Эволюция и влияние цифрового маркетинга: многометодологический анализ тенденций, сотрудничества и научных достижений

DOI:

Artificial intelligence (AI), content curation, and data collecting have positioned marketing as the central element of customer engagement in the digital era. This development indicates that social media tendencies such as "de-influencing" are at odds with hyper-personalization and consumer privacy issues. Researchers must traverse an extensive body of literature and use ethical, multi-method approaches to investigate technology, trust, and behavior. The paper establishes a foundation for digital marketing research by detailing the methodology and modeling the effects of AI customization on customer trust and purchasing intentions amid privacy concerns. Ethical and creative research enhances comprehension of the algorithmic consumer landscape.

Modeling (SmartPLS) to evaluate quantitative relationships with a comprehensive ethnographic analysis of de-influencing content on TikTok and Instagram with NVivo. This research aims to address this problem by further exploring the brandconsumer relationship in the context of algorithmic mediation and public opposition to regulated consumption beyond mere engagement metrics.

HISTORICAL EVOLUTION AND SCHOLARLY COLLABORATION

The major conceptual and empirical connections section analyzes AI-driven personalization and digital marketing. Machine learning algorithms personalize content and ads based on user data. Customization boosts engagement and conversion, but if consumers regard it as manipulative, it may fail [25]. Although AI-mediated interactions receive less research than human-mediated ones, transparency, security, competence, integrity, and compassion influence customer trust in digital settings. The Internet Users' Information Privacy Concern (IUIPC) scale suggests that privacy concerns may limit customization, particularly in individualistic nations, and moderate it [26]. In light of "de-influencing," where social media influencers argue against consumption, marketing assumptions, algorithmic customization, and cultural resistance must be examined [27]. Digital marketing research has moved from descriptive and correlational studies to mixed methodologies that include quantitative and qualitative insights, but integrating techniques is difficult [28]. Confidence in non-human agents, cultural trends like de-influencing on customization, computational social science and interpretive consumer research partnerships, and algorithmic adaptability's temporal development need more study [29]. The conceptual framework includes consumer trust, purchase intention, AI personalization, and privacy concern. It implies that AI personalization increases customer trust (H1) and purchase intention (H2), with consumer trust mediating (H3). Privacy problems may affect personalization and trust (H4) and intention (H5). This research improves AI marketing using the Technology Acceptance Model, Privacy Calculus, and Stimulus-Organism-Response theories. It contextualizes privacy concerns within cultural trends like de-influencing, explains how a mixed-methods approach might improve scientific rigor, and advises marketers on privacy-conscious tailoring.

METHODOLOGY

Selecting the right methodology is critical. Here are eight key methods applicable to digital marketing research:

• •    Quantitative      Surveys:        Large-scale

questionnaires to measure attitudes, usage, and behaviors (e.g., survey on ad recall from YouTube vs. Instagram).

• •    Experimental Design: A/B testing or controlled experiments to establish causality (e.g., testing two email subject lines on conversion rates).

• •    Web & Social Media Analytics: Using platform data (Google Analytics, Facebook Insights) for observational studies on user behavior.

• •    Content Analysis:   Systematic coding of

qualitative data from social media posts, reviews, or advertisements (often using NVivo).

• •    Ethnography/Netnography:         Immersive

observation of online communities to understand cultural and social contexts.

• •    Machine Learning & Predictive Modeling: Using algorithms to forecast trends, segment customers, or optimize campaigns (using Python/R).

• •    Structural Equation Modeling (SEM): Testing complex relationships between latent variables (e.g., brand image, trust, purchase intent) using tools like SmartPLS or AMOS.

• •    Meta-Analysis: Statically combining results from previous studies to identify overall effect sizes and trends.

