Cognitive,Normative,and Hedonic Pathways to Deepfake Acceptance: Integrating Knowledge-Attitude-Behavior and Goal Framing Theories

Deepfake in Education

Deepfake, a synthetic media generated through artificial intelligence and deep learning, has traditionally been associated with deception and misinformation (Ahmed, 2022; Vasist & Krishnan, 2023). Recent studies, however, emphasize its constructive potential in education, creative media, and geographic visualization (Roe et al., 2024; Zhao et al., 2021). In education, deepfake technology can enhance learning by animating historical figures, simulating training scenarios, and improving language acquisition (S. Alanazi et al., 2025; Gilbert & Gilbert, 2024). Nevertheless, concerns about authenticity and ethics remain. Therefore, integrating deepfake in education should involve media literacy initiatives that strengthen students’ critical evaluation and responsible engagement with digital content (Roe et al., 2024).

Given the dual nature of deepfake, presenting both educational potential and ethical risks, education is a critical domain for addressing these challenges. Instructional strategies enhance students’ ability to assess manipulated media (Aljalabneh, 2024), while initiatives like MIT’s Virtuality Lab integrate digital ethics into learning (Koperniak, 2022). Scholars emphasize embedding media literacy and ethical reasoning to promote constructive use (Roe et al., 2024) and prevent misuse (Naffi, 2024). This study advances beyond detection-focused approaches by examining how cognitive, normative, and hedonic factors shape university students’ evaluation and social acceptance of deepfake in higher education.

Theoretical Background

Conceptual Framework

Hedonic Goals and Social Acceptance

Perceived Enjoyment, Hedonic Enjoyment, and Social Acceptance

Based on GFT (Lindenberg & Steg, 2007), engagement with digital content, including deepfake, can be driven by hedonic goals emphasizing emotional experience and gratification. In entertainment-oriented contexts, such as synthetic media consumption, perceived enjoyment is particularly influential (M. Li & Wan, 2023; Stein et al., 2024; Xu & Thien, 2025). Defined as the immediate pleasure derived from interacting with AI-generated content, perceived enjoyment significantly affects user acceptance, especially in contexts of novelty, interactivity, and playfulness (van der Heijden, 2004; Venkatesh et al., 2012).

Hedonic enjoyment represents an outcome-based emotional response, including satisfaction, excitement, and a sense of adventure from sustained media engagement (Ahn et al., 2020; Horváth & Adıgüzel, 2018; Stein et al., 2024). Studies in social computing and immersive media show that it enhances prolonged affective connection and engagement with digital content (S. Alanazi et al., 2025; Gilbert & Gilbert, 2024). This corresponds to the hedonic goal frame, where behavior is driven by emotional gratification rather than instrumental or normative reasoning (Barbopoulos & Johansson, 2017).

Given the entertainment appeal of deepfake media (Cochran & Napshin, 2021), perceived and hedonic enjoyment are expected to influence users’ evaluation and acceptance of content. Based on this, the following hypotheses are proposed, as illustrated in Figure 1:

H7. Perceived enjoyment is positively associated with social acceptance.

H8. Hedonic enjoyment is positively associated with social acceptance.

OPEN IN VIEWER

Figure 1.

The present study’s research framework.

Moderating Effect of Fear of Missing Out

Fear of missing out (FOMO) is the anxiety that others are experiencing more rewarding events or opportunities (Tu et al., 2023). It reflects feelings of disconnection and missing out, often linked to the need for belonging, social anxiety, low self-esteem, fear of falling behind, and social isolation (Dutot, 2020; Good & Hyman, 2020; Przybylski et al., 2013; Wang et al., 2019), highlighting a strong drive for social connection.

Individual differences in FOMO influence knowledge, attitude, and perceptions toward emerging technologies. FOMO drives continuous connectivity to avoid missing trends or information (Groenestein et al., 2024). High-FOMO individuals may apply knowledge more actively to form favorable attitude, consistent with technology acceptance research showing that FOMO strengthens links between awareness, knowledge, and attitude (Gartner et al., 2022). Accordingly, this study hypothesizes that higher FOMO enhances the translation of deepfake knowledge into positive attitude.

FOMO influences moral reasoning and ethical evaluations, as individuals with high FOMO may align judgments with their attitude, especially when technology is perceived as socially beneficial (McKee et al., 2024). Ethical acceptability reflects moral legitimacy, while social acceptance denotes community approval (Taebi, 2017). High-FOMO individuals tend to follow social norms, so when deepfake is deemed ethically acceptable, they are more likely to socially accept it, supporting FOMO’s role in promoting adoption (Gartner et al., 2022).

Hedonic features, such as perceived entertainment or enjoyment of deepfake, may influence social acceptance, particularly among high-FOMO individuals. Research shows that FOMO-driven messaging enhances engagement with hedonic products (Munawar et al., 2021). Thus, if deepfake content is entertaining, high-FOMO individuals are more responsive, strengthening the link between hedonic appeal (perceived and hedonic enjoyment) and technology acceptance.

As shown in Table 1, FOMO has been studied as an independent, mediating, and moderating variable. Its moderating role in deepfake perception, awareness, and ethical responsibility remains underexplored. This study examines how FOMO moderates key indirect relationships within the deepfake acceptance framework, thereby extending the literature by assessing its influence on the strength of these associations. Thus,

H9a. FOMO positively moderates the relationship between knowledge and awareness.

H9b. FOMO positively moderates the relationship between knowledge and attitude.

H9c. FOMO positively moderates the relationship between attitude and awareness.

