When Algorithms Create Culture: An Integrative Model of Consumer Acceptance of AI-Generated Music ^†

Abstract

Background: The rapid advancement of generative artificial intelligence is transforming music composition from an exclusively human-centric activity into a hybrid human–algorithmic domain. Despite technological progress and growing commercial integration, consumer acceptance of AI-generated music remains empirically underexplored. Methods: This study formulates and empirically evaluates a multidimensional theoretical model integrating nine frameworks—including UTAUT2, parasocial interaction theory, anthropomorphism theory, authenticity theory, and innovation resistance theory—through a quantitative cross-sectional survey of 466 young adults aged 17–28. Confirmatory factor analysis and multiple regression analysis (with robust standard errors) were employed. Results: The model explained 63.6% of the variance in behavioral intention (R² = 0.636). Five constructs emerged as significant predictors: hedonic motivation (β = 0.136, p = 0.017), parasocial relationships (β = 0.121, p = 0.002), social influence (β = 0.126, p = 0.002), performance expectancy (β = 0.102, p = 0.019), and innovation resistance (β = −0.089, p = 0.029). Authenticity concerns, ethical AI concerns, anthropomorphic perceptions, and technological substitution fears were non-significant in the multivariate model. Conclusions: Young consumers’ acceptance of AI-generated music is primarily driven by experiential, social, and relational factors rather than ethico-cultural concerns. These findings have substantive implications for creative industries navigating algorithmic cultural production.

1. Introduction

Artificial Intelligence (AI) is no longer confined to automation, optimization, or predictive analytics. It has penetrated realms traditionally considered exclusively human—art, literature, and music. In recent years, generative AI systems have demonstrated the capability to compose symphonic pieces, create commercially viable pop songs, and replicate the voice characteristics of renowned musicians [1,2]. What was formerly considered science fiction has now materialized into a concrete market reality. An illustrative AI-generated composition in this idiom is provided in the Supplementary Materials (Audio S1).

The incorporation of AI into creative sectors represents one of the most significant shifts in modern cultural production. Creative AI refers to systems capable of generating original artistic outputs—text, images, and music—through machine learning frameworks such as deep neural networks and generative adversarial networks (GANs) [3,4]. In the realm of music, AI is now integrated throughout the entire value chain: composition, arrangement, production, recommendation algorithms, and predictive hit forecasting [5,6,7].

Yet, technological capability does not automatically translate into consumer acceptance. Although evidence indicates that listeners frequently find it difficult to differentiate between human and AI-generated music [8,9], research demonstrates a persistent labeling bias: music designated as AI-generated receives less favorable evaluations than that labeled as human-made [10,11]. This indicates that acceptance is not solely a function of quality but is also shaped by psychological and social factors.

The debate, therefore, extends beyond quality assessment. It encompasses: (1) authenticity and artistic legitimacy; (2) emotional attachment and parasocial bonds; (3) ethical transparency and intellectual property; (4) fear of professional displacement; and (5) resistance to algorithmic cultural homogenization.

While prior research has examined isolated determinants of AI acceptance—such as performance expectancy [12], trust and perceived humanness [11], or esthetic bias [9,10]—there remains a significant research gap. Existing studies rarely integrate psychological, sociological, ethical, and technological perspectives within a unified framework, particularly in the domain of music [13].

This paper is an extended and substantially revised version of a conference paper presented at HAICAI 2026 (Athens, 23–24 April 2026), published in the Lecture Notes in Networks and Systems series (Springer Nature). The current manuscript provides: (a) a full confirmatory factor analysis with discriminant validity testing; (b) comprehensive regression diagnostics, including heteroskedasticity tests and ordinal logistic robustness checks; (c) extended theoretical elaboration of the non-significant ethico-cultural constructs; and (d) substantially expanded Discussion and Conclusions sections with detailed implications for the creative industries.

This study offers three principal contributions. First, it expands the scope of technology acceptance research by exploring the symbolic and emotional dimensions of AI-generated cultural production, demonstrating that experiential and relational factors may outweigh normative and ethical considerations in shaping behavioral intention. Second, it empirically investigates the role of parasocial relationships with AI-generated artists, extending parasocial interaction theory beyond human media characters and providing quantitative evidence that relational attachment processes may operate independently of human ontology. Third, the findings challenge the dominant assumption that authenticity concerns, ethical apprehensions, and technological substitution fears constitute primary barriers to AI acceptance—at least among young consumers in digitally mediated cultural environments.