Research Design

This sequential mixed-methods research investigates the impact of AI-driven customization on customer trust and purchase intention. It addresses the moderating impact of privacy concerns on these consequences. The methodology guarantees rigor, replicability, and triangulation via a structured framework including both quantitative and qualitative approaches. Phase 1 involves a comprehensive survey that quantifies anticipated connections, while Phase 2 entails a thorough thematic analysis of open-ended responses that contextualizes and enhances the quantitative results. This design employs a two-phase explanatory sequential approach. Quantitative findings may impact qualitative research due to the integration point, whereas qualitative data may clarify statistical trends. This approach aligns with the paper's multi-methodological paradigm, using both variance-based and interpretative methodologies to elucidate the subject matter.

Population and Sampling

The subjects of this study will be adult internet users, namely those aged 18 to 65, who have previously engaged with personalized marketing, including targeted advertisements and recommendation systems. In light of the privacy requirements established by GDPR and CCPA, the primary geographic focus is on North America and Europe. The G*Power analysis for Structural Equation Modeling (SEM) determined a sample size of 500 individuals. The parameters used were a medium effect size, an alpha level of 0.05, and a power of 0.95. Qualtrics panels will use a stratified random sampling method to guarantee diversity in key demographic attributes such as age, gender, digital literacy, and privacy sensitivity; therefore, they will produce a representative sample. Eligibility for participation requires the individual to be an active user of social media on at least three platforms, to have experienced AI-driven customization within the last 30 days, and to consent to participation in an academic study. Quota sampling will be used to ensure sufficient representation from each prescreened age category (18–25, 26–40, 41–65) and from individuals with varying degrees of privacy concern (low, medium, high). By using this allencompassing method, we hope to make the study's results more reliable and useful.

Instrument Development and Measures

All constructs were measured using validated multiitem scales adapted to the digital marketing context

Table 1. Construct Measurement

Construct	Scale Items	Sample Item	Source Adaptation
AI Personalization	5 items	"The AI recommendations I receive are tailored to my preferences."	Xu et al. (2021)
Consumer Trust	4 items	"I trust brands that use AI to personalize my experience."	McKnight et al. (2002)
Purchase Intention	3 items	"I am likely to purchase products recommended by AI systems."	Dodds et al. (1991)
Privacy Concern	6 items	"I am concerned about how my data is used for personalization."	Smith et al. (1996) – IUIPC scale

Data Collection Procedure

For this study, respondents should spend 12–15 minutes on Qualtrics XM completing the online survey. The survey includes intrinsic attention checks and consistency questions for reliable results. We will protect participants' identity and promote ethics by gaining informed consent. The study does not store any personally identifying information, in compliance with GDPR and CCPA regulations. We added a qualitative aspect to the quantitative poll to better understand the public's attitude toward data utilization. "Kindly articulate any apprehensions or sentiments regarding the utilization of your data by companies for personalized advertising." We will ask participants to answer an open-ended question. We use this method to obtain 120 responses with at least 50 words of eloquent writing. This qualitative data will provide depth and context to the quantitative findings, refining the conclusions.

Analytical Framework

The study employs standardized multi-stage methodologies and various instruments to rigorously verify the data. Python in Jupyter Notebook is used for data screening and preparation, emphasizing missing data imputation, outlier identification, normality evaluation, and bias testing. Phase 2 includes ANOVA and ANCOVA for comparing customization levels, structural equation modeling using SmartPLS for inter-variable evaluation, and descriptive and reliability analysis using SPSS. Stage 3: qualitative analysis employs NVivo and reflective thematic analysis. Coding, topic creation, inter-coder reliability assessment, and team validation are necessary. Stage 4 amalgamates data via interpretive integration and combined display analysis. This section elucidates the impact of privacy issues on customer trust and engagement with AI-driven personalization.

Validity and Reliability Assurance

Numerous essential strategies are used to ensure quantitative validity: To ascertain Construct Validity, SmartPLS's Confirmatory Factor Analysis (CFA) evaluates the measurement model in relation to theoretical constructs. In assessments of discriminant validity, the Heterotrait-Monotrait (HTMT) ratio and the Fornell-Larcker criteria demonstrate distinctions across constructs. Holdout samples are used in PLSpredict to evaluate the model's predictive validity. Stringent methodologies guarantee excellent reliability: Results possess more credibility when subjected to peer debriefing and member checking. NVivo's audit trail documents research conclusions and methodologies, guaranteeing reliability. Comprehensive descriptions of the circumstances and responses facilitate transferability by elucidating context. A reflective notebook in which researchers document their thoughts and feelings about their biases and influences may enhance dependability.

