From Nudge to Hypernudge: AI and the New Consumer Manipulation Problem

By: Christophorus Ariobumi

Edited by: Stephen Shiwei Wang

Introduction

From planning travel itineraries to comparing products, artificial intelligence (AI) increasingly shapes how we make decisions. AI presents itself as a beacon of hope in the abundance of choices online. It helps us to plan, compare, or discuss products and services that ultimately influence our decision. But what if the choices we make online are not simply assisted by AI, but subtly engineered and planned? 

To understand this shift, it is useful to begin with the concept of a “nudge” in behavioral science. A nudge refers to subtle interventions in the choice architecture that steer decision-making without restricting options or altering economic incentives, therefore preserving their freedom of choice.1 Classic examples include automatic enrollment in retirement savings plans or structuring options to encourage healthier or more beneficial outcomes. Traditional nudge utilizes human’s bounded rationality, which means that humans are susceptible to confirmation bias that often makes decisions that appear rational but are constrained by the information available. 

However, the invention of AI boosted the capabilities to shape user outcomes. Hypernudge, coined by Karen Yeung, is an AI-driven dynamic intervention that leverages big data to influence human decision-making through a continuous feedback loop.2 Unlike traditional nudge, hypernudge does not provide a standardized and often static intervention. It operates dynamically through ongoing monitoring, predicting, and adjusting interventions based on the user’s behavior.3 Such practice is often invisible to the human eye, resulting in a blur between assistance and manipulation. An example of hypernudge is personalization of the recommended deals in Amazon based on each person’s search history and interaction in the application. 

Hypernudge may seem harmless, even helpful to humans in navigating complex choices, yet it raises a deeper concern. When influence becomes continuous, personalized, and even normalized on a daily basis, does it represent a fundamental problem to governance and ethics?

How Does Hypernudge Change Human Behavior?

At its core, traditional nudging influences consumer behavior through carefully presenting options so that people are more likely to choose a specific decision without restricting other options. As humans do not make perfectly rational decisions, nudge leverages the predictable pattern of human behavior such as default bias (sticking to the pre-set options), framing effects (how choices are presented), and limited options (focusing on the most visible) to guide their decision-making. Therefore, humans often “satisfice” or settle for decisions that are good enough given their cognitive limitations and environmental constraints.4 Ultimately, nudging works by aligning decisions with how people actually think, rather than how they ideally should.

The rise of AI-driven hypernudging fundamentally challenged the whole architecture of nudge. According to Mills, there are four core characteristics of hypernudge that distinguish it from the traditional nudge.5 First, the hypernudge must be personalized to the user. This means that it utilizes high-intensity methods such as AI and big data to reduce the problem of heterogeneity. Second, the nudge must constantly change in real time to reflect as much feedback during the session. Third, it has to possess the predictive capacity to evaluate predictions. Last, the technology itself can easily become hidden to human eyes. 

Through continuously evolving data used to personalize interventions, hypernudging fundamentally reshapes human behavior. Rather than simply presenting options, AI systems construct the decision pathway through which individuals arrive at choices in real time.6 What was once a static intervention becomes an adaptive system that continuously shapes both decisions and underlying preferences at scale. In this environment, influence is not a single moment of choice architecture, but an ongoing process embedded within the user’s interaction with digital platforms.

Amazon’s recommendation system is one prime example of hypernudge that is visibly seen in everyday life. Amazon utilizes AI and big data based on the user’s browsing behavior, purchase history, and other behavioral signals (clicks, time spent, etc.) to continuously personalize the recommendations. These recommendations are not neutral; they are strategically embedded in features such as flash sales, prominently ranked search results, and curated product listings, all designed to guide users toward purchase decisions. 

This influence extends beyond simple product recommendations. Amazon’s recently introduced AI-powered “Interest” feature actively tracks user preferences over time, surfacing new products that align with inferred interests.7 In doing so, the platform begins to function less like a passive marketplace and more like an automated personal shopper that delivers dynamic and continuously updated suggestions. As a result, consumer behavior is not only guided at the point of choice but is persistently shaped through an evolving, personalized decision environment.