2. Theoretical Background

The adoption of AI-generated music cannot be adequately explained through a singular theoretical lens. Unlike conventional technological innovations, AI music simultaneously engages technological, cultural, psychological, and ethical dimensions. It is not merely a tool; it operates as a creative agent embedded within a historically human-centric artistic tradition. Comprehending consumer behavior, therefore, requires a multifaceted theoretical framework capable of capturing both facilitating factors and resistance dynamics.

This study synthesizes nine complementary theoretical frameworks: the Unified Theory of Acceptance and Use of Technology (UTAUT2) [12], Anthropomorphism Theory [14,15], Authenticity Theory [16], Parasocial Interaction Theory [17,18], Curiosity Theory [19], Algorithmic Culture Theory [20], Technological Substitution perspectives [21,22], AI Ethics Frameworks [23], and Innovation Resistance Theory [24]. Together, these perspectives enable conceptualization of AI-generated music not merely as a technical artifact but as a socio-cultural phenomenon emerging at the intersection of human creativity and algorithmic agency.

2.1. Anthropomorphism and Acceptance

Anthropomorphism theory posits that individuals perceive non-human phenomena through human cognitive schemas, ascribing agency, intentionality, and emotional capacity to technological systems [14]. Three sub-dimensions are operationalized: animacy (perceived dynamism and expressive energy), human-likeness (resemblance to human emotional expression), and perceived sociability (capacity for social communication). Empirical support is documented in related domains: Hong et al. [25] demonstrated that anthropomorphic traits of an AI music generator influenced its acceptance as a musician, while Bagratuni et al. [13] established that liveliness and human-like qualities significantly predict the appeal of AI-generated singing voices.

H1a. 

Animacy positively influences consumer acceptance of AI-generated music.

H1b. 

Human-likeness positively influences consumer acceptance of AI-generated music.

H1c. 

Perceived sociability positively influences consumer acceptance of AI-generated music.

2.2. UTAUT2 Determinants and Acceptance

The UTAUT2 model [12] offers a comprehensive framework for understanding adoption in consumer contexts. Performance expectancy captures perceived functional advantages; social influence reflects approval from referent groups; and hedonic motivation operationalizes the enjoyment and esthetic satisfaction derived from a technology. A systematic review of 650 UTAUT2-based studies confirmed the model’s robustness across diverse domains, identifying hedonic motivation as the most frequently employed and consistently significant construct in consumer technology research [26]. In the context of AI-generated music, the hedonic path was found to be significant in approximately 88% of reviewed studies [27].

H2a. 

Performance expectancy positively influences consumer acceptance of AI-generated music.

H2b. 

Social influence positively influences consumer acceptance of AI-generated music.

H2c. 

Hedonic motivation positively influences consumer acceptance of AI-generated music.

2.3. Curiosity as an Exploratory Driver

Curiosity theory posits that individuals possess an intrinsic motivation to investigate novel stimuli [19,28]. AI-generated music, as a technologically and culturally novel phenomenon, may activate curiosity even among skeptical listeners, functioning as a stimulus for early-stage adoption. This is consistent with Kashdan et al.’s [29] two-dimensional model of trait curiosity—stretching (the motivation to seek new knowledge) and embracing (willingness to accept novelty and uncertainty)—both of which are likely activated when consumers encounter AI-generated music for the first time. This mechanism aligns with broader evidence that positive anticipated emotions such as curiosity increase behavioral intentions toward novel innovations [30].

H3. 

Curiosity positively influences consumer acceptance of AI-generated music.

2.4. Authenticity Concerns as a Cultural Barrier

Authenticity occupies a pivotal role in cultural consumption [16,31]. Music is conventionally associated with individual expression and emotional sincerity, with authenticity functioning as a socially constructed evaluative criterion established during the listening experience [32]. Empirical evidence confirms authenticity-based bias against AI-attributed music: Tubadji et al. [33] documented quality degradation upon AI authorship disclosure, while Ansani et al. [11] reported lower likeability, emotional valence, and quality ratings for AI-attributed performances.

H4. 

Authenticity concerns negatively influence consumer acceptance of AI-generated music.

2.5. Parasocial Relationships and Emotional Bonding

Parasocial interaction theory posits that audiences form unilateral emotional bonds with media figures [17]. In musical contexts, the emergence of AI artists raises the question of whether emotional attachment can form with non-human creators. Stein et al. [34] demonstrated that parasocial relationships with digitally constructed non-human personas can be as robust as those with human figures; Lim and Lee [35] further showed that disclosure of artificial origin does not inherently diminish parasocial formation when emotional narrative strategies are employed.

H5. 

