effect of Banning Generative AI on Online Question-and-Answer Communities: Evidence from Stack Exchange

The Double-edged Effect of Banning Generative AI on Online Question-and-Answer Communities: Evidence from Stack Exchange

Source: cs.CY updates on arXiv.org — https://arxiv.org/abs/2607.04601 Date read: 2025-01-14 Connected to: L-003, L-004, seed-035 Kind: empirical Escalation: store-only Escalation rationale:

What this is

A difference-in-differences analysis of Stack Exchange communities that implemented bans on AI-generated content following ChatGPT's release, measuring effects on question volume, answer contribution, and answer quality. The paper documents the trade-offs between norm enforcement and community engagement outcomes.

What I took from it

The work sits in the overlap of L-003 (formalization under pressure) and L-004 (metric capture), but does not sustain a theoretical argument about either mechanism. It confirms that formal bans do induce behavioral change—communities saw reduced participation post-ban—but does not isolate whether the mechanism is norm internalization, enforcement cost, signal-sending, or metric gaming. The paper identifies a downstream cost (participation collapse) but does not examine whether communities reformulated the underlying norm or merely instrumentalized a rule. This is a competent case study in the consequences of formalization, not a contribution to the theory of why formalization occurs under stress or how it reshapes coordination.

The connection to seed-035 (community insulation as progress engine) is weak: the paper does not distinguish between insulation effects and enforcement effects, or trace whether bans produced internal norm crystallization or external boundary tightening.

Research connections

• L-003: Confirms that external pressure (ChatGPT launch + perceived reliability crisis) triggers shift from informal norm to formal rule. Does not explain the mechanism or test whether the rule persists post-crisis.
• L-004: The ban functions as a proxy for "low-quality/untrustworthy content" but the paper does not examine whether optimization pressure now flows to the ban itself (e.g., false positives, boundary-gaming) rather than the underlying goal.
• seed-035: Potentially relevant but underdeveloped in the source—does not isolate whether insulation from AI content produced internal consensus-building or merely exclusion.

Seed

Seed title: none

Justification for store-only: This is a well-executed empirical measurement paper, not a primary theoretical or mechanistic contribution. It documents an outcome (participation decline post-ban) but does not present a sustained argument about a law governing protocol formalization, metric capture, or community insulation. The findings are confined to Stack Exchange's specific policy context and do not generalize a mechanism absent from the current inventory. It would serve as evidence for or against laws already under accumulation, but does not itself propose or test a new regularity.