Introduction 

​​​

How much of what we think, feel, or purchase online can be truly said to be our own choices? In the age of big tech, unsuspecting users are increasingly exposed to forms of psychological manipulation that shape and reinforce certain beliefs,  while also impairing decision-making processes. Persuasive technology is the key component that enables platforms to collect large amounts of data and generate cognitive profiles of consumers. The Elaboration Likelihood Model (ELM) provides a foundational framework for understanding manipulation strategies intended to target the peripheral route of thinking, relying on superficial cues and heuristics instead of logical thinking. This is enhanced through dark patterns which exploit cognitive biases in order to influence choices: ranging from purchasing certain products to political opinions. The current consent mechanisms in place are insufficient to protect users, focusing on simply informing users of manipulation but failing to address broader structural conditions that shape user behaviour. As manipulative practices become normalized and consent mechanisms fail, stronger regulations must be implemented to protect users’ autonomy and prevent technology from undermining individual authenticity, keeping its influence separate from social and political spheres.​​

​​

Persuasive Technology and Psychological Manipulation ​​
 

Artificial intelligence (AI) is integrated in most technologies today. It is a tool that can acquire user data and uncover connections invisible to humans, creating a cognitive profile that provides a foundation for persuasive technology. The definition of persuasive technology has been frequently adapted to accurately account for drawbacks,  which can appear as manipulation through the use of decision-making biases intensified by dark patterns and hypernudges, shaping human attitudes or behaviours (Faraoni, 2023). In other words, persuasive technology exploits human cognitive shortcuts to influence behaviour, without users' conscious awareness. Manipulation techniques of humans cannot be compared to algorithmic-driven manipulation as technology can collect large amounts of data in ways humans cannot. This allows it to influence decisions at a larger scale utilizing personal information, habits, preferences and cognitive processes. Faraoni (2023) even states that algorithms can know more about an individual than the individual themselves, demonstrating how powerful technological persuasion can be. For example, an AI-driven system can identify that a user is pregnant and track their emotions based on the frequency of social media posts of their baby, whereas a human can only assume that showing a picture of a baby will have an impact on decision-making processes (Faraoni, 2023). AI can then use this information to emotionally manipulate individuals into purchasing baby products, something that humans cannot achieve at the same scale. This example further demonstrates the gap between human manipulation and technological-driven manipulation and how they remain incomparable.

 

To understand how persuasive technology works, the ELM helps explain the two systems of processing persuasion and how each operates. System 1, is the primordial system, or peripheral route, which operates in an automatic and unconscious way through cognitive biases, while System 2 is the rational system, or central route, that relies on cognitive resources (Faraoni 2023; Hou 2025). Technology, specifically AI, shapes the psychological mechanism of human persuasion by reducing users’ capacity and willingness to engage in deeper cognitive elaboration, resulting in reliance on the peripheral route (Hou, 2025). Manipulation functions through  two key elements: it remains hidden from the individual and exploits cognitive vulnerabilities within decision-making processes (Faraoni, 2023). This allows individuals to rely on the peripheral routes’ superficial cues and cognitive shortcuts instead of careful and effortful evaluation. A study conducted by Hou (2025), analyzed 11,942 human-generated comments on various Chinese social media platforms to assess how users responded to AI-generated rumours as either logic-driven (the central route) or emotion-driven (the peripheral route). The goal of this study was to better understand public perception and patterns of reasoning. It was found that the hyper-realism of AI-generated content disrupts critical evaluation processes. Heuristic cues favoring emotional appeals over logical reasoning were predominant instead, highlighting  just how effectively AI-generated content persuasively targets system responses. 

 

Through the lens of the ELM, the harms of psychological manipulation begin with undermining authentic choice and consumer harm, even extending beyond individual decision-making by influencing political choices. Over time, the decisions individuals make begin to reflect externally engineered influences instead of their personal goals or values. This is evident through manipulative advertising strategies that motivate  consumers to purchase items they do not need or desire. It also undermines one’s  ability to carefully weigh their options, form intentions about them, and act in accordance with those intentions. This erosion of individual authenticity and impaired decision-making decreases one’s  ability to safeguard their interests long-term. This mechanism leads to collective harm where online manipulation threatens democracy, self-governance, and public trust (Susser et al., 2019). Thus, technological manipulation is not solely an ethical problem but a broader societal risk that is further systematically exploited by design strategies.

