Algorithmic allure: a theoretical framework for dark AI patterns and the erosion of informed consent in AI-driven digital marketing

1 Introduction

Generative AI is radically reshaping digital marketing by transforming how firms engage with consumers, tailor messages, and influence purchasing decisions. Where earlier marketing technologies primarily automated repetitive tasks and improved market segmentation, contemporary AI systems continuously learn from user behavior, generate bespoke content, predict future preferences, and refine persuasive strategies in real time (Bormane and Blaus, 2024). These capabilities have enhanced firms' ability to deliver more relevant customer experiences, yet they have also given rise to mounting ethical, legal, and social challenges concerning transparency, autonomy, and meaningful informed consent Cheong (2024). The rapid diffusion of generative AI has accelerated this shift: large language models, image synthesizers (Cheong, 2024), conversational agents, recommender systems, and autonomous AI agents are now integral components of digital marketing ecosystems (Huang and Rust, 2021). These technologies enable automated content production, hyper-personalized advertising, and instantaneous modification of persuasive messages based on fine-grained behavioral profiles (Peter et al., 2025). As a result, AI has moved from a decision-support role to being an active shaper of consumer perceptions, preferences (Grewal et al., 2025), and purchase behavior through continuously adaptive interactions.

For many years personalization was celebrated as a central benefit of digital marketing, reducing information overload and improving customer satisfaction (Kaplan and Haenlein, 2019).‏ However, recent technical advances have blurred the boundary between beneficial personalization and algorithmic manipulation (Kirk et al., 2024). Modern AI systems can detect affective states, cognitive biases, and situational vulnerabilities at the individual level, enabling message adaptation grounded in psychological weakness rather than coarse demographic attributes (Matz et al., 2024). Consequently, personalization increasingly functions as behavioral steering rather than merely enhancing the customer experience. These developments have intensified scholarly interest in algorithmic persuasion (Sposini, 2024), AI ethics, and digital transparency (Cheong, 2024). One prominent research strand examines “dark patterns”, interface designs intentionally crafted to nudge users toward firm-centric choices (Mathur et al., 2019). Although valuable, this literature remains fragmented: dark-pattern studies emphasize deceptive user interfaces, AI-in-marketing research prioritizes efficiency and personalization, AI ethics work foregrounds fairness and accountability, and consumer behavior scholarship focuses on trust and privacy. Few studies, however, explicate how contemporary AI integrates multiple persuasive tools into adaptive systems that can generate deceptive content, alter tactics instantaneously, exploit vulnerabilities, and progressively undermine consumers' capacity for genuinely informed decisions.

The emergence of generative and agentic AI during 2024–2026 has exposed limitations in current scholarship (Grewal et al., 2025). High-profile cases have revealed new modes of commercial deception that exceed traditional dark patterns. An ABC News visual investigation uncovered numerous fraudulent online sellers using AI-generated elderly personas (Katersky et al., 2025), synthetic video, and fabricated emotional narratives to market mass-produced goods as artisanal. In another instance, an Airbnb dispute involved allegations that AI-generated images were used to fabricate evidence supporting an unwarranted damage claim (TheGuardian, 2025). The U.S. Federal Trade Commission pursued action against DoNotPay for overstating AI capabilities in marketing materials, a practice now described as AI-washing (FTC, 2025; TheGuardian, 2025). Microsoft Research's Magentic Marketplace experiments further demonstrate how autonomous agents can be manipulated via prompt injection and bespoke persuasion strategies to produce harmful market behaviors (MicrosoftResearch, 2025). Regulators have begun to respond. China enacted Measures for Labeling AI-Generated Synthetic Content (Cyberspace Administration of China et al., 2025), which took effect in September 2025 and mandates a dual-labeling regime—visible markers for users and embedded metadata—to counter AI-driven misinformation (Sun, 2025). South Korea implemented a risk-based approach through its Framework Act on AI (effective January 2026) (Center, 2025), requiring mandatory labeling for AI-generated advertisements and treating “virtual humans” used to endorse products as potentially unfair advertising (Chang, 2025). In Latin America, Brazil is debating legislation that would impose fines up to BRL 50 million or 2% of revenue on high-risk AI systems lacking transparency (VerifyWise, 2025), signalling emerging regulatory attention in the region. Traditional dark patterns rely on fixed interface tricks: misleading buttons, hidden unsubscribe options, or preselected defaults (Brignull, 2010). By contrast, AI-driven systems are dynamic: they monitor user behavior, continuously update predictions, generate personalized content, and self-modify strategies. Manipulation thus increasingly originates not from static design elements but from covert, adaptive processes operating continuously in the background (Gasinski et al., 2026). Accordingly, existing theoretical accounts of dark patterns and digital persuasion are insufficient to capture these emergent practices. This study advances the concept of Dark AI Patterns: adaptive, AI-enabled mechanisms based on machine learning, generative models, behavioral prediction, and autonomous systems that influence consumer cognition and conduct. Dark AI Patterns intentionally reduce transparency, weaken decision autonomy (Maciaszczyk et al., 2026), and undermine informed consent, all while preserving the outward appearance of legitimate personalization and customer service.