H9d. FOMO positively moderates the relationship between attitude and ethical acceptability.

H9e. FOMO positively moderates the relationship between ethical acceptability and social acceptance.

H9f. FOMO positively moderates the relationship between perceived enjoyment and social acceptance.

H9g. FOMO positively moderates the relationship between hedonic enjoyment and social acceptance.

OPEN IN VIEWER

Table 1.

FOMO on Social Media and Deepfake Studies.

Topic	Role	Authors
Psychological well-being and social media engagementDV, social comparison	IV*(+), MED**	Reer et al. (2019)
Problematic social media use, emotional support from social media, phubbing behaviorDV	MED**	Fang et al. (2020)
Exposure to social media influencers, followers’ purchase intentionDV	MOD**	Tu et al. (2023)
Problematic social media useDV, social comparison, self-esteem	MOD**	Servidio et al. (2024)
Addictive social media use, social media flowDV, mindfulness	IV*(+)	Brailovskaia and Margraf (2024)
Perceived risks and benefits when self-disclosing, social media trust, intention to self-disclosureDV	IV*(+)	Schyff and Flowerday (2023)
Information security awareness (ISA)DV, personality traits, ISA knowledge, attitude, and behavior	IV*(+)	Hadlington et al.(2020)
Cognitive ability, sharing of deepfake on social mediaDV	IV*(+)	Ahmed et al. (2023)
Post-traumatic stress disorder, believe in fake newsDV	IV*(+), MED**	Gong and Ren (2023)
Social media use, blocking misinformation source behavior on social mediaDV	IV*(+), MED**	Ahmed et al. (2024)
Cognitive ability, sharing of deepfake intentionDV	IV*(+), MED**	Ahmed (2022)

Note. IV = independent variable; MOD = moderation; MED = mediation; DV = dependent variable.

*

⁽⁺⁾Positive and significant; **Non-directional relationship and significant.

Research Design and Sample

This study used an online survey distributed via social media to Malaysian university students, employing non-probability convenience sampling. With 28.68 million active users in 2024 (83.1% of the population), social media effectively reaches the 18 to 34 age group, encompassing typical university students (Howe, 2024). As deepfake content is primarily circulated on these platforms (B. Lee, 2025), this approach ensured access to relevant participants directly exposed to manipulated media, enhancing data validity.

A total of 242 responses were collected from students enrolled in business and information technology (IT) programs using non-probability convenience sampling. To capture diverse perspectives, university students from both disciplines were included, reflecting educational differences influencing digital ethics and cybersecurity awareness (Creswell & Creswell, 2022; Owa, 2024). IT students possess stronger technical and cybersecurity foundations, while business students approach technology strategically and ethically (Ghazali et al., 2024). Academic background shapes cybersecurity awareness and responses to emerging technologies like deepfake (Hong et al., 2023; Watson et al., 2021), and including different disciplines highlights within-population variation despite shared ICT proficiency (An et al., 2023; Daengsi et al., 2022).

To ensure data authenticity and minimize response bias, participants were informed of the study’s purpose, background, and confidentiality. Participation was voluntary, anonymous, and without incentives, and a brief explanation of deepfake technology was provided to ensure understanding. Ethical approval was obtained, and all procedures adhered to institutional guidelines. Table 2 shows the demographic profile: 117 males and 118 females (mean age 21.41), including 120 diploma, 105 undergraduate, and 10 postgraduate students; 122 from IT and 113 from Business programs. While 50.2% expressed concern about deepfake, only 25.1% reported frequent exposure.

OPEN IN VIEWER

Table 2.

Sample Characteristics.

Description	N	%
Gender
Male	117	49.8
Female	118	50.2
Age	Mean = 21.41 years old
Education level
Postgraduate	10	4.3
Undergraduate	105	44.7
Diploma/Foundation	120	51.1
Education field
Business	113	48.1
Information Technology	122	51.9
Deepfake concerned (Mean = 3.45; SD = 1.244)
Not at all concerned	22	9.4
Slightly concerned	27	11.5
Neutral	68	28.9
Moderately concerned	59	25.1
Extremely concerned	59	25.1
Deepfake exposure (Mean = 2.70; SD = 1.120)
Never	45	19.1
Rarely	49	20.9
Sometimes	82	34.9
Often	50	21.3
Always	9	3.8

Measures

The survey used a five-point Likert scale (“1 = strongly disagree” to “5 = strongly agree”) for all items, with detailed items and sources provided in Table A1 (see Appendix A). Operational definitions of the nine constructs are outlined in Table 3. All items were adapted from validated scales and reviewed by academicians in information systems and educational technology to ensure clarity, relevance, and contextual appropriateness.

OPEN IN VIEWER

Table 3.

Operational Definition.

Operational definition	References
Knowledge is defined as the university students cognitive grasp of deepfake concepts.	Pundir et al. (2021); Zhen et al. (2022)
Awareness is conceptualized as “behavior” and is defined as university students’ overall knowledge of information security threats associated with deepfake.	Bulgurcu et al. (2010)
Attitude is defined as the university students’ evaluation of the desirability of verifying the authenticity of deepfake content on social networking platforms before sharing.	Pundir et al. (2021)
Ethical acceptability reflects university students’ criteria for evaluating whether deepfake information aligns with their personal moral standards.	M. Li and Wan (2023)
Perceived enjoyment is defined as the extent to which university students’ find watching deepfake content inherently enjoyable, independent of any practical or functional benefits derived from the content.	M. Li and Wan (2023)
Hedonic enjoyment is defined as the positive emotional responses and sense of satisfaction university students experience while watching deepfake content.	E. Kim et al. (2021)
Social acceptance is defined as the university students’ behavioral adoption of deepfake technology.	M. Li and Wan (2023)
Fear of missing out is defined as the pervasive anxiety experienced by university students that other might be having more rewarding experiences without them.	Ahmed (2022)
Conscientiousness refers to university students’ traits such as orderliness, impulse control, and adherence to norms.	Ahmed et al. (2024)

Data Screening

Data were collected via Google Forms and screened in SPSS to ensure quality. The form prevented missing responses, and no missing data were found. Seven cases exhibiting straight-lining were removed following Hair et al. (2017). Mahalanobis distance values were calculated to detect multivariate outliers (Tabachnick & Fidell, 2013), and none were identified. Consequently, all 235 remaining responses were retained for analysis.

