Great Disappearance Acts Generative Search and Shadow Banning
Pith reviewed 2026-07-01 00:27 UTC · model grok-4.3
The pith
Generative search and shadow banning shift the internet toward centralized platform control that prioritizes profit over open expression and creator sustainability.
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
Generative search powered by RAG diverts users from source sites while shadow banning de-ranks or suppresses content through algorithmic moderation; both practices lack transparency, and frameworks like China's Regulation on Algorithmic Recommendations and the EU Artificial Intelligence Act provide incomplete remedies, resulting in centralized control that favors platform profit over innovation and open expression.
What carries the argument
Generative search via retrieval-augmented generation that produces direct answers and shadow banning via algorithmic de-ranking and traffic reduction, analyzed through legal claims of copyright issues and unfair competition alongside regulatory comparisons.
If this is right
- Independent websites and creators lose sustainable revenue streams as traffic is diverted.
- Free expression faces chilling effects from unaccountable visibility reductions.
- Existing regulations in major jurisdictions address only limited aspects of opacity and bias.
- Without new equitable frameworks, the open web loses diversity of sources and voices.
Where Pith is reading between the lines
- Similar visibility suppression could extend to other AI-mediated content distribution beyond search and social feeds.
- Licensing models for training data might need to include ongoing traffic compensation for source sites.
- Cross-border regulatory harmonization could become necessary if regional rules diverge in enforcement strength.
Load-bearing premise
That generative search and shadow banning function mainly as opaque tools driven by platform incentives without adequate transparency or redress options.
What would settle it
Empirical data showing that RAG search results consistently drive net traffic back to original websites or that shadow-banning decisions are publicly auditable with consistent appeal outcomes across user groups.
read the original abstract
The internet, once celebrated as a decentralized public sphere, is increasingly undermined by practices such as generative search and shadow banning, which divert traffic and suppress visibility. Generative search, powered by Retrieval Augmented Generation RAG, synthesizes content into direct answers, bypassing websites and depriving them of traffic and revenue. This threatens the sustainability of independent content creators, small enterprises, and the open web ecosystem. Shadow banning, a practice that intentionally reduces the visibility of social media posts through algorithmic moderation, exacerbates these issues by chilling free expression and limiting transparency and accountability. This article explores these opaque practices through a legal and regulatory lens. The first part examines the rise of generative search, analyzing its technological and legal implications, including copyright infringement, unfair competition, and unjust enrichment. It also evaluates potential solutions such as licensing agreements and agentic AI. The second part focuses on shadow banning: algorithmic dissuasion, de-ranking, and the reduction of traffic, with specific attention to Chinas Regulation on Algorithmic Recommendations RAR and the EUs Artificial Intelligence Act AIA. Both frameworks offer partial solutions but fall short of ensuring fairness, transparency, and redress mechanisms. Ultimately, the shift toward centralized control by dominant platforms prioritizes profit and risk management over innovation, fairness, and diversity in online expression. To counteract these trends, regulatory interventions, algorithmic transparency, and equitable frameworks are essential. Without such measures, the internet risks losing its character as a democratized public sphere for free expression and innovation.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper claims that generative search powered by RAG and shadow banning divert traffic from websites and suppress visibility on social media, threatening independent creators and the open web. It explores these through legal analysis, covering copyright infringement, unfair competition, unjust enrichment for generative search, and potential solutions like licensing and agentic AI. For shadow banning, it examines algorithmic de-ranking and evaluates China's RAR and the EU's AIA, finding both frameworks insufficient for transparency and redress. The paper concludes that dominant platforms' shift to centralized control prioritizes profit and risk management over innovation, fairness, and diversity, advocating for regulatory interventions, algorithmic transparency, and equitable frameworks.
Significance. If substantiated, the analysis would contribute to the understanding of how AI technologies and moderation practices affect online ecosystems and free expression. It identifies regulatory gaps in existing frameworks and highlights the need for better accountability mechanisms. However, the qualitative nature without empirical support limits its immediate applicability to policy-making or technical solutions in the field.
major comments (3)
- [Conclusion] The central claim in the conclusion that 'the shift toward centralized control by dominant platforms prioritizes profit and risk management over innovation, fairness, and diversity in online expression' is load-bearing for the call for regulatory intervention but rests on interpretive attribution rather than evidence; the legal analysis provides general descriptions of practices without traffic data, creator revenue studies, or analyses distinguishing profit/risk motives from legal compliance or quality control.