Against this backdrop, we propose Algorithmic Allure Theory (AAT) as an integrative framework. Rather than treating manipulative techniques as discrete tactics, AAT frames AI-mediated influence as a dynamic process involving algorithmic allure, cognitive priming, behavioral steering, and the gradual displacement of consumer autonomy. Using a multi-case qualitative design, we analyze seven internationally documented instances from 2024 to 2026 (see Figure 1 for an illustration of the AAT framework). Data were drawn from investigative journalism, regulatory reports, industry publications, and corporate documents, and analyzed with reflexive thematic analysis following Braun and Clarke (2022) to identify recurring mechanisms by which AI reshapes marketing practice and consumer decision-making.

This study makes four principal contributions: it establishes dark AI patterns as a distinct category of AI-driven deceptive marketing; introduces Algorithmic Allure Theory (AAT) to explain how adaptive AI systems progressively erode consumers' capacity for informed choice; proposes a conceptual taxonomy of deceptive AI practices; and derives practical implications for marketers, platform designers, and policymakers seeking to balance AI innovation with ethical standards and consumer protection.

2 Literature review

The integration of artificial intelligence into digital marketing represents one of the most significant shifts in contemporary commercial practice. Whereas earlier tools emphasized automation and basic market segmentation, current systems powered by machine learning (Rahman et al., 2026), generative models, and autonomous agents can deliver extreme personalization, produce real-time content, and exert predictive influence on behavior (White, 2023). Although these capabilities can enhance customer experiences, they also blur the line between beneficial personalization and covert manipulation: consumers appreciate tailored content but increasingly worry about the opaque mechanisms that generate it. Research on algorithmic persuasion shows that AI exploits cognitive biases through indirect influence strategies. Drawing on dual-processing theories such as the Elaboration Likelihood Model (Keng et al., 2025), scholars demonstrate that personalized messages frequently rely on peripheral cues—social proof, authority, scarcity—rather than promoting deliberative, rational processing. Nudge theory has consequently evolved into the concept of “fine- grainednudges” (Apriliani, 2025), where choice architectures dynamically adapt to continuous behavioral feedback, making manipulative interventions harder to detect.

Work on traditional dark patterns has documented many static deceptive interface designs, including hidden defaults and misleading UI elements (Mathur et al., 2019). Generative AI, however, introduces dynamic capabilities—synthetic content generation, affective targeting, and autonomous self-optimization—that far exceed the boundaries of fixed interface tricks (Gray et al., 2018). Actions that previously required substantial human effort are now scalable and continuously adaptive in real time (Zou, 2025). Ethics and governance frameworks—such as the OECD AI Principles (OECD, 2019), UNESCO recommendations (UNESCO, 2021), and the EU AI Act—emphasize transparency (EU, 2024), accountability, and human oversight, and they classify exploitative systems as high-risk. Yet regulatory emphasis on disclosure alone often proves inadequate: users may know an AI is present but cannot perceive how adaptive persuasion continually reshapes their decisions. Adobe's Trust Report and consumer studies show declining trust driven not only by lack of disclosure but also by feelings of inauthenticity and diminished autonomy (Adobe, 2026).

Taken together, the literature remains fragmented. Marketing scholarship highlights personalization benefits and efficiency; persuasion research isolates specific mechanisms; dark-pattern studies remain anchored to static UI designs; and AI ethics focuses on governance. There is scant theoretical work explaining how generative and agentic AI integrate these elements into interconnected, adaptive persuasive systems. This study addresses that gap by developing the concepts of Dark AI Patterns and Algorithmic Allure Theory.

3 Theoretical framework: algorithmic allure theory

3.1 Defining dark AI patterns

Building on the established concept of dark patterns (Brignull, 2010) and (Mathur et al., 2019), this study argues that artificial intelligence has profoundly extended both the mechanisms and reach of digital manipulation. We define Dark AI Patterns as adaptive (Schilke and Reimann, 2025), AI-powered persuasive mechanisms that deliberately employ machine learning, generative models, behavioral prediction (Ostinelli et al., 2025), synthetic content production, affective targeting, or autonomous self-optimization. These mechanisms influence consumer perception and behavior by reducing transparency (Park and Yoon, 2025), exploiting psychological vulnerabilities, and progressively weakening meaningful informed consent, while preserving the outward appearance of legitimate personalization, customer assistance, or commercial optimization (Sun et al., 2026). This definition extends traditional understandings of deceptive digital design along four principal dimensions (Apriliani, 2025; Gray et al., 2018; Keng et al., 2025; Mathur et al., 2019). First, manipulation is no longer confined to interface components; it encompasses the entire persuasive ecosystem constructed by AI. Second, persuasion becomes an adaptive, ongoing process rather than a fixed intervention, continuously evolving in response to consumer interactions. Third, AI enables a new register of synthetic authenticity through the creation of fabricated identities, narratives, and evidentiary artifacts that appear real. Fourth, optimization processes gain autonomy: systems continuously self-improve without constant human oversight (see Table 1 for the conceptual differences between conventional dark patterns and Dark AI Patterns). This shift indicates that the move from traditional dark patterns to Dark AI Patterns is not merely a technical upgrade but a conceptual transformation. AI relocates deception from isolated design elements to adaptive persuasive infrastructures capable of continuously reshaping digital choice environments.