A priori power analysis was conducted using GPower (Faul et al., 2009), following Hair et al. (2010), to assess sample adequacy. The measurement model included three arrowheads per endogenous construct and six interaction effects from two moderators. With α = .05, medium effect size (f² = 0.15; Singh, 2006), and power = 0.80, a minimum of 123 responses was required. As structural equation modeling typically requires at least 200 cases (Kline, 2011), the final sample of 235 responses exceeds both thresholds, supporting the dataset’s adequacy for analysis.

Common Method Bias

This study used a single-source, self-reported questionnaire, which could introduce common method bias (CMB; Podsakoff et al., 2003). A full collinearity test regressing latent variables on a common random variable showed variance inflation factor (VIF) values below 3.3 (Kock, 2015) for all constructs: ATT (1.455), AW (1.600), CS (1.251), EA (2.118), FOMO (1.138), HE (2.923), KTD (1.769), PE (2.776), and SA (2.147), indicating CMB is not a concern. Harman’s single-factor test revealed the first factor accounted for 22.387% of variance, below the 50% threshold (Podsakoff et al., 2003), further confirming minimal CMB risk.

Normality Assumption

Normality was assessed using latent variable scores, revealing Mardia’s multivariate skewness (β = 14.192, p < .05) and kurtosis (β = 125.606, p < .05), exceeding the ±3 skewness and ±20 kurtosis thresholds (M. K. Cain et al., 2017), indicating a violation of multivariate normality. Accordingly, a non-parametric approach was adopted, and bootstrapping was conducted using SmartPLS (Hair et al., 2021).

Assessment of the Measurement Model

The measurement model was assessed to establish construct reliability and validity (Table 4). Two items from the fear of missing out construct and two from conscientiousness were removed due to low outer loadings to improve average variance extracted (AVE). Specifically, the conscientiousness items “I am not someone who is careless” and “I have no difficulty getting started on tasks” were deleted. Although the Big Five traits were developed in Western contexts (Soto & John, 2017a), their expression varies across cultures. In this study, Malaysian students’ conscientiousness primarily reflected thorough and responsible task completion (mean = 3.550) and fulfilling task responsibilities (mean = 3.630), rather than task initiation (mean = 3.260) or avoiding carelessness (mean = 3.200), consistent with Muhamad et al. (2018).

OPEN IN VIEWER

Table 4.

Constructs Measurement.

Constructs	Items	Outer loadings	CR	AVE	CA	Mean (M)	Standard deviation (SD)
Knowledge	KTD1	0.758	.910	0.590	.885	3.490	0.944
	KTD2	0.789				3.710	0.803
	KTD3	0.723				3.740	0.789
	KTD4	0.739				3.730	0.888
	KTD5	0.819				3.690	0.857
	KTD6	0.768				3.560	0.867
	KTD7	0.779				3.920	0.783
Attitude	ATT1	0.900	.931	0.818	.889	3.860	0.960
	ATT2	0.892				3.870	1.022
	ATT3	0.921				3.930	1.025
Awareness	AW1	0.879	.918	0.788	.866	3.770	0.805
	AW2	0.885				3.660	0.748
	AW3	0.899				3.760	0.786
Ethical acceptability	EA1	0.869	.942	0.803	.918	2.890	0.956
	EA2	0.916				2.730	1.000
	EA3	0.930				2.720	0.998
	EA4	0.868				2.740	1.055
Social acceptance	SA1	0.904	.945	0.852	.913	3.090	1.091
	SA2	0.932				2.990	1.086
	SA3	0.932				2.900	1.076
Perceived enjoyment	PE1	0.933	.949	0.861	.919	2.940	0.998
	PE2	0.919				2.840	1.004
	PE3	0.931				2.780	1.031
Hedonic enjoyment	HE1	0.902	.961	0.860	.945	2.530	1.047
	HE2	0.955				2.510	1.103
	HE3	0.925				2.550	1.140
	HE4	0.927				2.500	1.137
Conscientiousness	CS1	0.673	.838	0.565	.749	3.580	0.846
	CS2 (deleted)	–				3.260	0.954
	CS3	0.810				3.550	0.790
	CS4	0.718				3.640	0.838
	CS5	0.798				3.630	0.807
	CS6 (deleted)	–				3.200	0.861
Fear of missing out (Ahmed, 2022)	FOMO1	0.702	.876	0.504	.837	3.010	0.947
	FOMO2 (deleted)	–				2.960	0.964
	FOMO3	0.655				2.900	1.003
	FOMO4	0.710				2.720	0.989
	FOMO5	0.761				3.190	0.948
	FOMO6	0.703				3.210	1.003
	FOMO7	0.744				3.000	0.918
	FOMO8 (deleted)	–				3.020	0.974
	FOMO9	0.688				3.140	0.962

Two items from the FOMO scale, “I fear my friends have more rewarding experiences than me” and “When I have a good time, it is important for me to share the details online (e.g., updating status),” were removed due to low outer loadings. This deletion aligns with evidence that FOMO expression varies across cultures. In collectivist societies such as Malaysia, younger individuals experience FOMO primarily through social comparison and staying informed about peers’ activities, rather than actively sharing personal experiences online (Ma’rof & Abdullah, 2024).