Another example can be drawn from the use of Generative AI (GenAI) when making decisions. GenAI, specifically the large language model (LLM), represents a significant development that might broaden options through presenting more options based on the customer’s preferences without the need to repeatedly feed information. However, in order to “intelligently” understand preferences, it involves surveillance that would violate privacy considerations with surveillance but making itself appear to be more transparent and user-directed. For instance, when using a large language model (LLM) to generate a travel itinerary, it does not automatically generate a one-size-fits-all itinerary to the users. It curates, filters, and generates the recommendations based on the conversation history to tailor the response based on what it evaluates as the user’s preferences. Therefore, the trade-off reflects a long-standing debate since the rise of modern technology: preserve privacy with less intelligent systems, or allow increasingly advanced technologies to quietly shape large parts of consumers’ lives?

Hypernudge and the Governance Implications

The examples from Amazon and GenAI systems show that decision pathways are not simply assisted, but subtly reconfigured. Through hypernudge, AI does not just assist human thinking—it constructs the conditions under which thinking occurs. As a result, there are several governance issues that emerge from hypernudging. 

First, hypernudging intensifies the issue of transparency through the proliferation of dark patterns, an interface that deceives users into making decisions that are not in their best interests.8 Through hypernudge, Amazon could exploit the ranking manipulation through personalization. Products labeled with “Best Seller” or “Recommended for You” are not neutral; they are AI-constructed pathways that manipulate users’ behavior. They tamper with the user’s perception of options without necessarily leading them to the best price or quality.9 While behavioral intervention is observable in traditional nudge, it remains hidden in hypernudge. Users couldn’t see why certain products appear first, or why certain products are labeled as “Recommended for You”. As a result, users cannot distinguish between what is relevant and what is strategically engineered as an option. AI influences in this context become embedded and non-transparent at the point of decision-making.

This issue is further exacerbated by GenAI’s systems. For example, when users ask for the best sneakers for hiking, responses are personalized based on their prior interactions and inferred preferences. While this may appear helpful on a surface level, it also narrows the information and options based on what the users are exposed to without any transparency. In a more extreme scenario, if advertising is embedded into GenAI’s outputs, the line between recommendation and promotion becomes increasingly blurred. 

Second, hypernudging creates dependency on their behavior and preferences. Through continuous exposure to similar products and recommendations, users’ preferences are gradually shaped to a narrow choice set. Hypernudging does not technically restrict choices, but it can shape people’s thoughts and decisions by exploiting cognitive biases without awareness, threatening a “right to mental self-determination” or the right to create one’s own thoughts rather than have them subtly constructed.10 This directly supports the idea that system-shaped outputs can be experienced as one’s own intentions and preferences, even though they are partially engineered. Over time, this can erode critical thinking as users start to rely heavily on AI and algorithms to shape their preferences. 

Third, hypernudging distorts market fairness and consumer welfare. Because AI systems determine how options are ranked and presented, it acts as a gatekeeper of visibility. Personalized recommendations and pricing strategies are explicitly used by retailers to increase profits by influencing consumer demand and willingness to purchase.11 On Amazon, products that appear at the top page or on the recommendation page are not necessarily the best in terms of price or quality, but those that are more likely to convert based on platform’s objectives. As a result, hypernudging can skew market competition through favoring certain brands or sponsored products while crowding out alternatives that may be better suited for the consumer. 