Table 1. Conceptual distinction between dark patterns and dark AI patterns.

Source: Prepared by the researchers based on previous studies.

3.2 Theoretical foundations of algorithmic allure theory

While the Dark AI Patterns concept describes the technical substrate of adaptive manipulation, it does not fully explain the psychological process through which such manipulation incrementally reshapes consumer judgment. Algorithmic Allure Theory (AAT) fills this gap by offering an integrative framework that explains how computational capacities are converted into persuasive power that affects human perception and undermines informed consent. Unlike conventional persuasion theories that often treat influence as a one-off message transmitted from marketer to consumer, AAT conceptualizes persuasion as a socio-technical, continuous process that emerges from the ongoing interplay among AI systems, digital environments, and human cognition. The central construct—Algorithmic Allure—captures the capacity of intelligent systems to create decision environments that, over time, appear more attractive, credible, personalized, and emotionally resonant. Consequently, consumers increasingly perceive algorithmic recommendations as trustworthy and aligned with their interests. Crucially, this felt allure is rarely experienced as external coercion; individuals tend to interpret algorithmic suggestions as expressions of their own preferences, thereby lowering their recognition of underlying commercial influence. AAT integrates elements from the Elaboration Likelihood Model (Petty and Cacioppo, 1986), Social Proof Theory (Cialdini, 2021), and Nudge Theory (Thaler and Sunstein, 2008), alongside insights from computational behavior analysis. Rather than treating these approaches as competing accounts, AAT frames them as interdependent processes operating within a broader adaptive system.

3.3 The Six sequential stages of algorithmic allure

AAT models algorithmic persuasion as a sequential, six-stage process that begins with foundational observation and culminates in algorithmic governance of the commercial environment: AI systems continuously collect behavioral, contextual, affective, and transactional data to build rich predictive profiles of consumers. This observation stage provides the informational substrate for all subsequent mechanisms. Machine-learning models translate raw data into psychological inferences—susceptibility to emotional appeals, price sensitivity, impulsivity, or decision fatigue. Generative AI augments this interpretation by analyzing natural language, vocal cues, and visual signals, enabling systems to infer not only what consumers do but also plausible reasons for their actions. This stage constitutes the core of AAT. Rather than sending uniform messages, AI dynamically constructs persuasive micro-environments tailored to each consumer: personalized recommendations, narratives aligned with current affective states, synthetic identities, adaptive pricing, fabricated testimonials, and context-sensitive visual elements. The high degree of personalization engenders a powerful illusion of authenticity; recommendations appear as sincere personal assistance rather than commercial persuasion, thereby amplifying algorithmic allure. Once attraction is established, AI subtly steers consumer behavior without overtly removing choice. Systems reorganize alternatives, reorder recommendations, reframe emotional contexts, time message delivery, and vary decision complexity. Unlike static dark patterns, this steering continually adapts based on each interaction, as systems learn which tactics elicit stronger behavioral responses. Over time, decision authority progressively shifts from the consumer to autonomous AI agents acting on their behalf. Consumers increasingly rely on agentic recommendations because they are faster, more convenient, and perceived as more objective. Repeated reliance reduces independent search and critical evaluation. However, as the study later demonstrates, these delegated agents themselves can become vectors for manipulation. In the final stage, AI transcends influencing isolated choices and becomes the infrastructural backbone of digital markets: controlling what consumers see, how products are presented, when prices change, and how persuasion strategies evolve through continual learning. Persuasion thus ceases to be merely a set of discrete messages and becomes embedded within market architecture.

• Stage 1: Algorithmic Observation (behavioral-surveillance profiling)

• Stage 2: Algorithmic Interpretation (inferential psychographic modeling)

• Stage 3: Algorithmic Attraction (adaptive micro-environment generation — the feedback-loop core of AAT)

• Stage 4: Behavioral Steering (reinforcement-driven choice-architecture modulation)

• Stage 5: Algorithmic Delegation (authority transfer via agentic reliance)

• Stage 6: Algorithmic Governance (infrastructural embedding in market architecture)

AAT asserts that the erosion of meaningful informed consent is not sudden or dramatic but emerges gradually across thousands of seemingly benign interactions. Each stage builds on the previous one to produce a cumulative effect that reshapes consumer autonomy while preserving the appearance of free choice. From the case material, the study inductively derives four interrelated theoretical propositions: These propositions are grounded in shared qualitative patterns across the examined cases rather than statistical testing. They therefore serve as conceptual foundations for future empirical work—experimental, survey, or ethnographic—aimed at unpacking consumer experiences within AI-augmented environments (see Table 2).