The remaining indicators’ outer loadings ranged from 0.655 to 0.933, exceeding the 0.50 threshold (Hair et al., 2017). Composite reliability values surpassed .70, and AVE values exceeded 0.50 for all constructs, confirming reliability and convergent validity. Cronbach’s Alpha values ranged from .749 to .945, meeting the .70 minimum for internal consistency (Hair et al., 2010). Discriminant validity was supported by HTMT values below 0.85 (Franke & Sarstedt, 2019) and the square roots of AVE exceeding inter-construct correlations (Fornell & Larcker, 1981), indicating all constructs were distinct (Table 5).

OPEN IN VIEWER

Table 5.

Discriminant Validity.

Construct	1	2	3	4	5	6	7	8	9	1	2	3	4	5	6	7	8	9
1. ATT										0.905
2. AW	0.467									0.413	0.888
3. CS	0.395	0.364								0.334	0.309	0.752
4. EA	0.192	0.098	0.112							−0.175	−0.088	−0.042	0.896
5. FOMO	0.153	0.165	0.171	0.313						−0.067	0.102	−0.021	0.284	0.710
6. HE	0.160	0.091	0.117	0.679	0.273					−0.146	−0.079	−0.011	0.634	0.248	0.927
7. KTD	0.503	0.627	0.447	0.110	0.180	0.086				0.457	0.560	0.373	0.032	0.120	−0.028	0.768
8. PE	0.038	0.092	0.072	0.597	0.271	0.801	0.179			0.030	0.081	0.007	0.549	0.247	0.746	0.151	0.928
9. SA	0.069	0.056	0.110	0.695	0.230	0.653	0.157	0.683		−0.049	−0.047	0.057	0.639	0.211	0.607	0.112	0.625	0.923

Note. Square root of AVEs in diagonal-bold; Columns 2 to 10 present the HTMT results, while columns 11 to 19 display the Fornell-Larcker criterion values.

Assessment of the Structural Model

Table 6 presents the structural model results, including path coefficients, standard errors, t-values, p-values, effect sizes, and inner VIF. Inner VIF values ranged from 1.150 to 2.932, below the 3.3 threshold, indicating no multicollinearity (Hair et al., 2019). Effect sizes (f²) indicate small to medium impacts (Cohen, 1988).

OPEN IN VIEWER

Table 6.

Summary of Hypotheses.

Hypothesis	β-value	t-Value	p-Value	Supported	Bias corrected confidence interval		VIF (≤3.3)	f 2	Effect size
Direct path					Lower	Upper
H1: KTD→AW	.443***	5.892	<.001	Yes	0.310	0.556	1.482	0.211	Medium
H2: KTD→ATT	.423***	7.306	<.001	Yes	0.315	0.510	1.200	0.208	Medium
H3: ATT→AW	.172**	2.556	.005	Yes	0.058	0.277	1.394	0.034	Small
H5: ATT→EA	−.172**	2.515	.006	Yes	−0.282	−0.053	1.150	0.030	Small
H6: EA→SA	.395***	5.350	<.001	Yes	0.271	0.515	1.984	0.171	Medium
H7: PE→SA	.341***	3.403	<.001	Yes	0.154	0.486	2.465	0.102	Small
H8: HE→SA	.108ns	1.015	.155	No	−0.054	0.294	2.932	0.009	Small
	Indirect effect						Direct effect
					Bias corrected confidence interval
Mediating path	β-value	t-Value	p-Value	Supported	Lower	Upper	β-value	p-Value	Results
H4: KTD→ATT→AW	.073*	2.322	.010	Yes	0.024	0.126	.443	<.001	Partial mediation
					Bias corrected confidence interval
Moderating path	β value	t-Value	p-Value	Supported	Lower	Upper
H9a: FOMO × KTD→AW	.083ns	0.957	.169	No	−0.050	0.231
H9b: FOMO × KTD→ATT	.140*	2.162	.015	Yes	0.038	0.250
H9c: FOMO × ATT→AW	.082ns	1.161	.123	No	−0.024	0.206
H9d: FOMO × ATT→EA	.169*	2.115	.017	Yes	0.042	0.299
H9e: FOMO × EA→SA	−.029ns	0.392	.347	No	−0.147	0.090
H9f: FOMO × PES→A	−.048ns	0.376	.354	No	−0.195	0.180
H9g: FOMO × HE→SA	.020ns	0.137	.445	No	−0.226	0.207
H10a: CS × KTD→AW	.026ns	0.376	.353	No	−0.083	0.139
H10b: CS × KTD→ATT	−.077ns	1.486	.069	No	−0.158	0.013
H10c: CS × ATT→AW	−.006ns	0.080	.468	No	−0.121	0.102
H10d: CS × ATT→EA	.054ns	0.767	.221	No	−0.056	0.177
H10e: CS × EA→SA	.033ns	0.452	.326	No	−0.080	0.161
H10f: CS × PE→SA	.101ns	1.031	.151	No	−0.038	0.275
H10g: CS × HE→SA	−.115ns	1.083	.139	No	−0.300	0.045

Note. ns = not significant; Cohen’s (1988) effect size threshold: Small (0.02), Medium (0.15), Large (0.35).