The Path Forward: Future of AI and Hypernudge

Despite the challenges outlined in the section above, governing hypernudge remains a significant task to implement due to its hidden nature. Current regulatory efforts already signal growing awareness of manipulative digital practices, such as the European Union (EU)’s Digital Service Act (DSA). Article 25(1) DSA prohibits online platforms from designing, organizing or operating their online interfaces in a way that deceives or manipulates users, or otherwise materially distorts or impairs their ability to make free and informed decisions.12 However, DSAs current approach to regulate dark patterns does not address the additional challenges of non-transparency resulting from the use of AI. In addition, the EU AI Act introduces new prohibitions on dark patterns without mentioning the term specifically. Articles 5(1)(a) and (b) of the EU AI Act prohibits subliminal techniques, purposefully manipulative or deceptive techniques or use of AI systems that exploit vulnerabilities based on age, disability or a specific social or economic situation that could cause significant harm.13

While the EU already took a step forward, regulations in the United States highlight both progress and limitations. In June 2023, the Federal Trade Commission (FTC) took action against Amazon for allegedly using deceptive interface designs to steer users into subscriptions of the Amazon Prime and make cancellation difficult.14 FTC’s action against Amazon recognized the harms of manipulative design, but it operates mainly on case-by-case interventions that address specific instances instead of the broader system of algorithmic behavioral influence. As hypernudging becomes embedded in AI-driven systems, enforcement that relies on identifying discrete violations may struggle to keep pace with continuously evolving practices. 

Discussion around hypernudging remains limited as it is not yet a mainstream regulatory category. The Organisation for Economic Co-operation and Development (OECD) have developed guidelines on dark patterns and online consumer protection, these frameworks focus primarily on interface-level manipulation mediated by technologies.15 They do not fully capture how AI systems dynamically construct decision environments, personalize influence, and shape preferences over time. As a result, existing governance tools tend to regulate the symptoms rather than the systemic mechanism of hypernudging itself. 

To address this gap, there is a need to push hypernudging into the center of policy discourse. This requires reframing the issue from a narrow concern about deceptive design toward a broader question of behavioral governance in AI-mediated environments. Policymakers should begin to discuss clear boundaries on acceptable choice architecture, including limits on manipulative personalization and behavioral targeting. 

Lastly, policymakers should advance the discussion around the advancement of explainable AI (XAI). XAI aims to make algorithmic decision-making more interpretable by providing insights into how outputs are generated and which factors influence them.16 In the context of hypernudging, XAI could help reveal why certain products or options are ranked higher and how user data shapes their recommendations. Normalizing XAI could mitigate the issue of transparency that enables users and policymakers to understand how AI influences its users in real time. However, XAI explanations could be overly technical and framed by the firms, which raised another concern about ethics and governance. Without expanding regulatory focus beyond interface design toward the underlying logic of algorithmic influence, current frameworks will remain insufficient to address the systemic risks posed by hypernudging.

 

 