• P1: As AI capabilities—especially real-time generative content, behavioral prediction, and autonomous self-optimization—expand, Dark AI Patterns become more complex and more concealed. Richer data access and more powerful models shift deception from simple UI tricks to deep contextualized personal influence that is difficult for consumers and regulators to detect.
• P2: Higher levels of algorithmic allure increase the likelihood that consumers will perceive algorithmic recommendations and environments as trustworthy, authentic, and congruent with their preferences. Case evidence shows personalization creates an emotional alignment that renders commercial suggestions experienced as genuine self-discovery rather than externally driven persuasion, thereby reducing critical scrutiny.
• P3: Continuous behavioral steering within adaptive environments diminishes independent cognitive evaluation and trains individuals to delegate judgments to algorithmic guidance. Repeated exposure conditions consumers to rely on algorithmic solutions, transferring cognitive burden from person to system.
• P4: Prolonged engagement with adaptive persuasive systems incrementally degrades the meaningfulness of informed consent, even where surface transparency and voluntariness are present. From this perspective, informed consent is not a one-time legal artefact achieved by disclosure but a fragile, ongoing cognitive state vulnerable to erosion within environments engineered to exploit human vulnerabilities—raising fundamental questions about the future of individual autonomy in AI-mediated marketplaces.

Table 2. Operationalizing the AAT propositions.

Source: Prepared by the researchers.

4 Materials and methods

4.1 Study design and strategy

Given the exploratory nature of this research and the relative novelty of AI-enabled deceptive practices, the study adopts a qualitative multiple-case design (Yin, 2018). Rather than testing predetermined hypotheses, the objective is theory building grounded in empirical phenomena. Instances of AI-mediated marketing deception are treated as practical manifestations of broader algorithmic mechanisms rather than as isolated regulatory failures. Following established approaches to theory construction from case studies (Eisenhardt, 1989), a multiple-case strategy was chosen to enable analytical replication across contexts instead of statistical generalization (Eisenhardt and Graebner, 2007). Each case functions as an independent context in which the emergent theoretical framework can be tested and refined. The unit of analysis is the AI-enabled persuasive mechanism within each case rather than the firm itself. This focus facilitates cross-contextual comparison while allowing identification of higher-order conceptual themes shared across diverse settings.

4.2 Case selection and data sources

Cases were selected using purposeful theoretical sampling (Patton, 2015), guided by five criteria: (1) clear evidence that AI (generative models, machine learning, large language models, or autonomous agents) played a central role in persuasive processes; (2) the case involved consumer-facing commercial communication; (3) robust public documentation from regulatory agencies, investigative journalism, or institutional reports was available; (4) the case had observable effects on consumer judgments, behavior, trust, or decision outcomes; (5) the incident occurred between January 2024 and June 2026, capturing the latest wave of generative AI marketing applications.

Applying these criteria yielded seven internationally documented cases spanning e-commerce, digital platforms, legal-tech applications, digital advertising, autonomous-agent experiments, and AI-managed retail platforms (see Table 3). This sampling strategy achieved extensive contextual variation while maintaining analytical focus on conceptually comparable phenomena. To strengthen credibility through triangulation (Lincoln and Guba, 1985), data were collected from multiple public sources: official regulatory documents and filings, government reports, corporate disclosures, investigative journalism, industry analyses, and peer-reviewed academic literature. For example, the DoNotPay case was primarily informed by U.S. Federal Trade Commission records (FTC, 2025), whereas the AI-generated artisan store case relied on a detailed ABC News visual investigation supplemented by related reporting and sector commentary (Katersky et al., 2025).

Table 3. Overview of the selected cases.

Source: Prepared by the researchers based on public reports and statistics.

4.3 Data analysis and trustworthiness

Data were analyzed using reflexive thematic analysis following Braun and Clarke (2022). Analysis proceeded through six iterative phases: immersion in the material, initial coding of persuasive mechanisms, development of preliminary themes (Braun and Clarke, 2022), systematic review and refinement of themes across cases, definition and naming of higher-order themes, and integration of themes into the emergent theoretical framework. Coding employed an abductive logic that combined deductive sensitivity to theoretically relevant constructs with inductive openness to novel patterns. The coding process prioritized conceptual salience, explanatory power, and theoretical coherence over mere frequency counts. Trustworthiness was assessed using Lincoln and Guba's (1985) criteria (Lincoln and Guba, 1985). Credibility was bolstered through methodological triangulation and reliance on authoritative sources, along with systematic cross-case comparisons. Transferability was supported by providing rich contextual descriptions of each case. Dependability was addressed through an audit trail documenting sampling decisions and analytic procedures. Confirmability was maintained by keeping explicit links between primary evidence, coded excerpts, and emergent themes that informed theory development.

Limitations associated with multiple-case qualitative designs are acknowledged: findings are interpretive and contextually grounded rather than statistically generalizable. Nevertheless, the approach is particularly well suited for investigating fast-moving, complex phenomena that are difficult to capture via surveys or controlled experiments. By offering a structured, replicable methodology for examining emergent AI-mediated deceptive practices, this study contributes a rigorous qualitative template that future empirical work—experimental, survey-based, or ethnographic—can adapt and test.