*

p < .05. **p < .01. ***p < .001.

In this model, knowledge shows a significant positive relationship with both awareness (β = .443, p < .001) and attitude (β = .423, p < .001). Attitude is positively associated with awareness (β = .172, p < .001) and negatively associated with ethical acceptability (β = –.172, p < .01). Ethical acceptability (β = .395, p < .001) and perceived enjoyment (β = .341, p < .001) are both significantly and positively associated with social acceptance. However, hedonic enjoyment does not significantly relate to social acceptance (β = .108, p > .05).

Attitude significantly mediates the relationship between knowledge and awareness. The indirect path yielded a t-value of 2.322 (p < .05). The direct effect of knowledge on awareness remains significant, indicating partial mediation (Hair et al., 2017).

The interaction between knowledge and FOMO on attitude is positive and significant (β = .140, p < .05), indicating that FOMO strengthens the effect of knowledge on attitude. Similarly, the interaction between attitude and FOMO on ethical acceptability is positive and significant (β = .169, p < .05), showing that FOMO moderates the negative effect of attitude on ethical acceptability.

Simple slope analyses (Figure 2) show that at high FOMO, greater deepfake knowledge predicts more positive verification attitude. At low FOMO, stronger verification attitude correspond to lower ethical acceptability, a relationship that diminishes at high FOMO. These results indicate that FOMO strengthens the positive effect of knowledge on attitude and attenuates the negative effect of verification attitude on ethical acceptability.

OPEN IN VIEWER

Figure 2.

The moderating effects of FOMO.

The structural model explains 54.1% of the variance in social acceptance (R² = .541), indicating moderate explanatory power (Hair et al., 2019). Cognitive, normative, hedonic, social-emotional, and personality factors collectively account for over half of students’ social acceptance of deepfake in educational contexts. Model fit indices yielded SRMR = 0.113, χ² = 1561.161, and NFI = 0.766. Although SRMR slightly exceeds 0.10 and NFI is below 0.90, variance-based PLS-SEM guidelines place greater emphasis on explanatory power over exact fit (Hair et al., 2022). Accordingly, the model can be regarded as adequate based on the R².

PLS-Predict assessed the model’s predictive ability for holdout samples (Shmueli et al., 2019). Social acceptance showed Q² = 0.247 (>0), and all indicators had Q² predict > 0, confirming predictive relevance (Table 7). RMSE comparisons with the linear model revealed minor deviations for some indicators (e.g., SA3 = +0.035; SA1 = +0.004; SA2 = −0.005), consistent with expectations in complex models. These small to moderate deviations do not compromise the evaluation of theoretical relationships (Shmueli et al., 2019). This is typical in the present study’s complex models integrating cognitive, normative, and hedonic constructs with moderating effects.

OPEN IN VIEWER

Table 7.

PLS-predict.

Construct	Q2 predict
SA	0.247
		RMSE
Indicators		PLS-SEM	LM	PLS-LM
SA1	0.198	0.978	0.974	0.004
SA2	0.207	0.970	0.975	−0.006
SA3	0.224	0.951	0.916	0.035

Additional Moderation Analyses

To assess robustness (Table A2), the study examined each moderator’s explanatory power (Jiang et al., 2021). Model 1 served as a baseline with seven direct relationships; Model 2 added FOMO, and Model 3 added conscientiousness. Bootstrapping in Model 3 showed that conscientiousness significantly and negatively moderated the knowledge–attitude relationship (β = −.108*, Bias-Corrected CI: −0.191, −0.024). Simple slope analyses (Figure 3) indicate the knowledge–attitude link is stronger at low and weaker at high conscientiousness. Including FOMO and conscientiousness as moderators increased explained variance in social acceptance by 3.64% (ΔR² = .019).

OPEN IN VIEWER

Figure 3.

Moderating effect of conscientiousness.

Cognitive (KAB Model), Normative Pathways, and FOMO

Recent cybersecurity research shows that the Knowledge-Attitude-Behavior (KAB) model effectively explains awareness and behavioral responses (Ahamed et al., 2024; Butavicius et al., 2020). Consistent with these findings, this study finds that deepfake knowledge (M = 3.691, SD = 0.652) significantly enhances both security awareness (M = 3.729, SD = 0.692, p < .001) and verification attitude (M = 3.888, SD = 0.907, p < .001). These results confirm that greater knowledge fosters critical verification attitude and heightens awareness of deepfake-related risks.

Among university students with high FOMO, greater deepfake knowledge significantly increases verification attitude, indicating that FOMO strengthens the knowledge–attitude link. This extends the KAB framework by showing that social-emotional factors moderate this relationship, with FOMO acting as a motivational force that enhances cognitive engagement with complex digital content (Kong et al., 2024). In the context of deepfake, students with higher FOMO demonstrate stronger verification attitude even when controlling for knowledge. These findings highlight the role of affective social pressures in amplifying knowledge effects, supporting calls to integrate emotional moderators into the KAB model (Hong et al., 2023) and aligning with evidence that social-emotional responses can drive decision making under high uncertainty (Loewenstein et al., 2001). Overall, KAB processes interact dynamically with individual social-emotional states in technology-mediated environments.

The study found a positive and significant relationship between attitude and awareness, consistent with Shahbaznezhad et al. (2021) and M. Alanazi et al. (2022). University students with stronger verification attitude demonstrate greater awareness of deepfake. Attitude mediates the relationship between knowledge and awareness, aligning with Ahamed et al. (2024), indicating that while knowledge enhances understanding, verification attitude determines how effectively it translates into heightened awareness of deepfake-related security risks.

