How young users push back against AI personalization on social media

Digital well-being, advertising literacy, and perceived relevance emerge as strong, direct predictors of resistance. On the other hand, perceived ad fatigue shows no direct path to resistance in the aggregate model. Instead, fatigue operates indirectly by altering how users judge relevance and by shaping literacy, supporting the view that annoyance alone tends to prompt avoidance rather than active counter-arguing.

A new peer-reviewed study published in Societies (2025) reports that young adults resist AI-personalized advertising most effectively when three factors align: robust digital well-being, high advertising literacy, and active appraisal of whether a message is personally relevant. Perceived ad fatigue on its own does not trigger resistance, but it does so indirectly when it shapes relevance judgments and literacy.

The article, “Algorithmic Burnout and Digital Well-Being: Modelling Young Adults’ Resistance to Personalized Digital Persuasion,” provides a structural model of resistance in data-driven media environments.

What the researchers asked and how they tested it

The team seeks to explain how psychological and cognitive resources convert AI-shaped exposure into resistance, focusing on four predictors: perceived ad fatigue (PAF), digital well-being (DWB), advertising literacy (ADL), and perceived relevance (PR), with resistance to persuasion (RTP) as the primary outcome. The model draws on the Elaboration Likelihood Model and the Persuasion Knowledge Model to position relevance and literacy as the cognitive filters through which affect and knowledge shape responses.

Researchers fielded a cross-sectional survey of 637 participants aged 18–35, all active users of at least one personalized social platform. Responses were collected using recall-based framing of daily exposure to targeted ads, a design aligned with prior persuasion and content-fatigue research.

They estimated relationships with Partial Least Squares Structural Equation Modeling (PLS-SEM), tested mediation through 10,000-resample bootstrapping, and ran multi-group analysis (MGA) to compare effects by age, gender, education, social-media use, ad-skipping, and digital burnout. The team deliberately specified a linear, compensatory baseline to identify the most stable direct and indirect paths before future nonlinear modeling.

The study is situated in a context where algorithmic personalization has intensified exposure and raised concerns about digital overload, especially among young adults who experience high-frequency, cross-device targeting. Prior work has documented the rise of algorithmic fatigue and the need for literacy to help users switch off persuasive intent; this study addresses the methodological gap by modeling resistance with mediators and segment differences.

What the data show about resistance to AI-personalized persuasion

Digital well-being, advertising literacy, and perceived relevance emerge as strong, direct predictors of resistance. On the other hand, perceived ad fatigue shows no direct path to resistance in the aggregate model. Instead, fatigue operates indirectly by altering how users judge relevance and by shaping literacy, supporting the view that annoyance alone tends to prompt avoidance rather than active counter-arguing.

Mediation testing clarifies these pathways. Perceived relevance functions as a proximal cognitive filter, translating both emotional load and user resources into resistance; advertising literacy provides a second route through which fatigue and well-being influence outcomes. The analysis indicates full mediation for PAF through relevance and literacy, and partial mediation for DWB, which still retains a strong direct link to resistance alongside its indirect effects.

The conceptual rationale is consistent with theory: relevance can reduce resistance when a message fits well yet increase resistance when intrusiveness and over-personalization trigger skepticism; literacy sharpens detection of tactics and supports selective attention. By integrating affect (fatigue), cognition (literacy, relevance), and self-regulation (well-being), the model reframes resistance as a product of interpretation and capability, not simply exposure.

The authors report that their approach advances prediction-oriented modeling of resistance and centers user agency in data-driven media environments. Their results map the psychological pathways through which individuals counter AI-optimized targeting, offering a baseline structure for more complex, possibly nonlinear specifications in future research.

Who is most affected and what platforms and policymakers should change

Segment analysis shows that resistance is context-dependent, varying across demographics and usage behaviors. Age moderates key paths: the link between literacy and resistance is weaker among the youngest adults, suggesting that familiarity with platforms does not always translate into critical counter-arguing. Gender differences also appear: well-being–to-relevance and literacy–to-resistance paths are stronger in males, while the direct fatigue–to-resistance path is more negative for females, consistent with disengagement rather than defiance. Education splits show that high-school graduates rely more on literacy to build resistance, whereas doctorate-level respondents show diminished incremental gains from literacy.

Digital burnout amplifies several relationships. When burnout is frequent, fatigue’s connection to literacy strengthens and its overall contribution to resistance grows, indicating that emotional overload can heighten vigilance and push users toward defensive processing. Ad-skipping weakens the influence of relevance on resistance among heavy skippers, implying that habitual avoidance dampens elaboration. Social-media intensity moderates how well-being and fatigue shape literacy and relevance, with non-users and light users showing stronger cognitive links than heavy users.

Put together, these results support a contingent framework of resistance: defenses are strongest where users have both skills (literacy) and self-regulatory capacity (well-being), and where they evaluate fit rather than reacting solely to volume or repetition. The study underscores practical levers for platforms and policymakers: build tools and curricula that raise advertising and algorithmic literacy, adjust frequency and intrusiveness management to avoid fatigue spirals, and promote user-control features that support healthier digital habits.

The authors argue that centering user agency is essential as data-driven media expand. They point to resilience- and safety-focused measures, literacy construction, well-being interventions, transparency, and exposure management, as more appropriate than engagement maximization for healthier ecosystems.