5 Results: a taxonomy of dark AI patterns

Comparative analysis across the seven cases reveals that contemporary commercial manipulation powered by AI far exceeds the static, interface-based deceptions emphasized in prior work (Gray et al., 2018; Mathur et al., 2019). AI no longer merely reshapes digital interfaces; it reconstructs identities, evidence, social signals, technical authority, authenticity, and even autonomous decision processes. Despite differences in sector, technology, and geography, the cases converge on a set of adaptive persuasive strategies that systematically reshape consumer judgments while diminishing users' recognition of commercial influence. This convergence suggests these instances are not isolated anomalies but evidence of an emergent class of algorithmic persuasion systems. Using reflexive thematic analysis, seven high-level themes emerged. The themes are not mutually exclusive: several cases combined multiple mechanisms simultaneously, indicating future AI-enabled manipulation will be increasingly hybridized and context-dependent.

5.1 Synthetic identity manipulation

The first theme concerns the algorithmic fabrication of commercial identities designed to cultivate trust and emotional connection. Synthetic Identity Manipulation denotes the use of generative AI to create or alter fictional commercial personas that possess perceived emotional credibility, apparent authenticity, and interpersonal trust to influence purchasing choices (Zhou et al., 2026). Unlike conventional identity impersonation in cybercrime, these personas do not necessarily replicate real individuals; rather, generative models construct wholly fabricated characters that are psychologically persuasive through familiar narrative tropes, emotionally charged language, and culturally congruent imagery. The clearest illustration appears in investigative reports about fraudulent e-stores that deployed AI-generated personas, synthetic videos, and fabricated emotional backstories to market mass-produced goods as artisanal (ABC News and related investigations, 2024–2025). These operations relied on repeated names and narratives, engineered urgency and clearance events, AI-generated video content, and limited opportunities for independent verification. Comparative analysis identified four recurring elements: emotionally laden hardship narratives, manufactured scarcity cues (Cialdini, 2021), intentionally imperfect visual markers to signal authenticity, and constrained verification channels.

Within Algorithmic Allure Theory (AAT), Synthetic Identity Manipulation bridges the Algorithmic Interpretation and Algorithmic Attraction stages: behavioral data identifies the most resonant emotional narratives, and generative models deploy those narratives to assemble convincing commercial identities.

5.2 Synthetic social proof

The second theme addresses the falsification of perceived collective endorsement. Social proof is a longstanding cognitive heuristic used under uncertainty (Cialdini, 2021); generative AI undermines the assumption that online reviews and user-generated content reflect genuine experience. Synthetic Social Proof refers to the algorithmic generation, amplification, or coordination of consumer opinions, reviews, recommendations, or behavioral signals to create the impression of broad market acceptance. The Rytr/Operation AI Comply case (FTC, 2025) is salient: services marketed the capacity to auto-generate persuasive customer reviews for distribution across platforms. Regulators concluded these practices enable production of misleading testimonials that can falsely signal product popularity and quality. Compared with human-authored fake reviews, AI-generated reviews scale more easily, align with context, vary linguistically, and can be tailored to different consumer segments. This theme frequently co-occurs with Synthetic Identity Manipulation and AI Washing, indicating these mechanisms often operate as an integrated persuasive ecosystem.

5.3 Synthetic evidence manipulation

The third theme shifts from pre-purchase influence to post-transaction reality construction. Synthetic Evidence Manipulation denotes creating, altering, or leveraging AI-generated visual, textual, audio, or multimodal artifacts to shape judgments about contractual claims, liability, or consumer responsibility. The Manhattan Airbnb dispute (TheGuardian, 2025) exemplifies this pattern: a host submitted images purportedly evidencing severe property damage to support a large claim, but investigative indicators suggested the images were AI-generated or manipulated, prompting the platform to retract the claim and remediate the guest. This case shows how photographs—long considered high-trust evidence (Messaris, 1997)—can be algorithmically falsified, creating an information asymmetry between those who can produce synthetic evidence and consumers lacking reliable detection tools.

5.4 AI washing

The fourth theme concerns overstated AI capabilities used for marketing persuasion. AI Washing occurs when organizations exaggerate or misrepresent the functions and reliability of purported AI systems to bolster credibility without sufficient technical basis. Operation AI Comply targeted such claims, notably DoNotPay, which was marketed as an “AI lawyer” despite regulatory findings that its legal capabilities were not consistently reliable (FTC, 2025). AI Washing leverages a technical authority bias: people tend to perceive AI outputs as more objective or trustworthy than human judgments (Logg et al., 2019), and exaggeration of capabilities exploits that bias to promote adoption or premium pricing without commensurate performance.

5.5 Agentic AI manipulation

The fifth theme marks a fundamental shift: persuasion aimed not only at humans but at autonomous AI agents. Agentic AI Manipulation involves designing signals, reputational cues, or prompt payloads intended to influence independent AI agents that make or recommend purchase decisions on behalf of users. Academic and industry multi-agent market simulations demonstrate that agentic systems can be swayed through fake authorities, manipulated reputation signals, and prompt injection techniques—producing vulnerabilities like algorithmic satisficing and undue deference to perceived authority. These weaknesses challenge consumer protection frameworks that presume human decision-makers and reveal novel vectors for manipulation in agentic marketplaces.