The study found that attitude negatively associates with ethical acceptability, consistent with Ramachandran et al. (2023). University students with stronger verification attitude perceive deepfake as less ethically acceptable, indicating that critical evaluation heightens ethical scrutiny, especially regarding authenticity and manipulation. Similar patterns appear in emerging technologies like ChatGPT, where frequent use raises concerns about reduced cognitive engagement, influencing ethical judgments (Acosta-Enriquez et al., 2024).

The study found that FOMO positively moderated the relationship between verification attitude and ethical acceptability of deepfake. Among students with low FOMO, stronger verification attitude corresponded to lower ethical acceptability, but this negative relationship weakened as FOMO increased. From the KAB perspective, high FOMO heightens attentional vigilance, strengthening the translation of knowledge into verification attitude (Hai & Xiong, 2025; Przybylski et al., 2013). From GFT, ethical acceptability reflects a normative goal, and high-FOMO individuals adjust moral judgments to align with social consensus, weakening the effect of verification attitude on ethical evaluations (McKee et al., 2024). Thus, FOMO operates dually: enhancing cognitive alignment with verification knowledge while attenuating normative application to moral judgment, highlighting how social-emotional pressures shape knowledge–attitude–judgment processes in deepfake contexts.

Normative and Hedonic Pathways (Goal Framing Theories), Social Acceptance, and FOMO

The study found a positive relationship between normative goal frame (ethical acceptability) and social acceptance, consistent with Ari et al. (2024) and Taebi (2017). University students who perceived deepfake as ethically acceptable were more likely to express social acceptance, including supporting its use in coursework, applying it in studies, and engaging with deepfake-generated learning content.

However, unexpectedly, not all hedonic goals significantly influenced university students’ acceptance of deepfake. Perceived enjoyment (i.e., finding deepfake content fun) positively and significantly predicted social acceptance, whereas hedonic enjoyment (e.g., thrill, sense of adventure, or excitement) did not. This aligns with Ruiz et al. (2024) but contrasts with Marjerison et al. (2022). The difference may reflect contextual factors: deepfake is technically complex and ethically ambiguous (M. Li & Wan, 2023), which may reduce thrill-oriented hedonic responses. Hedonic motivations encompass multiple dimensions, including enjoyment, passing time, and behavioral intention (Brandtzaeg & Følstad, 2018), yet the novelty and ethical concerns of deepfake may dampen hedonic excitement, while perceived enjoyment represents an immediate psychological reaction (Cyr et al., 2009), making it a stronger predictor of educational acceptance (f² = 0.102 vs. f² = 0.009).

In summary, social acceptance depends more on accessible and relevant enjoyment than on thrill alone. Previous studies indicate that hedonic motives are less influential when ethical or functional factors are salient (Cyr et al., 2009; van der Heijden, 2004), consistent with this study’s finding that perceived enjoyment predicted acceptance, whereas thrill-oriented hedonic enjoyment did not. According to GFT, normative goals can override hedonic goals when ethical responsibility and social appropriateness are emphasized (Lindenberg & Steg, 2007). In educational contexts, university students may therefore prioritize ethical acceptability and responsible use of deepfake (f² = 0.171) over adventure or excitement associated with hedonic enjoyment.

Moderating Roles of FOMO and Conscientiousness in Cognitive, Normative, and Hedonic Pathways

In this study, conscientiousness and FOMO significantly moderated the knowledge–attitude relationship but not the knowledge–awareness or attitude–awareness links. Attitude is shaped by both cognitive and motivational–emotional processes (Ajzen, 2001). FOMO, as an emotional–motivational factor, influences evaluations through heightened sensitivity to social belonging and exclusion (Przybylski et al., 2013), explaining its effect on the knowledge–attitude link. By contrast, awareness reflects cognitive recognition of an issue (Hong et al., 2023) and is less dependent on affective or motivational states.

Conscientiousness, characterized by self-discipline, organization, and goal directed behavior (Costa & McCrae, 1992), relates more strongly to effortful behaviors and evaluative judgments than to perceptual or reflective processes. It predicts task performance and the translation of knowledge into deliberate action (Komarraju et al., 2011) and is more consistently linked to behavioral outcomes than to awareness or perception (McCrae & Costa, 2008). In contrast, awareness involves reflection, salience detection, and memory retrieval (Hong et al., 2023), which depend on cognitive internalization (Aikins, 2004) rather than goal directed effort (Petty & Cacioppo, 1986), and moral evaluations are shaped primarily by contextual and normative cues (Graham et al., 2009).

These findings are consistent with the present results. Both FOMO and conscientiousness moderated the knowledge–attitude relationship, highlighting their role in shaping evaluative processes. However, only FOMO influenced the attitude–ethical acceptability relationship, reflecting its affective and motivational basis, while conscientiousness did not significantly moderate awareness or ethical acceptability.

Conscientiousness emphasizes responsibility, persistence, and future-oriented planning (Bogg & Roberts, 2013; Wilmot & Ones, 2019), affecting behaviors that require sustained effort, whereas FOMO reflects sensitivity to social belonging and fear of exclusion (Przybylski et al., 2013), making it more relevant to evaluative judgments or decisions under social comparison. Trait activation theory suggests traits influence behavior only when situations align with their motivational relevance (Tett & Burnett, 2003). Because hedonic and normative motivations are immediate or socially guided, the effects of conscientiousness and FOMO are limited: hedonic motives involve spontaneous affective responses (Kawabata & Mallett, 2022), and normative motives are shaped by social norms and collective expectations. This aligns with evidence that conscientiousness predicts long-term planning and self-regulation (Chakraborty et al., 2023), while FOMO primarily shapes evaluative judgments rather than immediate or norm-driven responses (Elhai et al., 2020).