 

Persuasive Technology and Psychological Manipulation ​​​

​

The most prominent mechanism of societal risk materializes through the use of dark patterns. These patterns can be defined as deceptive or distorted practices of digital interfaces used to manipulate consumers towards choices that benefit corporations rather than users. In other words, dark patterns  are digital tricks that motivate consumers  to benefit companies instead of themselves.  These practices rely on manipulative nudges, a form of choice architecture that alters behaviour without forbidding options or significantly changing economic incentives. Instead, this works by structuring digital environments to make one choice more likely to happen. These nudges are designed in this way to influence decision-making processes to favour pre-selected choices by the designer. Online persuasion is significantly more dangerous than face-to-face persuasion as individuals are less likely to notice, thus left unable to activate defensive mechanisms. Human cognition is limited in its ability to recognize the cognitive biases embedded in technology, and this imbalance is intensified when dark patterns are deployed through AI. Under these circumstances, second-generation dark patterns, known as hypernudges, emerge. Hypernudges operate dynamically through real-time data, personalizing outputs based on user engagement (Faraoni, 2023). This creates a continuous feedback loop in which outputs feed back into inputs. This constant reconfiguration of choice architecture allows for algorithms to detect correlations across datasets that exceed human cognitive capacity.

 

By reshaping choice architecture, algorithms enable interfaces to manipulate users, which Sterelny’s framework calls hostile dark patterns (Timms, 2025). Timms (2025) characterizes all dark patterns as hostile patterns meaning that digital interfaces work in ways to undermine user interests while benefiting platforms or services. Many user-interface designers argue that their platforms were designed without intentionality to manipulate users. However, Timms (2025) argues that deliberately exploiting human behaviour is sufficient to classify an interface as problematic, even though this is not  necessary to undermine user autonomy or operate as a hostile dark pattern. Some patterns go further, amounting to ‘digital pickpocketing,’ by which  user autonomy, decision-making processes, or cognition are  not merely influenced, but entirely bypassed. 

 

Dark patterns employ various methods to serve platforms or services . Obstruction introduces challenges to complicate task completion, like requiring users to navigate complex menus when simply trying to search for a recipe, promoting frustration that leads to compliance. Sneaking is another method that conceals important information, like hidden costs or subscriptions, resulting in users agreeing to terms unintentionally. In addition, social engineering leverages emotional manipulation, creating a sense of urgency or fear of missing out. For example, online stores often employ urgency-based messages such as ‘last chance’ or ‘only one item left in stock,’ which encourages rapid decision-making and increases compliance that benefits platforms or services. Interface interference manipulates visual choice architecture, exemplified by preselected options that exploit users’ cognitive biases to maintain preselected choices instead of selecting equally available alternative choices . Alongside these tactics, forced action compels users to complete certain actions before obtaining the service they want, like providing personal information before being able to view a webpage. These deceptive strategies highlight the multifaceted ways in which dark patterns insidiously mislead users to make unintentional or harmful decisions (Timms, 2025).

 

Often dismissed as user-interface issues, Leiser and Santos (2024) instead argue that dark patterns are embedded structurally in system architecture. They provide a three-tier visibility threshold for dark patterns, consisting of: visible dark patterns, darker patterns, and the darkest patterns. Visible dark patterns are overt and easily recognizable manipulation tactics that directly influence human decision-making, such as nagging or preselection, and are more likely to be detected by regulators and auditors. In contrast, darker patterns are more subtle and covert practices, operating behind the scenes, like requiring multiple selection options to increase user effort in navigating a webpage, and these are less detectable by regulators.Further, the darkest patterns are the most deceptive form of computational manipulation, relying on advanced algorithmic frameworks designed to utilize cognitive biases and personalize users’ cognitive profiles through hypernudges without conscious awareness. Currently, regulatory decisions do not reference dark patterns explicitly; however, the identification of  these patterns in enforcement actions could help highlight their illegality and discourage manipulative practices. Future oversight must extend beyond the user interface to address the entire system architecture where these dark patterns are embedded.

​

Consent and User Autonomy
​

The primary harm of online manipulation lies in its violation of individual autonomy. Autonomy is a foundational normative principle in liberal democratic societies, grounded in the belief that individuals have the ability to self-govern and are also able to collectively participate in democratic processes.That is to say, , autonomy is not only crucial to individual decision-making, but is also a prerequisite to collective decision-making which is targeted by online manipulation. Autonomy can be understood as having the psychological or cognitive competencies to make one’s own decisions and act upon them (Susser et al., 2019). When digital systems interfere with these competencies, they undermine the conditions necessary for individual agency and collective autonomous decisions.