5.6 Synthetic authenticity

The sixth theme highlights a paradox of AI-generated content: high technical quality does not guarantee persuasive legitimacy if audiences detect a lack of genuine human intent. Synthetic Authenticity refers to AI-produced content that is technically polished but judged by audiences as emotionally hollow or inauthentic (Zhou et al., 2026). The McDonald's Netherlands AI Holiday ad case (Adformatie, 2025) illustrates this dynamic: the campaign drew criticism not for AI use per se but for perceived emotional emptiness—viewers labeled the content “robotic” or “soulless”. Adobe's Trust Report (2025) supports this pattern, showing many consumers lose trust in brands that deploy AI content without transparent signaling or respect for human authenticity.

5.7 Autonomous commercial manipulation

The seventh and most advanced theme describes systems that progressively remove meaningful human oversight. Autonomous Commercial Manipulation denotes AI systems that independently manage commercial decisions—catalog descriptions, customer interactions, dynamic pricing, and operations—with minimal human supervision, generating risks of amplified or emergent manipulative behaviors. Documented AI-managed e-commerce platforms illustrate that full delegation can yield short-term efficiency gains but also produce exaggerated claims, inconsistent policies, or price volatility perceived as unfair, complicating accountability attribution.

5.8 Comparative synthesis of themes

Viewed together, the seven themes constitute interlocking components of a computationally integrated architecture that incrementally reshapes consumer autonomy. Each pattern undermines a distinct dimension of trust—interpersonal trust, collective endorsement, evidentiary trust, technological authority, agentic reliability, emotional authenticity, and institutional accountability—and their sequential interplay reflects an evolving persuasive arc that moves from drawing attention to reconstituting perceptual reality itself. Importantly, the erosion of meaningful informed consent often arises not from straightforward informational deficits but from continuous optimization of the informational and persuasive environment by adaptive AI systems. This empirical architecture underpins the six-stage Algorithmic Allure Theory and sets the stage for the ensuing theoretical discussion.

6 Discussion

6.1 Reframing digital persuasion in the generative AI era

The findings indicate that digital marketing has entered a new evolutionary phase. AI is no longer merely a supporting tool; it is the infrastructure through which persuasion is designed, executed, optimized, and governed. Generative AI shifts emphasis away from crafting individual persuasive messages toward constructing the entire informational environment that surrounds consumers. This shift requires scholars to move analytical focus from discrete persuasive content to the architecture of persuasion itself.

6.2 Extending existing persuasion theories

The results corroborate classical persuasion accounts but also extend them. The Elaboration Likelihood Model explains how Dark AI Patterns exploit peripheral routes—authority, scarcity, and social proof—yet Algorithmic Allure Theory (AAT) adds that AI does more than deploy these cues: it creates the context that confers those cues their meaning and tailors that context to each person. Social proof theory faces a new challenge in the generative AI ability to manufacture the appearance of consensus, while nudge theory must accommodate choice architectures that evolve continuously in real time rather than remain relatively static (see Table 4).

Table 4. Comparing algorithmic allure theory (AAT) with its parent persuasion theories.

Source: Prepared by the researchers.

As Table 4 shows, AAT does not merely relabel its parent theories but specifies the mechanism that distinguishes AI-mediated persuasion from the discrete, message-level events these theories were built to explain: a closed observation–action feedback loop in which the outcome of one persuasive attempt is fed back into the system and used to generate the next attempt within the same interaction. This mechanism implies a clear boundary condition: AAT's dynamics should dominate over classical, message-level persuasion routes only where a system has (i) sustained, individual-level behavioural data, (ii) generative capability to vary content in near real time, and (iii) the autonomy to act on inferred outcomes without a new human-authored message each time. Where any of these conditions is absent — for example, a static, one-off advertisement with no behavioural feedback channel — persuasion should follow classical ELM dynamics instead. This yields a falsifiable, AAT-specific prediction (Proposition P2, Table 2): perceived authenticity of an algorithmic recommendation should rise with the frequency of adaptive re-targeting even when message content is held constant, a pattern that social proof theory (which locates authenticity in the content of peer signals) and ELM (which locates attitude change in argument quality or peripheral cues, not re-targeting frequency) do not independently predict.

6.3 Algorithmic allure as an integrative theoretical framework

AAT functions as a holistic integrative frame that unites human cognition, machine learning, autonomous agents, and market governance in a single explanatory account of AI-augmented persuasion. Rather than isolating individual techniques, AAT emphasizes the coherence of the six algorithmic stages and expands the unit of analysis from single decisions to algorithmically mediated commercial systems.

6.4 Informed consent as a fragile construct in the face of algorithms

One of the study's central contributions is reconceptualizing informed consent. Traditional policy assumes disclosure and information provision suffice to secure consent, yet the cases show consumers may recognize AI presence while still remaining embedded in systems engineered to exploit their vulnerabilities. Hence informed consent should be treated as a dynamic, ongoing cognitive state continually reshaped by adaptive systems—not a one-time legal formality. Sustainable, ethical marketing will increasingly depend on balancing computational effectiveness with genuine human presence and respect for consumer autonomy. Consequently, the normative question shifts from whether AI can persuade more effectively to under what conditions AI-enabled persuasion remains ethically legitimate.