Both FOMO and conscientiousness were tested as potential moderators; however, conscientiousness did not show a significant effect when analyzed alongside FOMO. When examined independently, conscientiousness negatively moderated the positive relationship between deepfake knowledge and verification attitude, contrary to the hypothesized positive effect. Specifically, the association between knowledge and verification attitude was stronger for students with lower conscientiousness and weaker for those with higher conscientiousness, indicating that while knowledge generally enhances verification attitude, its impact diminishes as conscientiousness increases.

A possible explanation is that highly conscientious individuals approach complex or controversial information, such as deepfake, with caution and analytical rigor (Komarraju et al., 2011). Rather than directly translating knowledge into strong verification attitude, they may adopt a more measured or skeptical stance, especially when ethical guidelines or established procedures are ambiguous. Research indicates that conscientious individuals often avoid uncertain environments without structured frameworks (Soto & John, 2017b). In the context of deepfake, where technological and ethical boundaries remain fluid (M. Li & Wan, 2023), conscientious students may hesitate to convert knowledge into confident verification attitude.

Conscientiousness involves diligence, accuracy, and adherence to rules (Soto & John, 2017a). Research indicates that highly conscientious individuals are careful, avoid risks, and aim to maintain life satisfaction (Aumeboonsuke & Caplanova, 2021; Boyce et al., 2016). While these traits are advantageous in structured settings, they may prompt caution in ambiguous contexts. University students reported moderate concern about deepfake (Mean = 3.45) but low exposure (Mean = 2.70). Thus, even with substantial deepfake knowledge, conscientious students’ cautious and rule-oriented tendencies may limit the translation of knowledge into stronger verification attitude.

Theoretical Implications

As deepfake technology becomes increasingly prevalent in education and entertainment (Chadha et al., 2021; Gilbert & Gilbert, 2024; Roe et al., 2024), understanding university students’ perceptions and responses is crucial. This study applies the Knowledge-Attitude-Behavior (KAB) model and Goal Framing Theory (GFT) to examine students’ awareness and acceptance of deepfake in academic contexts. While the KAB model effectively explains behavior change in health education (Rimpeekool et al., 2016) and information security (Ahamed et al., 2024), prior research shows that knowledge alone rarely alters attitude or behavior, highlighting the role of additional motivational factors (B. H. Nguyen & Le, 2024). In the context of deepfake, this challenge is amplified by the technical complexity of creating and detecting manipulated media (Verdoliva, 2020) and ethical ambiguities regarding consent and misinformation (Chesney & Citron, 2019; Vaccari & Chadwick, 2020). Findings reveal that students’ verification attitude shapes their ethical acceptability judgments, which in turn influence social acceptance, thereby extending the KAB model and demonstrating the relevance of GFT in explaining how cognitive attitude and normative perceptions jointly determine acceptance of emerging technologies.

Drawing on GFT, this study highlights that behavior is guided by multiple simultaneously active goals, particularly normative and hedonic orientations. A normative goal frame emphasizes societal implications, with individuals feeling responsible for the ethical and social consequences of deepfake use (M. Li & Wan, 2023). In contrast, a hedonic goal frame prioritizes immediate enjoyment, leading to favorable attitude toward deepfake for amusement or creativity, even when risks are recognized (Gestoso & Bakayeva, 2024; W. Li & Zhao, 2024). Findings indicate that these motivational orientations strongly shape whether university students perceive deepfake as entertainment or an ethical concern, which in turn influences their social acceptance of the technology.

The results indicate that social pressures, specifically FOMO, moderate the relationship between university students’ deepfake knowledge and verification attitude. Higher FOMO strengthens this relationship, suggesting that social factors heighten critical engagement (Przybylski et al., 2013) with emerging deepfake technology (F. Y. Lee et al., 2025). Additionally, FOMO moderates the negative relationship between verification attitude and ethical acceptability: at low FOMO, stronger verification attitude leads students to view deepfake as less ethically acceptable, whereas at high FOMO this effect weakens. Theoretically, these findings extend the KAB model by showing that attitude does not uniformly translate into ethical evaluations but depend on social pressures. From the perspective of GFT, normative goals are more influential under low FOMO, while higher FOMO conditions dilute the weight of ethical considerations.

The findings extend existing models by showing that socially embedded motivations, such as FOMO, can recalibrate the influence of attitude on ethical acceptability, enhancing the explanatory power of both KAB and GFT perspectives. They highlight the importance of incorporating social factors into models of knowledge, attitude, and technology acceptance, as these factors shape how individuals interpret information and assess ethical implications in deepfake contexts (Table 8).

OPEN IN VIEWER

Table 8.

Summary of Significant Hypotheses and Findings, Theoretical Contributions, and Practical Implications for University Students’ Engagement with Deepfake Technology.

