When Is Self-Disclosure Optimal? Incentives and Governance of AI-Generated Content
Juan Wu 1 , Zhe (James) Zhang 2 , Amit Mehra 2 1 University of Science and Technology of China 2 University of Texas at Dallas · 2026
Read When Is Self-Disclosure Optimal? Incentives and Governance of AI-Generated Content by Juan Wu 1 , Zhe (James) Zhang 2 , Amit Mehra 2 1 University of Science and Technology of China 2 University of Texas at Dallas (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-18654.txt.
Introduction
Juan Wu, Zhe Zhang, and Amit Mehra study when a maker should tag their work as made by AI. Smart tools have cut the cost of text, shots, and clips, so feeds fill with work of mixed worth. Sites now ask makers to tag AI work. Tags are backed by checks that can miss, and by fines for those who skip the tag. Makers must weigh the gain of a hid tag next to the risk of a fine.
The team model that choice and ask what rules make self-tags work. If the fine is too light, few will tag. If the check is too weak, tags lose their worth. Their work maps the mix of check and fine that makes tags a fair sign. Sites could use that map to set rules that keep trust in feeds.
Catalog record
| CREATOR | Juan Wu 1 , Zhe (James) Zhang 2 , Amit Mehra 2 1 University of Science and Technology of China 2 University of Texas at Dallas |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 72 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-18654
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
When Is Self-Disclosure Optimal? Incentives and Governance of AI-Generated Content Juan Wu 1 , Zhe (James) Zhang 2 , Amit Mehra 2 1 University of Science and Technology of China 2 University of Texas at Dallas Abstract Generative artificial intelligence (Gen-AI) is reshaping content creation on digital platforms by reducing production costs and enabling scalable output of varying quality. In response, platforms have begun adopting disclosure policies that require creators to label AI-generated content, often supported by imperfect detection and penalties for non-compliance. This paper develops a formal model to study the economic implications of such disclosure regimes. We compare a non-disclosure benchmark, in which the platform alone detects AI usage, with a mandatory self-disclosure regime in which creators strategically choose whether to disclose or conceal AI use under imperfect enforcement. The model incorporates heterogeneous creators, viewer discounting of AI-labeled content, trust penalties following detected non-disclosure, and endogenous enforcement. The analysis shows that disclosure is optimal only when both the value of AI-generated content and its cost-saving advantage are intermediate. As AI capability improves, the platform’s optimal enforcement strategy evolves from strict deterrence to partial screening and eventual deregulation. While disclosure reliably increases transparency, it reduces aggregate creator surplus and can suppress high-quality AI content when AI is technologically advanced. Overall, the results characterize disclosure as a strategic governance instrument whose effectiveness depends on technological maturity and trust frictions. ifundefinedshowcaptionsetup 1 Introduction Generative artificial intelligence (Gen-AI) is rapidly transforming creative production on digital content platforms. Recent advances in large-scale generative models, such as GPT-5.2 and Gemini Ultra, have substantially expanded the scope, speed, and scalability of content creation. These technologies enable creators to produce text, images, audio, and video at unprecedented scale while significantly reducing production costs. For platforms, the diffusion of Gen-AI promises a larger and more diverse content supply, lower entry barriers for creators, and expanded monetization opportunities through increased engagement and advertising revenue. At the same time, the proliferation of Gen-AI poses a fundamental governance challenge for content platforms. While AI tools can enhance productivity and content quality, they also raise concerns about authenticity, credibility, and user trust. Prior research shows that audiences often perceive human-generated content as more authentic and emotionally resonant, leading to higher engagement and attachment ( Wan et al. 2017 , Kim and Dennis 2019 ) . As AI-generated content becomes increasingly human-like, however, distinguishing between human and machine authorship has become both technically difficult and strategically consequential. Viewers may value transparency regarding content creation, yet such transparency can also trigger skepticism and reduce engagement, particularly in credibility-sensitive domains. Historically, platforms have relied on algorithmic detection to identify AI-generated or manipulated content, using linguistic patterns, metadata, and behavioral signals. However, detection accuracy remains imperfect and costly to maintain. Empirical evidence suggests that state-of-the-art detectors often achieve accuracy below 80% and perform unevenly across languages, formats, and adversarial paraphrasing ( Weber-Wulff et al. 2023 , Sadasivan et al. 2023 , Kalpesh et al. 2023 ) . Detection models also require continuous retraining and labeled data, making them an incomplete solution at scale or leading to huge costs. Moreover, detection alone does not address the strategic incentives of creators, who may choose whether to disclose or conceal AI usage based on expected audience responses and platform enforcement. In response, platforms have increasingly turned to disclosure policies that require creators to self-report AI usage. Since early 2024, Meta has mandated labels such as ‘‘Imagined with AI’’ on Facebook and Instagram, while YouTube has implemented an ‘‘Altered or Synthetic Content’’ disclosure requirement for AI-generated or manipulated media. 