This fragility, however, does not imply consumers lack all capacity for resistance; evidence points to emergent forms of public pushback and discernment.

6.5 Consumer resistance as market self-correction

Consumers are not purely passive recipients of Dark AI Patterns; they are developing adaptive resistance strategies that blunt manipulative effects. The phenomenon of “synthetic authenticity” illustrates this dynamic: McDonald's Netherlands' AI-generated holiday ad provoked widespread criticism as emotionally hollow (Katerskyet al., 2025), signaling growing public sensitivity to the difference between technical polish and genuine human intent (TheGuardian, 2025). The cultural uptake of the term “AI slop” as Merriam-Webster's 2025 Word of the Year exemplifies increasing skepticism toward low-quality AI output (Merriam-Webster, 2025).

Quantitative indicators align with this trend: a Billion Dollar Boy study found consumer preference for AI-generated creator content dropped from 60% in 2023 to 26% in 2026 (Billion-Dollar-Boy, 2025), and Adobe's Trust Report shows sizable declines in brand trust when AI content is used without disclosure (Adobe, 2026). Such “passive resistance”—withdrawing trust and favoring brands that foreground human authenticity—constitutes a natural constraint on more blatant forms of algorithmic manipulation.

6.6 Toward ethical AI: transparency, empowerment, and human-centered algorithmic design

Despite rising concerns, AI also offers opportunities to enhance transparency and empower consumers when governed by ethical principles (Radanliev, 2025). Explainable AI, transparent recommender mechanisms, and privacy-respectful personalization can improve public understanding of algorithmic decisions and give consumers greater control over their data and choice (Ibitoye et al., 2025). Ethical design practices that foreground transparency, bias mitigation, accountability, and human oversight can support consumer autonomy rather than degrade it (Fukukawa and Trivedi, 2025).

In sum, the core challenge for AI-enabled marketing is not the persuasive power of algorithms per se but the governance contexts in which that power is exercised. Moving from algorithmic manipulation to ethical persuasion requires reframing transparency as more than disclosure—toward meaningful consumer comprehension and agency—and designing AI systems that preserve dignity, autonomy, and trust.

7 Contributions and implications

7.1 Theoretical contributions

This study delivers five principal theoretical contributions. First, it conceptualizes Dark AI Patterns as a distinct theoretical construct that extends the traditional dark-pattern literature (Mathur et al., 2019) from static interface tricks to adaptive, AI-generated persuasive ecosystems. Second, the study develops Algorithmic Allure Theory (AAT), a six-stage model that synthesizes insights from technology acceptance frameworks, the Elaboration Likelihood Model, Social Proof theory, Nudge theory, and Persuasion Knowledge theory into an integrated systems-level account. AAT reframes persuasion as a continuous computational process rather than a series of discrete behavioral interventions. Third, the paper reconceptualizes informed consent as a dynamic cognitive state vulnerable to gradual erosion through continuous algorithmic optimization, rather than as a one-time legal formality achievable by disclosure alone. This repositioning creates a conceptual bridge between marketing ethics, behavioral economics, and AI governance. Fourth, the study proposes a systematic taxonomy of seven AI-enabled manipulative mechanisms organized by persuasive logic rather than by the underlying technical stack, which enhances theoretical durability as technologies evolve. Fifth, the research advances a contextually grounded, dynamic understanding of consumer vulnerability: vulnerability emerges from an interaction between individual characteristics and the adaptive optimization capacity of AI systems, not solely from fixed personal traits. Collectively, these contributions tighten conceptual links between marketing research and AI governance literatures that have historically developed in relative isolation. By shifting the unit of analysis from individual persuasive acts to algorithmically mediated market systems, the study offers a new vocabulary and analytical lens for scholars investigating AI-mediated influence.

7.2 Practical and policy implications

For marketing practitioners, the findings caution that technical sophistication alone does not guarantee sustainable persuasive success or durable brand trust. Firms should balance traditional performance metrics with trust-related indicators—perceived authenticity, transparency, and relationship quality—and adopt internal AI-marketing ethics policies. Recommended operational practices include mandatory human review for high-impact campaigns, verification protocols for AI-generated content, explicit labeling of synthetic media, and periodic audits of personalization systems.

For platform operators and AI developers, the results endorse Ethical-by-Design approaches. Technical priorities include detection mechanisms for synthetic identities, AI-generated reviews, and fabricated evidence; defenses against prompt injection and related attacks on autonomous shopping agents; and the deployment of explainable AI interfaces that explicate why a recommendation or decision was produced. Such measures can reduce deception vectors and strengthen consumer comprehension.

Policymakers should recognize that existing consumer-protection regimes—largely oriented toward false advertising and unfair contract terms—are insufficient for algorithmically constructed informational environments. Priority regulatory actions include: These proposals align with emerging governance frameworks such as the EU AI Act, OECD AI Principles, and recent FTC enforcement actions.