 

These manipulative practices have become increasingly normalized, especially with the increased use of targeted advertising and AI tools becoming embedded in various institutions. Online manipulation should not be understood as an isolated occurrence, but as recurring with current consent mechanisms breaking down, failing to address overarching structural conditions that shape user behaviour (Susser et al., 2019). Some scholars, such as Jacobs (2020), argue that these dynamics are especially concerning for vulnerable people. In this view, those with a diminished capacity to safeguard their own interests must be increasingly protected from persuasive technology, ensuring that their autonomy is not impeded on. However, this vulnerability-focused approach is contested. Brenncke (2024) asserts that protection from autonomy breaching by persuasive technology is not solely limited to vulnerable people, but rather extends to all consumers, since all consumers are subject to online manipulation. This dissent demonstrates that vulnerability is only one important factor: digital manipulation is ultimately a structural issue that affects users broadly.

 

Despite the prevalence of manipulation, most regulatory frameworks of consent rely on ‘notice-and-consent’ mechanisms. This is the practice of notifying individuals that they are the target of manipulation. However, alerting individuals of this manipulation is not nearly enough to neutralize the harm. This limitation is captured by the ‘transparency paradox,’ which is the idea that too little information prevents decision-making, while too much information overwhelms users and discourages meaningful engagement (Susser et al., 2019). For consent mechanisms to work effectively and create a meaningful impact, individuals understand the source of the manipulation and what strategies they employ to meet their specific goal.. In practice, such conditions are not met, rendering ‘notice-and-consent’ as an ineffective consent tool.

 

The failure of the ‘notice-and-consent’ model is further highlighted by Richards and Hartzog (2019).. They identify unwitting consent, which undermines knowledge when users fail to understand legal agreements or the practical risks of those agreements. Conversely, coerced consent targets voluntariness, where individuals are forced to choose between consent or the loss of an essential asset. Alternatively, incapacitated consent applies where individuals, such as children, are legally unable to provide consent. Although these consent models are present, the practical conditions of awareness and voluntary consent fall short. This is influenced by a lack of vocabulary amongst scholars, advocates, and consumers to critique digital consents, allowing for this persistence of weak consent practices. These dynamics suggest that consent is almost entirely incompatible with the modern realities of data and technology. The widespread use of consent in digital contexts, along with weak legal policing, has left consumers vulnerable to data breaches, identity theft, and surveillance. Richards and Hartzog (2019) instead suggest a trust-based approach, which would include parties in information relationships to protect the data in their care, treat users fairly, and abstain from seeking consent for practices that would render individuals unreasonably vulnerable. Thus, trust is a necessary condition for protecting users where consent alone has proven insufficient.

 

The ethical failures of procedural consent are highlighted by Facebook’s emotional manipulation study conducted by Facebook and Cornell University. In the study by Flick (2015), the researcher manipulated users’ feeds to display both positive and negative posts to observe effects on subsequent posts. The researcher did not obtain informed consent for this experiment, promoting significant public backlash. This case highlights the gap between formal agreement to platform terms and informed consent from its users, demonstrating how easily users can be subjected to emotional influence without conscious awareness. This underscores the need for greater platform responsibility in consent practices and reveals how online consent functions as a procedural formality instead of a valid protection of autonomy. Thus, failures of consent mechanisms are evident in digital landscapes and leaves users vulnerable despite being designed to protect autonomy.      
 

Regulatory Gaps

​

The inadequacy of consent mechanisms reveals a broader regulatory incapacity to govern technology, especially AI-driven systems. AI exposes fundamental shortcomings of traditional regulatory models, challenging Easterbrook’s ‘Law of the Horse’ argument that general legal principles are sufficient to govern advancing technology. Easterbrook’s view emerged in a time period where digital systems were static and had limited societal consequences; by contrast, contemporary technologies pose increased systemic risks and are rapidly changing. This leads to increased strain on existing regulatory frameworks. The rapid, exponential advancement of AI has intensified the pacing problem,as constant innovation widens the gap between technological development and effective regulatory frameworks.. Regulatory models often fail because they are not designed for dynamic technologies, and instead remain mostly reactive with post-hoc responses, intervening only after harm has been done. In addition, regulators often lack the technical knowledge of developers, further hindering effective oversight (Currie et al., 2025). Currie et al. (2025) also note that AI’s ability to shift between minimal-risk and high-risk profiles can contribute to failures of regulations. These structural barriers underpin the gap between static governance frameworks and the dynamic evolution of current technological systems.