Parasocial relationships with AI artists positively influence consumer acceptance of AI-generated music.

2.6. Technological Substitution and Algorithmic Culture Concerns

In the music industry, AI presents a potentially direct substitutive threat, as creative output—rather than merely routine labor—becomes subject to automation [36]. Herington et al. [37] document widespread professional anxiety concerning job displacement among musicians. Algorithmic culture theory [20] further posits that computationally optimized cultural production may foster creative homogenization. Recent empirical analysis of AI music training datasets revealed a significant Western-centric bias, with approximately 86% of training data sourced from the Global North [38].

H6a. 

Job loss concerns negatively influence consumer acceptance of AI-generated music.

H6b. 

Creative uniformity concerns negatively influence consumer acceptance of AI-generated music.

2.7. Ethical Concerns and Innovation Resistance

AI ethics frameworks prioritize transparency, fairness, and accountability as foundational principles [23,39]. In AI-generated music, ethical concerns encompass undisclosed AI authorship, opaque training data, and creator compensation. Empirical research in AI-mediated markets demonstrates that ethical attitudes significantly shape trust and acceptance [40]. Innovation Resistance Theory [24] further explains principled resistance to innovations perceived as threatening established values. Longoni et al. [41] demonstrate that consumers resist AI providers even when objectively superior due to uniqueness neglect; Ansani et al. [11] provide experimental evidence of persistent AI composer bias unaffected by objective music quality.

H7a. 

Perceived ethical violations negatively influence consumer acceptance of AI-generated music.

H7b. 

Innovation resistance negatively influences consumer acceptance of AI-generated music.

3. Research Model and Hypotheses

The research model conceptualizes Behavioral Intention (BI) as the central endogenous construct, operationalized as the individual’s conscious willingness to listen to, recommend, and integrate AI-generated music into everyday consumption patterns. BI is predicted by both enabling mechanisms (hedonic motivation, social influence, parasocial relationships, performance expectancy) and inhibitory forces (innovation resistance), consistent with UTAUT2 theory [12] and the behavioral intention–acceptance relationship [42].

Table 1. Summary of research hypotheses.

H	Theoretical Framework	Predictor	Dir.	Theory Source	Outcome
H1a	Anthropomorphism	Animacy	(+)	[14]	Consumer Acceptance (BI)
H1b	Anthropomorphism	Human-Likeness	(+)	[14]	Consumer Acceptance (BI)
H1c	Anthropomorphism	Perceived Sociability	(+)	[14]	Consumer Acceptance (BI)
H2a	UTAUT2	Performance Expectancy	(+)	[12]	Consumer Acceptance (BI)
H2b	UTAUT2	Social Influence	(+)	[12]	Consumer Acceptance (BI)
H2c	UTAUT2	Hedonic Motivation	(+)	[12]	Consumer Acceptance (BI)
H3	Curiosity Theory	Curiosity	(+)	[19]	Consumer Acceptance (BI)
H4	Authenticity Theory	Authenticity Concerns	(−)	[16]	Consumer Acceptance (BI)
H5	Parasocial Interaction	Parasocial Relationships	(+)	[17]	Consumer Acceptance (BI)
H6a	Tech. Substitution	Job Loss Concerns	(−)	[21]	Consumer Acceptance (BI)
H6b	Algorithmic Culture	Creative Uniformity Concerns	(−)	[20]	Consumer Acceptance (BI)
H7a	AI Ethics	Perceived Ethical Violations	(−)	[23]	Consumer Acceptance (BI)
H7b	Innovation Resistance	Innovation Resistance	(−)	[24]	Consumer Acceptance (BI)

Note. BI = Behavioral Intention. (+) = positively hypothesized; (−) = negatively hypothesized. Conceptual model figure is available from the corresponding author upon request.

summarizes the full set of research hypotheses derived from the theoretical framework.

4. Materials and Methods

4.1. Research Design

This study employs a quantitative, theory-testing, cross-sectional research design with a post-positivist epistemological orientation. Data were collected at a single point in time using a structured self-administered questionnaire. The methodological approach is explanatory rather than descriptive; the focal objective is evaluating hypothesized causal linkages within a theoretically grounded structural model. The research process comprised two sequential phases: (1) pilot testing for instrument validation and refinement, and (2) primary data collection for hypothesis testing via multiple regression analysis.