• Standardized disclosure requirements for AI-generated commercial content.
• Mandatory human oversight for high-impact autonomous systems.
• Independent algorithmic auditing mechanisms.
• Clear liability rules for the use of synthetic evidence in commercial disputes.
• Stricter sanctions for AI washing and deceptive AI capability claims.

Consumer education is equally crucial. Digital literacy programs should extend beyond cybersecurity to include AI literacy: teaching people how to critically evaluate emotionally charged commercial narratives, recognize markers of synthetic identities and media, preserve transactional evidence, and interrogate unsupported claims about AI capabilities.

At an organizational level, the findings suggest a shift from AI-driven persuasion toward responsible AI-enabled persuasion: commercial success should be evaluated not only by behavioral impact but by adherence to transparency, accountability, authenticity, human oversight, and respect for consumer autonomy. The study crystallizes five guiding principles for responsible AI marketing: Transparency; Human Oversight; Authenticity; Algorithmic Accountability; and Consumer Autonomy.

7.3 Limitations and directions for future research

This research faces several limitations. The multiple-case qualitative design yields rich theoretical depth but does not permit immediate statistical generalization; the prevalence of Dark AI Patterns in the broader marketing ecosystem therefore remains an empirical question for large-scale quantitative work. The analysis relied primarily on documentary sources—regulatory filings, investigative journalism, and institutional reports—and despite systematic triangulation (Lincoln and Guba, 1985), source-related biases may persist. Finally, rapid advances in generative and agentic AI mean the specific cases analyzed will evolve even as the underlying mechanisms remain relevant.

Future research should (1) estimate the prevalence of Dark AI Patterns via surveys and experiments; (2) measure consumer awareness and resistance to algorithmic persuasion; (3) empirically test the propositions of Algorithmic Allure Theory across sectors and cultures; and (4) evaluate the effectiveness of regulatory and design interventions aimed at preserving meaningful informed consent in increasingly autonomous digital markets.

8 Conclusion

This study set out to understand how artificial intelligence transforms digital persuasion into adaptive systems capable of gradually undermining consumers' meaningful informed consent. Through a qualitative multiple-case analysis of seven internationally documented incidents, the research identified seven recurring mechanisms—Synthetic Identity Manipulation, Synthetic Social Proof, Synthetic Evidence Manipulation, AI Washing, Agentic AI Manipulation, Synthetic Authenticity, and Autonomous Commercial Manipulation—that together form a taxonomy the study terms Dark AI Patterns. This taxonomy extends prior literature on interface-based deception into the era of generative and agentic AI. Crucially, these patterns do not operate in isolation; they are interdependent components of a broader computational architecture that incrementally reshapes consumer autonomy and the very notion of consent. Persuasion is no longer confined to discrete marketing messages or UI designs; it is embedded within algorithmic environments that continuously observe users, interpret their behavior, predict preferences, and steer decisions.

To explain this shift, the study introduced Algorithmic Allure Theory (AAT). The theory describes how AI-mediated persuasion unfolds across six sequential stages: Algorithmic Observation, Algorithmic Interpretation, Algorithmic Attraction, Behavioral Steering, Algorithmic Delegation, and Algorithmic Governance. The study also reframes informed consent as a dynamic cognitive process rather than a static legal disclosure. Across all seven Dark AI Patterns, the erosion of consent rarely stemmed from a total lack of information. Instead, it resulted from the continuous algorithmic optimization of the surrounding informational environment. Even seemingly transparent interactions can compromise consumer autonomy when AI systems strategically shape perception, trust, and behavior. These findings have important implications for both scholarship and practice. For researchers, the proposed taxonomy and theoretical framework offer a solid foundation for future studies on AI-enabled persuasion, consumer resilience, digital ethics, and algorithmic governance. For practitioners and policymakers, the results underscore the need to embed transparency, accountability, explainability, and meaningful human oversight into AI-driven marketing systems. Responsible innovation should be judged not only by technological power or commercial results, but by how well AI supports—rather than undermines—consumer autonomy. Organizations that successfully balance algorithmic intelligence with authentic human judgment are likely to build stronger long-term trust and sustainable competitive advantage. In conclusion, the rise of Dark AI Patterns marks a transition from the age of digital persuasion to the age of algorithmic persuasion. Understanding and governing this shift is essential to ensure that artificial intelligence serves as a tool for responsible innovation rather than an invisible architecture of commercial manipulation. The future of digital marketing will depend less on raw technological capability and more on society's ability to guide that capability with transparency, accountability, and respect for human agency.

Generative AI statement

The author(s) declared that generative AI was used in the creation of this manuscript. During the preparation of this manuscript, the authors used ChatGPT (version GPT-4, OpenAI) to improve the readability and language of the English text. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the published article.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcomm.2026.1935069/full#supplementary-material

Supplementary material Supplementary material at the source

Footnotes

Edited by: Tereza Semerádová, Technical University of Liberec, Czechia

Reviewed by: Mohammad Al Haj Eid, Princess Sumaya University for Technology, Jordan

Najwa Jadallah, Palestine Technical University Kadoorie, Palestine