Resistance or compliance? The impact of algorithmic awareness on people's attitudes toward online information browsing

The study investigates how algorithmic awareness shapes users’ compliance versus resistance when browsing online information. In the algorithm-saturated media environment, algorithms reduce cognitive burden but may confine users to information cocoons and affect creativity. Prior work shows mixed responses to algorithmic recommendations (privacy concerns, resistance, or cooperation), yet the psychological mechanisms remain unclear. The authors focus on endogenous experiences—perceived power and locus of control—as potential mediators linking algorithmic awareness to behavioral outcomes under different browsing purposes (purposeful vs purposeless). Research hypotheses: H1: Algorithmic awareness significantly affects users’ online browsing behavior. H2: Algorithmic awareness affects perceived power across contexts. H3: Perceived power affects online browsing behavior, with internal control (H3a) and external control (H3b) playing different mediating roles depending on browsing context.

The literature situates algorithmic awareness as users’ perception of algorithms’ presence, logic, and impact in media environments. Awareness can foster two behavioral orientations: resistance (disrupting or avoiding algorithmic curation) and compliance (strategically leveraging recommendations). Studies document tactics of resistance (e.g., TikTok users’ intentional misclicks) and drivers of compliance (relevance, perceived efficacy, boundary perception). Awareness may amplify engagement while not necessarily reducing platform fatigue, reflecting a paradox of informed use. Perceived power—users’ subjective sense of influence and control—can be enhanced by media technologies and is theorized to shape cognition and behavior, enabling selective attention and task focus. Algorithmic perception can increase satisfaction, trust, perceived usefulness, and a sense of information control, potentially reducing privacy concerns. Locus of control distinguishes internal versus external attributions over outcomes, with implications for how users negotiate algorithmic influence. Gaps remain in identifying the pathways from awareness to behavior (compliance vs resistance) and in explaining divergent choices under similar exposure; experimental evidence is limited compared to qualitative and survey-based work.

Design: 2 (algorithmic awareness: present vs absent; between-subjects) × 2 (browsing purpose: purposeful vs purposeless; within-subjects) mixed factorial design. The study examined compliance with algorithmic recommendations and the mediating roles of perceived power and sense of control (internal vs external). A pilot study preceded the lab experiment to curate neutral materials. Participants: Power analysis (G*Power, f=0.2, α=0.05, 1-β>0.80) indicated a minimum of 52. Seventy-five undergraduates were recruited (30 yuan incentive); after manipulation check exclusions (n=7), 68 remained (10 male, 58 female). Ethics approval: BNU School of Journalism and Communication IRB (BNUJ&C20241227022); informed consent obtained. Pilot study: An expert panel sourced 158 short texts (~100 words) with images from mobile-era platforms (e.g., Xiaohongshu/Rednote) across daily life and film/TV categories. N=199 participants rated liking, positivity/negativity, and familiarity on 7-point scales. Thirty neutral items (15 daily life; 15 film/TV) were selected based on central tendency and minimal group differences (no significant differences across categories). Materials and measures: Emotional scale (7 items: excitement, calmness, happiness, tension, relaxation, anxiety, sadness; 7-point Likert; Cronbach’s α=0.85) administered pre- and post-task. Compliance scale (5 items; 7-point Likert) assessed liking of materials, continued browsing intention, desire to see more, liking of push method, and liking of that procedure segment. Perceived power: Revised General Perceived Power Scale (Anderson & Galinsky, 2006; 7-point Likert). Sense of control: Adapted Levenson (1981) subscales for internal control and powerful others (external) control (7-point Likert). Procedure: After consent and pretest emotions, participants were randomly assigned to algorithmic awareness (told information was recommended based on behavioral habits) or no awareness (told information was completely random). All participants completed both purposeless browsing (free browsing while waiting; press J to advance) and purposeful browsing (task: find a film suitable for a class meeting), with 15 trials per condition, order randomized (E-Prime 3.0). Critically, actual presentation was random for all participants (algorithm-neutral). Post-task measures included emotions, compliance, perceived power, and sense of control. Manipulation check: A single-choice item (algorithmically filtered vs randomly filtered) and a 7-point belief item on algorithmic filtering verified awareness activation. Analysis: Mixed ANOVA tested main and interaction effects on compliance. PROCESS (Hayes) mediation models were applied: Model 4 for simple mediation by perceived power; Model 6 for chained mediation via perceived power → internal/external control → compliance. Bootstrap samples=5,000; 95% bias-corrected CIs.