4.2. Sampling Strategy and Participants

The target population comprised young adult consumers aged 17–28 who actively use digital music services. This age group was selected on theoretical and empirical grounds: young adults constitute the primary early adopters of digital cultural innovations, exhibit the highest levels of streaming platform engagement, and represent the demographic most likely to encounter and evaluate AI-generated music in naturalistic listening environments [26]. Having matured within algorithmically mediated media contexts, they possess the experiential familiarity necessary for a valid assessment of AI-generated cultural content.

A non-probability convenience sampling method was employed. Although probability sampling would strengthen external validity, the exploratory and emergent nature of AI-generated music research warrants this approach in initial theory-development contexts [43]. Participants were recruited through online platforms, including social media and academic communities, enabling broad geographic reach within the target age range. The sample was intentionally restricted to young adults aged 17–28, reflecting the demographic most actively engaged with AI-generated content and digital music platforms. This design choice maximizes internal homogeneity and ensures that observed variation in acceptance reflects attitudinal rather than generational differences. The methodological implication is that findings should be interpreted as specific to this cohort; their applicability to older consumers, who may relate differently to algorithmic mediation and AI authorship, cannot be assumed without further cross-age replication.

4.3. Instrument Development and Pilot Testing

All constructs were measured using multi-item Likert-type scales adapted from previously validated instruments. Prior to primary data collection, a pilot study with 33 participants assessed the clarity, reliability, and internal consistency of the measurement instrument. Cronbach’s alpha coefficients ranged from 0.765 to 0.909 across constructs, exceeding the conventional threshold of 0.70 [44]. Item-level analysis identified conceptual redundancy in one Behavioral Intention item (“I would accept AI-generated hit songs in the charts”), which was removed to improve discriminant clarity. The survey did not restrict participants to a specific music genre, as the objective was to assess general acceptance dispositions toward AI-generated music rather than genre-specific evaluations. Participants reported their primary genre preferences via an open-ended item. Responses revealed a heterogeneous sample: the most frequently reported preferences were Greek popular and art music (laika/entechna; 22.7%), Pop (11.4%), Hip-hop/Rap/R&B (10.7%), and Rock/Metal/Alternative (7.9%), with the remainder distributed across Electronic/EDM, K-pop/J-pop, Classical, Jazz/Soul, and mixed genre profiles. This diversity reflects the naturalistic listening habits of the target demographic and supports the ecological validity of the instrument across a broad stylistic spectrum. Genre-specific variation in AI music acceptance was not modeled in the present study; given that the technical complexity of AI-generated content varies substantially across genres, this constitutes a relevant direction for future research.

4.4. Measurement Model: Confirmatory Factor Analysis

A confirmatory factor analysis (CFA) was conducted using maximum likelihood estimation to assess the measurement model. The multi-construct model demonstrated good fit: χ²(624) = 1232.005, p < 0.001, CFI = 0.953, TLI = 0.945, RMSEA = 0.046 (90% CI [0.042, 0.050]), SRMR = 0.052. All standardized factor loadings were statistically significant (p < 0.001) and ranged from 0.54 to 0.95. Composite reliability (CR) exceeded 0.70 for all constructs. Average variance extracted (AVE) exceeded 0.50 for all constructs except Authenticity (AVE = 0.49), which was retained on the basis of three converging criteria: composite reliability exceeded the recommended threshold (CR = 0.76 > 0.70), all factor loadings were statistically significant (p < 0.001), and the AVE shortfall of 0.01 is within the range deemed acceptable in the literature when CR is satisfactory [45]. This decision is consistent with established practice in measurement model evaluation.

Discriminant validity was assessed via the Fornell–Larcker criterion and the heterotrait–monotrait (HTMT) ratio. Human-likeness and Perceived Sociability exhibited an HTMT of 0.897, marginally below the 0.90 criterion. A competing single-factor model merging these constructs demonstrated significantly inferior fit (Δχ²(13) = 22.85, p = 0.043), supporting their empirical distinctiveness. Common method bias was evaluated using a Harman single-factor test; the single-factor model fit substantially worse than the proposed model (CFI and TLI below acceptable thresholds; RMSEA exceeding recommended limits), indicating that common method variance does not constitute a critical threat to validity.

4.5. Ethics Statement

This study was conducted in accordance with the Declaration of Helsinki and applicable national ethical guidelines for non-interventional survey research. All participants provided informed consent prior to their participation. The survey was anonymous and voluntary. No sensitive personal data were collected. In Greece, non-interventional survey studies of this type conducted within academic contexts do not require formal institutional ethics board approval; the study was conducted in compliance with the ethical standards of the institutions of all co-authors (University of West Attica and European University of Cyprus).