- [First part] The first part's examination of generative search implications (copyright infringement, unfair competition, unjust enrichment) and solutions (licensing agreements, agentic AI) remains descriptive without specific case examples, case law citations, or assessment of how RAG leads to the claimed harms.
- [Second part] The second part's evaluation of China's RAR and the EU's AIA notes that both 'fall short of ensuring fairness, transparency, and redress mechanisms' but offers no detailed comparison, implementation examples, or specific shortcomings to support this assessment.
minor comments (2)
- The abstract contains grammatical issues such as 'Chinas' and 'EUs' which should read 'China's' and 'EU's'.
- The manuscript would benefit from clearer section headings, numbering, and more specific references to legal cases or regulatory texts to support the analysis.
Simulated Author's Rebuttal
Thank you for the constructive feedback. We appreciate the opportunity to clarify and strengthen our analysis. Below, we respond to each major comment and indicate the revisions we will make to the manuscript.
read point-by-point responses
-
Referee: [Conclusion] The central claim in the conclusion that 'the shift toward centralized control by dominant platforms prioritizes profit and risk management over innovation, fairness, and diversity in online expression' is load-bearing for the call for regulatory intervention but rests on interpretive attribution rather than evidence; the legal analysis provides general descriptions of practices without traffic data, creator revenue studies, or analyses distinguishing profit/risk motives from legal compliance or quality control.
Authors: We recognize that this claim is interpretive, derived from the patterns observed in the legal and regulatory shortcomings analyzed in the paper. As a legal analysis rather than an empirical study, we do not present primary data on traffic or revenues. We will revise the conclusion to frame this as a synthesis of the identified regulatory gaps and platform incentives, adding caveats to distinguish it from empirical findings. This will better support the call for intervention while acknowledging the interpretive nature. revision: partial
-
Referee: [First part] The first part's examination of generative search implications (copyright infringement, unfair competition, unjust enrichment) and solutions (licensing agreements, agentic AI) remains descriptive without specific case examples, case law citations, or assessment of how RAG leads to the claimed harms.
Authors: The analysis in the first part is grounded in legal doctrines applicable to generative search. To enhance specificity, we will incorporate relevant case law citations, such as those involving AI-generated content and copyright, and provide a clearer step-by-step assessment of how RAG systems bypass traditional web traffic by delivering synthesized responses. This will address the descriptive nature while maintaining the paper's focus on regulatory implications. revision: yes
-
Referee: [Second part] The second part's evaluation of China's RAR and the EU's AIA notes that both 'fall short of ensuring fairness, transparency, and redress mechanisms' but offers no detailed comparison, implementation examples, or specific shortcomings to support this assessment.
Authors: We agree that a more detailed evaluation would strengthen this section. In the revised manuscript, we will include a comparative analysis of specific provisions in the RAR and AIA, with examples of their implementation and explicit identification of shortcomings regarding transparency and redress for shadow banning and algorithmic moderation. revision: yes
Circularity Check
No circularity: qualitative legal analysis with no derivations or fitted claims
full rationale
The paper is a qualitative discussion of legal and regulatory issues around generative search and shadow banning. It contains no equations, no parameter fitting, no predictions derived from inputs, and no self-citation chains that serve as load-bearing premises for a mathematical or empirical result. The central claim is an interpretive conclusion about platform incentives, but it is not constructed by redefining or renaming its own inputs. All enumerated circularity patterns require a derivation step that reduces to its own premises by construction; none are present here.
Axiom & Free-Parameter Ledger
Reference graph
Works this paper leans on
-
[1]
These factors are not necessarily the exclusive determinants of the fair use inquiry and do not mechanistically resolve fair use issues
Copyright Infringement To see whether there is copyright infringement, there should be substantial similarity and access. 74 Substantial similarity can be assessed at different abstraction levels (chapters, sections, paragraphs, sentences, and words, letters). Changing each word with a synonym, does not necessarily immunize one from copyright infringement...