 

The European Union has introduced various legislative measures as current solutions to address online manipulation tactics that impede decision-making processes. The Digital Services Act (DSA) prohibits online platforms from creating or operating in a deceptive manner, intended to protect users from nudging that distorts their decision-making processes. In addition, the DSA requires risk assessments on cases of potential user manipulation. However, this measure remains largely reactive, and risk assessments may not capture the full scale of emerging AI behaviours. The Digital Markets Act (DMA) highlights the need for consent for personalized advertising and prohibits gatekeepers from using online interfaces to impair autonomy. Nevertheless, these consent requirements remain insufficient, as they do not prevent subtle manipulative techniques from influencing user decisions. The Artificial Intelligence Act (AIA) bans AI practices that exploit the vulnerabilities of specific groups based on age or disability (Faraoni, 2023). However, even with these prohibitions, enforcement still remains difficult, as subtle manipulative techniques are difficult to detect in practice, and the law struggles in anticipating the ways in which AI can influence behaviour. Thus, these measures are only a start to solving online manipulation, underscoring the persistent regulatory gaps in existing frameworks.     

 

To address these gaps, policymakers must prioritize adaptable  and anticipatory regulation, accounting for risks in policy design. Risk-based approaches must balance the classification of risk with the initiation of democratic processes to legitimize regulatory decisions. Embedding risk in policy design must be an ongoing process to ensure adaption across technical, legal, and ethical domains. This requires multi-level governance architecture with the ability to evolve at the same speed of dynamic technology, accounting for the evolving societal harm as it surfaces. At the global level, AI regulation varies significantly. From market-driven in the United States, a sectoral approach which relies on voluntary frameworks, to a state-centric model in China, a model aligned with ideological control. These differences underscore the need for global coordination to standardize regulatory frameworks. Instead of pursuing rigid uniformity, policymakers should value interoperability to create a harmonious and coherent global governance that supports responsible innovation while protecting fundamental rights (Currie et al., 2025). In the current age of big tech, this coordinated and risk-based approach becomes necessary to safeguard all users.    

Conclusion

 

​The rise of persuasive technology reveals a threat to autonomy by enabling widespread psychological manipulation. Through the exploitation of cognitive biases and platforms’ increased reliance on superficial cues, digital landscapes increasingly shape beliefs, emotions, and behaviours as supported by the ELM. This process is further exacerbated by dark patterns driving users toward preselected choices, discouraging authentic choice. Existing consent frameworks do not protect users,and are only a form of superficial disclosure that fails to address how manipulation is embedded structurally in choice architecture. To safeguard autonomy and democratic values, more adaptive regulations must account for dynamically evolving technology. Global cooperation is crucial in order to develop standardized legal frameworks to effectively combat online manipulation.
​​​​

​

​

References 

 

1. Brenncke, M. (2024). A theory of exploitation for consumer law: Online choice architectures, dark patterns, and autonomy violations. Journal of Consumer Policy, 47(1), 127–164. https://doi.org/10.1007/s10603-023-09554-7 

2. Currie, W. L., Leimeister, J. M., Schlagwein, D., & Willcocks, L. (2025). Rethinking technology regulation in the age of AI risks. Journal of Information Technology, 40(3), 236–245. https://doi.org/10.1177/02683962251378815 

3. Faraoni, S. (2023). Persuasive technology and computational manipulation: Hypernudging out of mental self-determination. Frontiers in Artificial Intelligence, 6, 1216340. https://doi.org/10.3389/frai.2023.1216340 

4. Flick, C. (2015). Informed consent and the Facebook emotional manipulation study. Research Ethics, 12(1), 14–28. https://doi.org/10.1177/1747016115599568 

5. Hou, Z. (2025). Analyzing the persuasion mechanism of AI-generated rumors via the elaboration likelihood model. Frontiers in Psychology, 16, 1679853. https://doi.org/10.3389/fpsyg.2025.1679853 

6. Jacobs, N. (2020). Two ethical concerns about the use of persuasive technology for vulnerable people. Bioethics, 34(5), 519–526. https://doi.org/10.1111/bioe.12683 

7. Leiser, M., & Santos, C. (2024). Dark patterns, enforcement, and the emerging digital design acquis: Manipulation beneath the interface. European Journal of Law and Technology, 15(1). https://ejlt.org/index.php/ejlt/article/view/990 

8. Richards, N. M., & Hartzog, W. (2019). The pathologies of digital consent. Washington University Law Review, 96, 1461–1503. 

9. Susser, D., Roessler, B., & Nissenbaum, H. (2019). Technology, autonomy, and manipulation. Internet Policy Review, 8(2). https://doi.org/10.14763/2019.2.1410 

10. Timms, R. G. (2025). All ‘dark patterns’ are ‘hostile patterns’: A hostility framework for understanding problematic digital interfaces. Ethics and Information Technology, 27(4). https://doi.org/10.1007/s10676-025-09856-z