5. Results

5.1. Sample Characteristics and Descriptive Statistics

The final sample consisted of 466 valid responses. Cronbach’s alpha coefficients for the primary sample ranged from 0.706 to 0.926 across constructs, confirming satisfactory internal consistency. The dependent variable, Behavioral Intention (BI), had a mean of 5.807 (SD = 2.768) on a composite scale ranging from 2 to 14 (median = 6). Distribution diagnostics indicated near-symmetry (skewness = 0.117) and slight platykurtosis (kurtosis = −0.754), with no evidence of floor or ceiling effects.

Among the predictor constructs, Authenticity Concerns (M = 15.659), AI Ethics Concerns (M = 16.961), and Innovation Resistance (M = 16.127) recorded notably high mean scores, indicating that participants expressed substantial concerns in these domains. Parasocial Relationships (M = 8.983) and Hedonic Motivation (M = 8.215) showed moderate levels. Full descriptive statistics are presented in

Table 2. Descriptive statistics for study variables (N = 466).

Variable	N	Mean	SD	Min	Max	Skew.	α
Behavioral Intention (BI)	466	5.807	2.768	2	14	0.117	0.706
Animacy	466	9.393	3.890	3	21	—	0.91
Human-likeness	466	6.380	2.617	2	14	—	0.73
Perceived Sociability	466	6.388	2.585	2	14	—	0.75
Performance Expectancy	466	9.034	3.758	3	21	—	0.82
Social Influence	466	9.845	3.789	3	21	—	0.82
Hedonic Motivation	466	8.215	4.025	3	21	—	0.93
Curiosity	466	6.906	2.914	2	14	—	0.82
Authenticity Concerns	466	15.659	3.850	3	21	—	0.76
Parasocial Relationships	466	8.983	4.789	4	28	—	0.92
Technological Substitution	466	13.676	4.665	3	21	—	0.88
Algorithmic Culture	466	13.740	4.191	3	21	—	0.82
AI Ethics Concerns	466	16.961	3.936	3	21	—	0.85
Innovation Resistance	466	16.127	4.027	3	21	—	0.88

Note. SD = standard deviation; Skew. = skewness (reported for BI only); α = Cronbach’s alpha.

.

5.2. Measurement Model Results

The CFA results confirmed the adequacy of the measurement model.

Table 3. CFA fit indices.

Index	Value
χ2	1232.005
df	624
p	<0.001
CFI	0.953
TLI/NNFI	0.945
RMSEA	0.046
RMSEA 90% CI	[0.042, 0.050]
SRMR	0.052
GFI	0.882
AGFI	0.853

presents the key fit indices, demonstrating acceptable model fit across all evaluated criteria. Standardized factor loadings, composite reliability (CR), and average variance extracted (AVE) for all constructs are reported in

Table 4. Standardized factor loadings, composite reliability (CR), and average variance extracted (AVE).

Construct	Items (Std. Loadings)	CR	AVE
Animacy	Qq4 = 0.717, Qq5 = 0.937, Qq6 = 0.942	0.91	0.76
Human-likeness	Qq7 = 0.668, Qq8 = 0.774	0.73	0.52
Perceived Sociability	Qq9 = 0.796, Qq10 = 0.748	0.75	0.60
Performance Expectancy	Qq11 = 0.786, Qq12 = 0.661, Qq13 = 0.842	0.82	0.60
Social Influence	Qq14 = 0.751, Qq15 = 0.745, Qq16 = 0.800	0.82	0.59
Hedonic Motivation	Qq17 = 0.899, Qq18 = 0.909, Qq19 = 0.887	0.93	0.81
Curiosity	Qq20 = 0.766, Qq21 = 0.918	0.82	0.71
Authenticity	Qq22 = 0.550, Qq23 = 0.756, Qq24 = 0.770	0.76	0.49 *
Parasocial Relationships	Qq25 = 0.816, Qq26 = 0.873, Qq27 = 0.890, Qq28 = 0.891	0.92	0.76
Technological Substitution	Qq29 = 0.904, Qq30 = 0.902, Qq31 = 0.699	0.88	0.68
Algorithmic Culture	Qq32 = 0.479 †, Qq33 = 0.873, Qq34 = 0.835	0.82	0.56
AI Ethics	Qq35 = 0.795, Qq36 = 0.891, Qq37 = 0.717	0.85	0.64
Innovation Resistance	Qq38 = 0.863, Qq39 = 0.813, Qq40 = 0.854	0.88	0.71

Note. CR = composite reliability; AVE = average variance extracted. * AVE marginally below the 0.50 threshold; retained on the basis of satisfactory composite reliability (CR = 0.76 > 0.70), statistically significant factor loadings (p < 0.001), and an AVE shortfall (0.01) within the range deemed acceptable in the literature when CR is adequate [45]. ^† Item retained to preserve content validity; removal did not improve CR or AVE.