1 1 1 https://about.fb.com/news/2024/02/labeling-ai-generated-images-on-facebook-instagram-and-threads/ ; https://blog.youtube/news-and-events/disclosing-ai-generated-content/ These policies typically combine voluntary self-disclosure with algorithmic detection as a backstop and impose penalties—such as reduced visibility, downranking, or removal—when undisclosed AI usage is detected. 2 2 2 https://about.fb.com/news/2024/04/metas-approach-to-labeling-ai-generated-content-and-manipulated-media/ and https://transparency.meta.com/policies/community-standards/misinformation . Yet disclosure practices vary substantially across platforms, and several major platforms (e.g., X, Reddit, Bluesky) have not adopted mandatory disclosure regimes. Despite their growing prevalence, the economic consequences of disclosure policies for AI-generated content remain poorly understood. Disclosure is typically framed as an ethical or transparency-enhancing intervention, yet it also operates as a strategic governance instrument that reshapes creator incentives, content composition, and platform value. By altering how AI adoption maps into engagement and revenue, disclosure affects not only whether creators use AI, but also how and when they choose to reveal that usage. Identical AI-generated content can generate markedly different engagement outcomes depending on whether AI use is voluntarily disclosed, revealed through platform detection, or strategically concealed. As a result, disclosure does not merely shift adoption margins; it fundamentally changes creators’ strategic calculus regarding transparency, concealment, and content positioning. In this sense, disclosure functions as an endogenous incentive mechanism that reallocates AI usage across creator types and content domains, rather than as a simple binary constraint on technology adoption. This paper addresses a set of interrelated research questions concerning the governance of AI-generated content on digital platforms. 1. Under what conditions should a platform mandate disclosure of AI usage rather than rely on algorithmic detection alone? 2. How do disclosure policies reshape creators’ strategic incentives, not only regarding whether to adopt Gen-AI, but also whether to disclose, conceal, or selectively reveal its use? 3. How should platforms optimally design enforcement intensity, such as penalties for non-disclosure, as the quality and cost efficiency of Gen-AI technologies evolve over time? 4. How do disclosure regimes affect platform profitability, creator surplus, and aggregate content quality, and when do transparency gains come at the cost of reduced ecosystem value? By answering these questions within a unified formal framework, we clarify when disclosure policies function as effective governance tools and when they instead become frictions that distort AI adoption and value creation. Specifically, we develop a formal model comparing two regimes: a non-disclosure benchmark, in which the platform alone detects and labels AI-generated content, and a self-disclosure regime, in which creators are required to label AI usage subject to imperfect detection and penalties for non-compliance. The model captures three key features of Gen-AI ecosystems. First, AI reduces production costs and may improve content quality, but the magnitude of these effects varies across creators. Second, viewers discount content labeled as AI-generated and further penalize creators whose non-disclosure is revealed by the platform, reflecting both algorithm aversion and trust concerns ( Castelo et al. 2019 , Logg et al. 2019 , Longoni et al. 2019 ) . Third, detection is imperfect, creating strategic incentives for concealment. The analysis delivers several insights. Disclosure is optimal only within an intermediate region in which AI-generated content is valuable but not dominant and cost savings are substantial but not overwhelming. In this region, disclosure improves transparency and reallocates AI usage across creator types in a way that preserves trust without severely distorting productive efficiency. Outside this region, disclosure becomes counterproductive. When AI quality is low, disclosure primarily suppresses adoption without improving content value. When AI quality is high, disclosure-induced credibility discounts and enforcement costs distort efficient AI adoption and reduce aggregate content quality. The results further show that disclosure is not neutral across market participants. Although self-disclosure reliably increases transparency and can raise platform profit by mitigating trust externalities, it does so by shifting compliance and credibility costs onto creators. Aggregate