Works Cited

  1. Thaler, Richard H, and Cass R Sunstein. 2008. Nudge: Improving Decisions about Health, Wealth, and Happiness. New York: Penguin Books.
  2. Yeung, Karen. 2016. “‘Hypernudge’: Big Data as a Mode of Regulation by Design.” Information, Communication & Society 20 (1): 118–36. https://doi.org/10.1080/1369118x.2016.1186713.
  3. Degli Esposti, Sara. 2014. “When Big Data Meets Dataveillance: The Hidden Side of Analytics.” Surveillance & Society 12 (2): 209–25. https://doi.org/10.24908/ss.v12i2.5113. 
  4. Tagliabue, Marco. 2022. “Tutorial. A Behavioral Analysis of Rationality, Nudging, and Boosting: Implications for Policymaking.” Perspectives on Behavior Science, January. https://doi.org/10.1007/s40614-021-00324-9. 
  5. Mills, Stuart. 2022. “Finding the ‘Nudge’ in Hypernudge.” Technology in Society 71 (November): 102117. https://doi.org/10.1016/j.techsoc.2022.102117. 
  6. Richarde, Ana Paula Merenda, Diego Costa Pinto, Marlon Dalmoro, and Paulo Henrique Muller Prado. 2025. “The Power of GenAI Nudges: How Generative AI Shapes Consumer Empowerment and Goal Desirability.” International Journal of Information Management 85 (July): 102955. https://doi.org/10.1016/j.ijinfomgt.2025.102955. 
  7. Constantino, Tor. 2025. “Amazon’s AI Just Got Smarter, More Predictive and More Personal.” Forbes. March 31, 2025. https://www.forbes.com/sites/torconstantino/2025/03/31/amazons-ai-just-got-smarter-more-predictive-and-more-personal/. 
  8. Faraoni, Stefano. 2023. “Persuasive Technology and Computational Manipulation: Hypernudging out of Mental Self-Determination.” Frontiers in Artificial Intelligence 6 (July). https://doi.org/10.3389/frai.2023.1216340.  
  9. Morozovaite, Viktorija. 2022. “Hypernudging in the Changing European Regulatory Landscape for Digital Markets.” Policy & Internet, October. https://doi.org/10.1002/poi3.329. 
  10. Faraoni, Stefano. 2023. “Persuasive Technology and Computational Manipulation: Hypernudging out of Mental Self-Determination.” Frontiers in Artificial Intelligence 6 (July). https://doi.org/10.3389/frai.2023.1216340.  
  11. Zhou, Chi, Danyang Bai, Tieshan Li, and Jing Yu. 2024. “Personalized Recommendation, Behavior-Based Pricing, or Both? Examining Privacy Concerns from a Cost Perspective.” Omega 133 (December): 103223. https://doi.org/10.1016/j.omega.2024.103223. 
  12. Troge, Thorsten. 2023. “Does AI Enhance the Risk of Dark Patterns and How Does EU Law Regulate Them?” Taylorwessing.com. Taylor Wessing. May 9, 2023. https://www.taylorwessing.com/en/interface/2023/ai—are-we-getting-the-balance-between-regulation-and-innovation-right/does-ai-enhance-the-risk-of-dark-patterns-and-how-does-eu-law-regulate-them. 
  13. European Parliament. 2025. “AT a GLANCE Digital Issues in Focus.” https://www.europarl.europa.eu/RegData/etudes/ATAG/2025/767191/EPRS_ATA(2025)767191_EN.pdf. 
  14. FTC. 2023. “FTC Takes Action against Amazon for Enrolling Consumers in Amazon Prime without Consent and Sabotaging Their Attempts to Cancel.” Federal Trade Commission. June 21, 2023. https://www.ftc.gov/news-events/news/press-releases/2023/06/ftc-takes-action-against-amazon-enrolling-consumers-amazon-prime-without-consent-sabotaging-their. 
  15.  OECD. 2024. “Six ‘Dark Patterns’ Used to Manipulate You When Shopping Online.” OECD. 2024. https://www.oecd.org/en/blogs/2024/09/six-dark-patterns-used-to-manipulate-you-when-shopping-online.html. 
  16. Ali, Sajid, Tamer Abuhmed, Shaker El-Sappagh, Khan Muhammad, Jose M. Alonso-Moral, Roberto Confalonieri, Riccardo Guidotti, Javier Del Ser, Natalia Díaz-Rodríguez, and Francisco Herrera. 2023. “Explainable Artificial Intelligence (XAI): What We Know and What Is Left to Attain Trustworthy Artificial Intelligence.” Information Fusion 99 (101805): 101805. https://doi.org/10.1016/j.inffus.2023.101805.

Author Bio

Christophorus Ariobumi is a Master of Public Administration candidate at Cornell University’s Brooks School of Public Policy, specializing in science and technology policy. He is also an Environmental Finance and Impact Investing (EFII) Fellow, driven by his interest in the intersection of digital technology and sustainability. His work examines urban policy, gig and labor economy, and digital sustainability. Prior to Cornell, he worked as a public policy and government affairs consultant in Indonesia, advising public and private sector clients on navigating complex regulatory environments, developing stakeholder engagement strategies, and analyzing policy impacts across sectors.

Related Posts

Scroll to Top