.

5.3. Regression Diagnostics

Prior to hypothesis testing, a comprehensive set of OLS regression diagnostics was performed. Multicollinearity was assessed using Variance Inflation Factors (VIF); all VIF values ranged from 1.134 to 3.748, well below the threshold of 10 [43], with tolerance values uniformly above 0.267. Heteroskedasticity was evaluated through White’s test (F(20,445) = 1.56, p = 0.059); the non-significant result supports the homoskedasticity assumption. As an additional robustness check, the model was re-estimated using heteroskedasticity-consistent (robust) standard errors; results were substantively identical. Residual normality was confirmed through distribution diagnostics, with the large sample size (N = 466) further mitigating concerns about minor departures [46].

A robustness check using ordinal logistic regression (cumulative logit model) confirmed full convergence with the OLS findings: AIC = 1589.26 vs. intercept-only AIC = 2028.87; χ²(20) = 479.60, p < 0.001; concordance c = 0.846. The identical pattern of significant and non-significant predictors across both estimation strategies strengthens confidence in the reported results.

5.4. Bivariate Analysis

Spearman correlation analysis revealed that Behavioral Intention was positively and significantly associated with Animacy (ρ = 0.598), Human-likeness (ρ = 0.507), Perceived Sociability (ρ = 0.508), Performance Expectancy (ρ = 0.668), Social Influence (ρ = 0.617), Hedonic Motivation (ρ = 0.729), Curiosity (ρ = 0.558), and Parasocial Relationships (ρ = 0.668; all p < 0.001). Negative associations were observed for Authenticity (ρ = −0.473), AI Ethics (ρ = −0.254), and Innovation Resistance (ρ = −0.584; all p < 0.001). Technological Substitution (ρ = −0.040, p = 0.392) and Algorithmic Culture (ρ = −0.059, p = 0.204) were not significantly related to Behavioral Intention. Full results are reported in

Table 5. Spearman correlations with Behavioral Intention (N = 466).

Predictor	Spearman ρ	p-Value
Animacy	0.598	<0.001
Human-likeness	0.507	<0.001
Perceived Sociability	0.508	<0.001
Performance Expectancy	0.668	<0.001
Social Influence	0.617	<0.001
Hedonic Motivation	0.729	<0.001
Curiosity	0.558	<0.001
Authenticity	−0.473	<0.001
Parasocial Relationships	0.668	<0.001
Technological Substitution	−0.040	0.392
Algorithmic Culture	−0.059	0.204
AI Ethics	−0.254	<0.001
Innovation Resistance	−0.584	<0.001

.

5.5. Multiple Regression Analysis and Hypothesis Testing

The main hypotheses were tested using multiple linear regression with Behavioral Intention as the dependent variable (N = 466), employing heteroskedasticity-consistent (robust) standard errors. The overall model was highly significant (F(20,465) = 63.46, p < 0.001) and explained 63.6% of the variance in Behavioral Intention (R² = 0.636, Adjusted R² = 0.620). Full results are reported in

Table 6. Multiple regression results predicting Behavioral Intention (N = 466).

Predictor	β	Std. Error	t	p-Value	VIF	Hypothesis
Animacy	0.061	0.045	1.35	0.178	2.88	H1a—Not supported
Human-likeness	−0.011	0.054	−0.20	0.844	2.52	H1b—Not supported
Perceived Sociability	−0.005	0.053	−0.09	0.928	2.30	H1c—Not supported
Performance Expectancy	0.102	0.043	2.36	0.019	3.29	H2a—Supported **
Social Influence	0.126	0.039	3.19	0.002	2.20	H2b—Supported ***
Hedonic Motivation	0.136	0.056	2.40	0.017	3.75	H2c—Supported **
Curiosity	0.077	0.044	1.74	0.082	2.02	H3—Marginal †
Authenticity	−0.010	0.036	−0.28	0.778	2.32	H4—Not supported
Parasocial Relationships	0.121	0.038	3.19	0.002	2.16	H5—Supported ***
Tech. Substitution	0.010	0.027	0.36	0.715	2.10	H6a—Not supported
Algorithmic Culture	−0.016	0.030	−0.55	0.585	2.13	H6b—Not supported
AI Ethics	0.044	0.036	1.23	0.219	2.20	H7a—Not supported
Innovation Resistance	−0.089	0.041	−2.19	0.029	3.02	H7b—Supported **
Gender	−0.098	0.185	−0.53	0.595	1.20	Control
Age	0.007	0.034	0.20	0.844	1.64	Control
Education	0.137	0.151	0.91	0.365	1.43	Control
Daily Listening	−0.079	0.093	−0.84	0.400	1.13	Control
Music Chart Engagement	0.082	0.114	0.72	0.473	1.13	Control
Prior AI Music Listening	0.201	0.177	1.13	0.258	1.13	Control
AI Recognition Ability	0.019	0.089	0.21	0.835	1.29	Control