1985
-
[2]
semi-plagiarism
Plagiarism and Semi-Plagiarism Some generative search results do not provide sources. This lack of attribution leads to plagiarism,89 which Posner describes as academic fraud. 90 Fraud is a tort, and often a crime, but plagiarism is neither.91 In case of most plagiarism two parties are harmed: the original author of the content who was deprived of his rig...
1976
-
[3]
hot news,
Unfair Competition International News Service v. Associated Press (1918) (INS v AP )95 can shed light on the question whether generative search is fair competition. During WWI, the allied forces did not want to provide INS with news because its owner, William Randolph Hearst, favored Germany. Therefore, INS hired people to read East Coast newspapers that ...
1918
-
[4]
63.41 percent of all U.S. web traffic referrals from the top 170 sites initiated on Google.com
Unjust Enrichment and Tort of Negligence Google Search has been the dominant intermediary between internet users and websites, almost worldwide.100 For a decade, over 90 percent of internet users made use of Google Search.101 Website holders, too, were dependent on Google Search for the vast majority of their traffic referrals. 102 93OFFICE OF ALUMNI AND ...
2024
-
[5]
a thriving ecosystem of publishers and creators
Licensing Deals Generative search deprives website holders of the revenue that can be generated from internet users visiting their websites: the fee for subscriptions to have access to their website, or the pay per clicks on advertisements on their website. However, generative search providers can compensate website holders for this deprivation and indemn...
2023
-
[6]
Agentic Retrieval-Augmented Generation: A Survey on Agentic RAG
Agentic AI Fenwick, Jurcys, and Loikkanen posited AI agents as a possible remedy against the traffic and revenue loss of websites: 116 Google’s AI agent could gather information from websites to generate search results and remunerate AI agents of these websites for this information. While RAG is reactive, responding to input prompts with contextually rele...
work page internal anchor Pith review Pith/arXiv arXiv 2024
-
[7]
Other circumstances provided for by laws and administrative regulations
Reining in Recommendation Algorithms China’s copyright system—which is a hybrid of exceptions and limitations, the three-step test, and fair use principles —provides an example of a system within which unauthorized works uploaded on a platform are not always copyright infringements. China ’s third amendment to the Copyright Law (2020) provides 13 limitati...
2020
-
[8]
Unauthorized but Legal User-Generated Content This article is investigating the legitimacy and chilling effect of this practice in light of RAR 132 and AIA,133 and takes user-generated content that was created unauthorizedly but possibly legal if it did not infringe copyright as a case study. The results of this case study are to a large extent also appli...
1994
-
[9]
18 Hao) ( 最高人民法院印发《关于充分发挥知识产权审判职能作用推动社会主义文化大发展大繁荣和促进经 济自主协调发展若干问题的意见》的通知（法发〔2011〕18号)) [Notice of the Supreme People ’s Court on Issuing the Opinions on Issues concerning Maximizing the Role of Intellectual Property Right Trials in Boosting the Great Development and Great Prosperity of Socialist Culture and Promoting the Independent and Coordinated Develo...
2011
-
[10]
of the Supreme People’s Court)] (promulgated by the Judicial Comm. Sup. People’s Ct., Dec. 16, 2011, effective Dec. 16,
2011
-
[11]
beneficence, non-maleficence, autonomy, justice, and explicability
Sup. People ’s Ct. Gaz., Dec. 16, 2011, https://gongbao.court.gov.cn/Details/90857967e518c766c368851b1b705a.html. 130Ganea et al., supra note 127. 13117 U.S.C. § 107. 132RAR, supra note 122. 133Commission Regulation 2024/1689 of June 13, 2024, Laying Down Harmonised Rules on Artificial Intelligence and Amending Regulations (EC) No 300/2008, (EU) No 167/20...
-
[12]
ERRORS, MISTAKES, OR INACCURACIES ON THE SERVICE
-
[13]
ANY INTERRUPTION OR CESSATION OF THE SERVICE
-
[14]
THE REMOVAL OR UNAVAILABILITY OF ANY CONTENT. 158 Since June 2022, 500 hours per minute were uploaded to YouTube. 159 Because the manual assessment of copyright infringement and exceptions or limitations of copyright is not scalable, platforms increasingly rely on automatic content recognition filters, such as YouTube ’s Content ID. 155Danny Friedmann, Si...
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.