Manipulation and emotions: Algorithmic awareness manipulation succeeded; participants in the awareness group reported higher belief in algorithmic filtering (M=4.69, SD=1.85) than the no-awareness group (M=2.50, SD=1.08), t(66)=5.882, p<0.001, Cohen’s d>1. Emotional states showed no significant pre-post differences across excitement, pleasure, peace, relaxation, tension, anxiety, or sadness (all ps>0.05), indicating neutral stimuli did not alter affect. Compliance (mixed ANOVA): Main effect of browsing purpose: F(1,66)=7.83, p=0.01, η²=0.11; compliance higher with purpose (M=5.37, SD=0.14) than without (M=5.09, SD=0.13). Main effect of algorithmic awareness: F(1,66)=10.61, p<0.01, η²=0.14; compliance higher when users believed items were algorithmically recommended (M=5.36, SD=0.18) vs random (M=4.83, SD=0.18). Interaction not significant: F(1,66)=1.28, p=0.26, η²=0.02. Simple effects: Under algorithmic recommendation belief, compliance was higher in purposeful (M=5.83, SD=0.19) than purposeless browsing (M=5.44, SD=0.19), F(1,66)=7.72, p=0.01, η²=0.11. Under random belief, purposeful (M=4.75, SD=0.19) vs purposeless (M=4.91, SD=0.19) did not differ, F(1,66)=1.39, p=0.24. Mediation during purposeful browsing: Model 4 showed algorithmic awareness positively predicted compliance (β=0.78, SE=0.27, p=0.01) and marginally increased perceived power (β=0.62, SE=0.32, p=0.06). Perceived power positively predicted compliance (β=0.23, SE=0.10, p=0.03). Indirect effect via perceived power significant: Effect=0.14, SE=0.10, 95% CI [0.00, 0.42], accounting for ~18% of total effect. Chain mediation (Model 6): Awareness → perceived power → internal control → compliance was significant: Effect=0.09, SE=0.07, 95% CI [0.01, 0.29]. The path via external control was not: Effect=−0.00, SE=0.01, 95% CI [−0.06, 0.01]. Mediation during purposeless browsing: Awareness increased compliance (β=0.62, SE=0.27, p=0.02), but perceived power did not predict compliance (β=0.12, SE=0.10, p=0.22), and the indirect effect via perceived power was not significant: Effect=0.08, SE=0.09, 95% CI [−0.04, 0.33]. Gender checks (Table 2) showed no significant differences in perceived power, internal/external control, or compliance (all ps>0.05). Overall: Algorithmic awareness robustly increased compliance across contexts, with psychological empowerment (perceived power and internal control) mediating effects specifically during purposeful browsing.

Findings address the core question of why users sometimes comply with and sometimes resist algorithmic recommendations. Algorithmic awareness, even in an algorithm-neutral setting, increased users’ willingness to comply with recommended content. Under purposeful browsing, awareness bolstered perceived power, which strengthened internal control, fostering higher compliance—users experienced algorithms as tools that could be consciously leveraged for efficient goal attainment. In purposeless browsing, awareness still elevated perceived power, but this did not translate into compliance via empowerment pathways; behavioral compliance may instead reflect heuristic processing and cognitive offloading without a felt increase in agency. Theoretically, the study integrates algorithmic awareness with perceived power and locus of control, demonstrating that internal control, not external control, links empowerment to compliance during goal-directed tasks. Practically, results suggest that transparent cues and explanations can transform passive curation into reflective, agency-oriented interactions, while also highlighting risks of algorithmic overreliance and information cocoons if users lapse into technological unconsciousness. The work refines understanding of how psychological empowerment mechanisms mediate user–algorithm power dynamics and clarifies boundary conditions by browsing purpose.

This research demonstrates that algorithmic awareness increases users’ compliance with recommended content across browsing contexts; during purposeful browsing, the effect is partially transmitted through perceived power and, in turn, strengthened internal control. The study contributes an experimentally validated mechanism linking awareness to behavior via empowerment, clarifying when and how users delegate decision-making to algorithms as instrumental aids. Implications include designing transparent, explanatory recommendation cues that promote conscious, goal-aligned use while counteracting technological unconsciousness and over-narrowed exposure. Future research should broaden samples beyond predominantly female undergraduates, test high-involvement and affectively charged content in more ecologically valid interfaces (e.g., short-video feeds, news apps), examine longitudinal dynamics of awareness and empowerment, and explore boundary conditions across cultures, ages, occupations, and levels of algorithmic literacy.

Generalizability is constrained by a predominantly female undergraduate sample aligned with the platform’s user base; gender effects were nonsignificant but caution is warranted for broader populations. The laboratory setting with affectively neutral stimuli enhances internal validity but may underestimate real-world emotional responses and ecological validity relative to personalized, high-stakes feeds. All stimuli were randomly presented; while this isolated psychological mechanisms, it differs from live algorithmic environments. Future work should diversify demographics, incorporate ecologically rich interfaces and emotionally varied, high-relevance content, and assess longitudinal and cross-cultural boundary conditions.