Note. R² = 0.636, Adjusted R² = 0.620, F(20,465) = 63.46, p < 0.001. Robust standard errors reported. ^† p < 0.10; ** p < 0.05; *** p < 0.01.

. A summary of hypothesis testing outcomes is presented in

Table 7. Summary of hypothesis testing results.

H	Construct	Direction	β	p	Supported?
H1a	Anthropomorphism—Animacy	(+)	0.061	0.178	No
H1b	Anthropomorphism—Human-likeness	(+)	−0.011	0.844	No
H1c	Anthropomorphism—Sociability	(+)	−0.005	0.928	No
H2a	UTAUT2—Performance Expectancy	(+)	0.102	0.019	Yes **
H2b	UTAUT2—Social Influence	(+)	0.126	0.002	Yes ***
H2c	UTAUT2—Hedonic Motivation	(+)	0.136	0.017	Yes **
H3	Curiosity	(+)	0.077	0.082	Marginal †
H4	Authenticity Concerns	(−)	−0.010	0.778	No
H5	Parasocial Relationships	(+)	0.121	0.002	Yes ***
H6a	Job Loss Concerns	(−)	0.010	0.715	No
H6b	Creative Uniformity Concerns	(−)	−0.016	0.585	No
H7a	AI Ethics Concerns	(−)	0.044	0.219	No
H7b	Innovation Resistance	(−)	−0.089	0.029	Yes **

Note. ^† p < 0.10; ** p < 0.05; *** p < 0.01. Marginal support = significant at 10% but not 5% threshold.

.

6. Discussion

The empirical results reveal a theoretically coherent pattern in which experiential, social, and relational factors dominate the acceptance of AI-generated music among young consumers, while ethico-cultural concerns—despite commanding high mean scores—fail to achieve statistical significance in the multivariate model. This finding does not invalidate ethico-cultural critique; rather, it evidences a well-documented divergence between attitudinal concern and behavioral consequence that has been documented in the privacy literature [47] and may reflect the normalizing effects of prolonged exposure to algorithmically mediated content.

6.1. Hedonic Motivation and the Affective Marketplace

Hedonic Motivation emerged as the strongest predictor of Behavioral Intention (β = 0.136, p = 0.017) and, in the univariate model, accounted for 51.7% of variance—the highest proportion among all predictors. This finding is consistent with the characterization of music as an experience commodity whose primary value lies in emotional engagement and esthetic gratification [12,27]. Young consumers appear to evaluate AI-generated music primarily on the basis of its emotional impact rather than the identity or ontological status of its creator. This constitutes a practically significant finding for the creative industries: optimizing the emotional quality of AI-generated music may be more effective in driving adoption than emphasizing algorithmic transparency.

6.2. Social and Relational Dynamics

Social Influence (β = 0.126, p = 0.002) and Parasocial Relationships (β = 0.121, p = 0.002) achieved identical levels of statistical significance with comparable effect sizes, reflecting complementary dimensions of a broader social-relational framework. Social influence confirms UTAUT2 predictions [12] and aligns with research emphasizing social identity’s role in music consumption [48]. The Parasocial Relationships finding is particularly noteworthy: it extends parasocial interaction theory [17,18] by demonstrating that unilateral emotional bonds can form with non-human creators, supporting Stein et al. [34] evidence on virtual influencers and Lim and Lee [35] work on AI origin disclosure and parasocial engagement. That relational attachment appears to operate independently of human ontology represents a conceptually significant extension of parasocial theory into the domain of algorithmic cultural production.

6.3. Performance Expectancy and the Sufficiency Threshold

Performance Expectancy, while statistically significant (β = 0.102, p = 0.019), exhibits the smallest effect size among the four significant predictors. This pattern suggests that technical competence functions as a necessary but insufficient condition for acceptance: AI-generated music must be perceived as technically adequate, yet this perception alone does not account for the richness of acceptance variance. The affective and relational dimensions provide substantially greater explanatory power—a conclusion aligned with evidence that esthetic judgments of AI-generated music are substantially influenced by factors extrinsic to sound quality itself [10,11].