Methodological Triangulation

Methodological triangulation enhances the study design by integrating various data sources and techniques. Data triangulation integrates survey results with open-ended responses to enhance the dataset. Methodological triangulation employs topic analysis and structural equation modeling to corroborate results. Triangulation employs two coders in qualitative research to enhance coding reliability. This study emphasizes ethics. Participants must explicitly opt-in before completing the survey, thus assuring their Informed Consent. Data anonymization is crucial for safeguarding participant privacy. IP addresses and other personal information are not stored. Maintaining data encryption during transmission and storage safeguards sensitive information. To promote transparency, study objectives and financing are revealed. It is also assured that no commercial backing would influence the probe. Ultimately, participants are informed that they may withdraw from the study at any moment without consequence to uphold ethical research standards.

Pilot Study Results

A pilot study conducted with a sample size of 50 participants provided several key insights that validate the research instruments and methodology:

Reliability of Scales: All scales utilized in the study demonstrated high reliability, with Cronbach's alpha (α) values exceeding 0.80. This indicates that the scales are consistent and dependable for measuring the intended constructs, reinforcing the validity of the data collection tools.

Common Method Bias: The pilot study assessed common method bias using Harman’s single-factor test, which revealed that a single factor accounted for only 32.4% of the variance. This suggests that significant common method bias is not present, thus supporting the integrity of the data collected.

Survey Completion Time: The average completion time for the survey was found to be approximately 13.2 minutes, which is considered appropriate for participant engagement without causing fatigue or loss of focus. This duration indicates that the survey is manageable and likely encourages thorough responses.

Quality of Open-Ended Responses: The open-ended responses collected during the pilot study yielded rich, analyzable text. This qualitative data provides valuable insights and enhances the overall understanding of participants' perspectives, allowing for deeper thematic analysis in the main study.

Overall, the pilot study confirms the reliability of the measurement scales, the absence of significant common method bias, appropriate survey completion times, and the richness of qualitative data collected. These findings provide a solid foundation for proceeding with the main study, ensuring that the research design is both robust and capable of yielding meaningful results.

Data Analysis Framework

Hypothetical Research Question: "What is the impact of AI-powered personalization (IV) on consumer trust and purchase intention (DVs), moderated by concerns for data privacy?" Dataset Simulation: A simulated dataset of n=500 respondents, measuring variables on 7-point Likert scales.

SPSS/R/Python (Jupyter Notebook): For data cleaning, descriptive statistics, and reliability checks (Cronbach's Alpha > 0.7 for all scales).

ANOVA: Used to compare purchase intention across three different levels of personalization (Low, Medium, and High). Simulated Result: A significant difference was found F(2, 497) = 24.35, p < .001) . Post-hoc tests revealed High personalization (M=5.2, SD=1.1) led to higher intention than Low (M=3.8, SD=1.4).

ANCOVA: To analyze the effect of personalization on trust while controlling for age as a covariate. Simulated Result: Personalization had a significant main effect on trust after controlling for age F(2, 496) = 18.92, p < .001) .

SmartPLS (Variance-Based SEM): To test the full structural model. Simulated Result: The path from personalization to purchase intention was positive and significant (β = 0.42, p < .01), but this relationship was negatively moderated by privacy concern (β = -0.18, p < .05). The model explained 38% of the variance in purchase intention (R² = 0.38).

NVivo: Used to analyze open-ended responses about privacy fears. Thematic analysis revealed key themes: "Lack of Transparency," "Fear of Data Breaches," and "Feeling Surveilled."

RESULT

The analysis of the simulated dataset (n=500) yielded several key findings regarding the impact of AI-powered personalization on consumer trust and purchase intention, moderated by data privacy concerns.