Hypotheses (significant findings)	Key findings	Theoretical implications	Practical recommendations
H1: Knowledge → Awareness	Higher knowledge increases deepfake security awareness	Highlights the KAB model’s role in guiding awareness of deepfake technology	Integrate deepfake content; promote critical thinking and fact-checking
H2: Knowledge → Attitude	Knowledge positively relates to deepfake verification attitude	Indicates that cognitive knowledge shapes critical attitude toward deepfake	Enhance digital literacy; include verification exercises
H3: Attitude → Awareness	Stronger verification attitude increases security awareness	Highlights reciprocal links between attitude and awareness in digital literacy	Use reflection exercises to reinforce awareness
H4: Attitude mediates Knowledge → Awareness	Attitude partially mediates knowledge’s effect on awareness	Demonstrates cognitive mechanisms connecting knowledge to awareness	Emphasize critical evaluation in teaching materials
H5: Attitude → Ethical acceptability	Stronger verification attitude decreases ethical acceptability	Indicates cognitive attitude drive normative evaluations	Integrate ethics content; encourage moral reflection
H6: Ethical acceptability → Social acceptance	Higher ethical acceptability promotes social acceptance in learning	Demonstrates the role of normative evaluations in deepfake acceptance	Guide responsible deepfake use; provide ethical frameworks
H7: Perceived enjoyment → Social acceptance	Enjoyment increases acceptance; hedonic thrill does not	Highlights hedonic influences on technology acceptance	Include educational and creative deepfake activities
H9b: FOMO moderates Knowledge → Attitude	Higher FOMO strengthens the knowledge- attitude link	Highlights social pressures can amplify cognitive influences	Leverage peer discussions and social influence in programs
H9d: FOMO moderates Attitude → Ethical acceptability	Higher FOMO weakens the effect of attitude on ethical acceptability	Extends the KAB model by showing attitude does not uniformly relate to normative judgments under social pressure.	Integrate peer dialogue, ethical reasoning, and emotional awareness exercises

Practical Implications

Deepfake technology presents both opportunities and challenges. It offers university students avenues for entertainment, social connection, and potential educational applications. However, despite awareness of associated risks, particularly misinformation, some students may still share deepfake content due to limited verification behaviors (Sharma et al., 2023). This underscores the importance of understanding how deepfake knowledge, verification attitude, and security awareness interact among university students, who are active social media users and future societal influencers.

This study demonstrates that deepfake knowledge positively relates to verification attitude and awareness, with verification attitude acting as a mediator. This has clear implications for education and media literacy. Enhancing students’ understanding of deepfake can improve their awareness of both its benefits and risks. Universities could implement strategies such as integrating deepfake case studies into media literacy curricula, designing exercises to reinforce verification behaviors, and providing opportunities for critical evaluation of manipulated content. These initiatives capitalize on the link between knowledge and verification attitude, equipping students to engage responsibly with digital content and navigate emerging technological challenges.

The study shows that among university students with higher FOMO, greater deepfake knowledge enhances verification attitude, indicating that social-emotional pressures can be leveraged positively when students are well-informed. Educational programs could incorporate collaborative exercises, peer discussions, or simulated social scenarios to mimic FOMO while emphasizing critical evaluation. Conscientiousness was found to negatively moderate the knowledge–attitude relationship, with students low in conscientiousness benefiting more from knowledge in forming positive verification attitude. This suggests personality influences how effectively knowledge translates into proactive behavior. To support all students, universities could provide clear ethical guidelines, step-by-step verification procedures, checklists for assessing credibility, and practical workshops to build verification skills. Combining knowledge-building, structured training, and consideration of social-emotional and personality factors can enhance digital literacy, critical evaluation, and responsible engagement with deepfake technology.

The study finds that verification attitude negatively associates with ethical acceptability of deepfake, indicating that students who critically evaluate digital content are less likely to view deepfake as ethically acceptable, especially in manipulative or deceptive contexts. This highlights the importance of promoting both technical literacy and ethical reflection in university education (Kumar, 2024). Integrating ethics-focused content into media and communication courses can help students assess the moral implications of deepfake and foster responsible, informed digital citizenship.

The study further shows that students with strong verification attitude and low FOMO are more likely to perceive deepfake as ethically unacceptable. This underscores the need for media literacy initiatives addressing both cognitive and social-emotional factors. Educational programs could combine critical verification training with strategies to manage FOMO, such as guided reflection on digital habits and prioritizing credible information sources. Peer discussion groups evaluating deepfake examples and scenario-based exercises simulating real-world exposure can enhance ethical decision-making. Policymakers and curriculum designers might also integrate digital ethics modules to foster responsible engagement with emerging technologies.

Given that ethical acceptability positively associates with social acceptance of deepfake in educational contexts, universities can implement targeted strategies to enhance ethical judgment and promote responsible use. Institutions may integrate structured ethical frameworks into coursework, helping students assess the potential impacts, risks, and benefits of deepfake technology. Framing deepfake as a tool for creative and academic applications, such as digital storytelling, media production, or historical simulations (Hendrickson, 2025; Lundberg & Mozelius, 2025), can foster both ethical awareness and responsible engagement. Additionally, providing clear ethical guidelines, facilitating peer discussions, and using scenario-based exercises that illustrate responsible use and potential misuse can strengthen students’ ethical reasoning. Linking ethical considerations to practical activities encourages thoughtful and constructive adoption of deepfake technology in academic work.

The positive association between perceived enjoyment and social acceptance indicates that students are more likely to embrace deepfake in academic contexts when the experience is engaging. Universities can capitalize on this by designing learning activities that highlight the educational and creative potential of deepfake, such as interactive projects, simulations, or digital storytelling exercises. To ensure enjoyment does not lead to mere entertainment or misuse, instructors should embed clear learning objectives, ethical guidelines, and reflective discussions. Framing activities around skill development, critical thinking, and creative application links enjoyment to purposeful learning, fostering acceptance grounded in both engagement and educational value.