6.4. Innovation Resistance as Friction, Not Barrier

Innovation Resistance (β = −0.089, p = 0.029) constitutes a statistically significant inhibitory force with a modest effect size, suggesting it operates as friction rather than an absolute barrier. This pattern is consistent with Ram and Sheth’s [24] original conceptualization of resistance as principled opposition to value-threatening innovations, but indicates that among young consumers, this resistance is manageable and likely to diminish through familiarity, positive experience, and social normalization.

6.5. The Non-Significance of Ethico-Cultural Concerns

Authenticity Concerns (p = 0.778), Technological Substitution (p = 0.715), Algorithmic Culture concerns (p = 0.585), and AI Ethics concerns (p = 0.219) were non-significant in the multivariate model. This does not indicate that these dimensions are irrelevant to AI discourse; the high mean scores for these constructs (M = 15.66, 16.96, and 16.13, respectively) confirm their attitudinal salience. The findings instead suggest that in routine consumption contexts, immediate experiential gratification may override abstract normative reasoning—a pattern consistent with the privacy paradox literature [47]. This dissociation of ethical judgment from consumption behavior is further consistent with research on moral decoupling, whereby consumers psychologically separate assessments of moral conduct from assessments of performance in order to support options they find ethically contestable [49].

6.6. Anthropomorphism as Potential Mediator

All three anthropomorphic dimensions (Animacy, Human-likeness, and Perceived Sociability) demonstrated strong bivariate associations with Behavioral Intention (all p < 0.001) but lost significance in the multivariate model, likely due to collinearity with Hedonic Motivation and Parasocial Relationships. This pattern suggests that anthropomorphic perceptions may function as mediators rather than independent direct predictors of acceptance. Future research employing structural equation modeling could profitably examine these mediating pathways, contributing to a more fine-grained understanding of how anthropomorphic cues shape acceptance through affective and relational mechanisms.

7. Conclusions

This study formulated and empirically evaluated a multidimensional integrative model of consumer acceptance of AI-generated music among young adults (N = 466), synthesizing nine theoretical frameworks. The multiple regression model explained 63.6% of the variance in Behavioral Intention (R² = 0.636), indicating strong explanatory power. Four constructs were confirmed as significant positive predictors—Hedonic Motivation, Social Influence, Parasocial Relationships, and Performance Expectancy—while Innovation Resistance emerged as a significant inhibitory force. Ethico-cultural concerns, despite high attitudinal salience, did not significantly predict behavioral intention in the multivariate context.

This empirical pattern describes a theoretically coherent acceptance structure: affective and social dynamics prevail, while normative concerns recede, suggesting that AI-generated music is experienced as an affective artifact integrated into youth cultural practice rather than primarily as a technological or ethical disruption. The convergence of social and relational drivers implies that commercial strategies focused on emotional experience quality, social normalization via influencers, and the cultivation of AI artist identities with narrative coherence are likely to be most effective in driving adoption.

Three principal theoretical contributions are advanced. First, UTAUT2 is extended into a domain of cultural consumption governed by emotional and symbolic values, demonstrating the continued relevance of functional predictors within an affective marketplace. Second, the empirical validation of parasocial attachment to non-human artistic creators constitutes one of the initial quantitative confirmations that relational processes in digital culture may be decoupling from human ontology. Third, the documented non-significance of ethico-cultural concerns challenges established assumptions in the literature and reframes the central research question: rather than asking whether AI can produce authentic art, future inquiry may more productively examine how affective, social, and relational dynamics are being restructured around non-human creators.

The study’s limitations include the age-restricted convenience sample (17–28), common method bias [50] inherent to the cross-sectional self-report design, the absence of genre-specific controls, and the inability to establish causal ordering. The exclusive focus on young adults constrains inference to other demographic cohorts: older consumers may exhibit substantially different acceptance patterns, potentially governed by esthetic conservatism, professional identity concerns, or lower engagement with algorithmic platforms. The significant predictors identified here may attenuate or be displaced by other constructs in non-youth populations, making cross-age replication a priority. Future studies should employ longitudinal designs, experimental labeling conditions (disclosed vs. undisclosed AI authorship), and broader demographic sampling. Research employing structural equation modeling would allow examination of the mediating role of anthropomorphic perceptions and other indirect pathways that the present direct-effects framework was not designed to capture. The full survey instrument is provided in Appendix A.