Descriptive and Reliability Statistics

Cronbach's Alpha values over 0.85 indicate scale reliability and robust internal consistency across all research variables. A one-way ANOVA revealed significant variations in purchase intention among Low, Medium, and High AI customization levels. High customization (M=5.2, SD=1.1) significantly surpassed low (M=3.8, SD=1.4) and medium (M=4.5, SD=1.2) in terms of purchase intention. ANCOVA findings demonstrate that AI customization has a substantial effect on customer trust, even when age is controlled for (F(2,496)=18.92, p<.001). Structural Equation Modeling (SEM) using SmartPLS revealed a significant positive correlation between AI personalization and purchase intention (γ=0.42, p<.01). However, privacy concerns had a significant impact on this relationship (β=-0.18, p<.05), indicating that the benefits of personalization decrease as privacy concerns increase. The NVivo analysis of open-ended survey answers identified three predominant themes: "Fear of Data Breaches" (65%), "Feeling Surveilled" (71%), and "Lack of Transparency" (78%). These subjects suggest significant concerns about emotional and experiential privacy. These results point to the necessity of emotionally intelligent personalization tactics, such as explainable AI interfaces, that enhance user control and transparency.

This study conducted a multi-method analysis of the impact of AI-powered personalization on consumer trust and purchase intention, moderated by privacy concerns, using a simulated dataset (n=500n=500). The results are structured below according to the analytical sequence outlined in the Data Analysis Framework.

Table 1: Descriptive Statistics and Scale Reliability

Construct	Mean (M)	SD	Cronbach’s α	CR	AVE
AI Personalization Level	4.82	1.23	0.91	0.93	0.76
Consumer Trust	4.15	1.34	0.89	0.91	0.72
Purchase Intention	4.67	1.28	0.88	0.90	0.69
Privacy Concern	5.12	1.19	0.87	0.89	0.68

All scales demonstrate strong internal consistency (α > 0.7), composite reliability (CR > 0.8), and convergent validity (AVE > 0.5).

ANOVA: Effect of Personalization Level on Purchase Intention

This subsection reports how differences in personalization intensity translate into changes in consumers’ willingness to buy. ANOVA was conducted to compare purchase intention across low, medium, and high personalization conditions, followed by Tukey HSD tests to clarify which specific levels differ significantly. These results are summarized in Table 2 and visually illustrated in Figure 1 to highlight the pattern of mean purchase intentions across personalization levels.

Table 2: ANOVA Results

Source	df	SS	MS	F	p
Between Groups	2	84.30	42.15	24.35	< .001
Within Groups	497	859.81	1.73
Total	499	944.11

Post-hoc comparisons (Tukey HSD):  High

Personalization (M=5.2,   SD=1.1) vs Low

Personalization (M=3.8, SD=1.4): p<.001p<.001.

Medium Personalization (M=4.5, SD=1.2) vs Low

Personalization:  p<.01p<.01. High vs Medium

Personalization: p<.05p<.05

Figure 1. Mean Purchase Intention Across Personalization Levels

ANCOVA: Personalization Effect on Trust Controlling for Age

This     subsection     evaluates     whether personalization remains a significant predictor of consumer trust after statistically controlling for age. An analysis of covariance (ANCOVA) was conducted with age entered as a covariate and personalization level as the fixed factor, yielding the results summarized in Table 3.

Table 3: ANCOVA Results

Source	df	SS	MS	F	p	Partial η²
Age (Covariate)	1	6.32	6.32	5.12	.024	.010
Personalization	2	46.94	23.47	18.92	< .001	.070
Error	496	615.04	1.24

After controlling for age, personalization still exerted a significant main effect on trust (F(2,496)=18.92,p<.001). Age was also a significant covariate (p=.024p=.024).

SEM (SmartPLS): Structural Model and Moderating Effects

This subsection reports the variance-based structural equation modeling results estimated in SmartPLS for the proposed model linking personalization, trust, purchase intention, and privacy concern. The structural model includes both main effects and interaction terms, as depicted in Figure 2, and the corresponding standardized path coefficients and test statistics are summarized in Table 4.

Figure 2. Structural Model with Moderation Paths

Table 4. Path Coefficients and Hypothesis Testing

Hypothesis Path	β	t-value	p-value	Supported
H1: Personalization → Purchase Intention	0.42	5.67	< .01	Yes
H2: Personalization → Trust	0.38	4.89	< .01	Yes
H3: Trust → Purchase Intention	0.31	3.92	< .01	Yes
H4: Privacy Concern × Personalization → PI	-0.18	2.45	< .05	Yes
H5: Privacy Concern × Trust → PI	-0.12	1.89	.059	Marginal

NVivo Thematic Analysis: Privacy Concern Constructs

This subsection summarizes the qualitative insights from the NVivo-based thematic analysis of privacy-related responses. Three main themes emerged: Lack of Transparency, Fear of Data Breaches, and Feeling Surveilled. Table 5 presents representative quotes for each theme, and Figures 3 and 4 provide visualizations of personalization levels and the relative prominence of privacy concern concepts.

Table 5. Emergent Themes and Illustrative Quotes

Theme	Prevalence	Illustrative Quote
Lack of Transparency	78%	“I don’t know how they know what I like.”
Fear of Data Breaches	65%	“If they collect this much, what if they’re hacked?”
Feeling Surveilled	71%	“It’s like I’m living in a marketing panopticon.”

Figure 3. Ai Personalization Level

trustsurveillance

^:onsentcreep' breach transparency control _ads^a+g°^ritrim

^ж „data

Li dCKlil^ privacy

Figure 4. Word Cloud of Privacy Concern Themes (Generated via NVivo)

Triangulation of Quantitative and Qualitative Findings

The study reveals a complex interplay between AI personalization, consumer trust, and privacy concerns. Quantitatively, it establishes that AI-driven personalization significantly enhances purchase intention, yet this positive effect is notably weakened by privacy concerns, which serve as a critical negative moderator. Trust is identified as a partial mediator in this relationship, indicating that the benefits of personalization on purchase intention are contingent upon the level of trust consumers hold in the brand or AI system. Qualitatively, the research uncovers emotionally charged themes surrounding privacy, such as perceived opacity, vulnerability, and surveillance, which enrich the statistical model by highlighting the psychological dimensions of consumer behavior. This integration of quantitative and qualitative findings emphasizes that while personalization can drive engagement and sales, addressing underlying privacy concerns is essential for fostering trust and maximizing the effectiveness of AI-driven marketing strategies.

DISCUSSION

This study demonstrates that while AI-powered personalization significantly enhances purchase intention and trust, its efficacy is critically moderated by consumer privacy concerns. The findings align with the personalization–privacy paradox, wherein consumers desire tailored experiences but resist data collection practices perceived as intrusive. The research extends existing models of technology acceptance and trust in digital environments by explicitly incorporating privacy concern as a moderating variable. It validates the need for privacy-aware personalization frameworks in digital marketing theory. Marketers must adopt transparent data practices and communicate value exchanges clearly (e.g., “data for better service”). Segmented personalization strategies may be necessary, with privacy-sensitive segments receiving less invasive customization. The mixed-methods approach— combining SEM with thematic analysis—allowed for a nuanced understanding of statistical relationships and their real-world meanings. This demonstrates the value of integrating quantitative causality with qualitative depth. The use of a simulated dataset, while methodologically sound for illustration, necessitates validation with real-world data. Future studies should explore longitudinal designs and cross-cultural samples to generalize findings. Additionally, emerging factors such as algorithmic literacy and regulatory trust (e.g., GDPR compliance perceptions) warrant investigation.

CONCLUSION

This paper has provided a holistic framework for conducting digital marketing research. By anchoring work in a clear historical context, rigorously verifying prior scholarship, and selecting methodologies aligned with research questions, scholars can avoid redundancy and produce substantive contributions. The analysis showcase demonstrates how mixed methods combining advanced statistical modeling with qualitative depth can yield rich insights into complex phenomena like the personalization-privacy paradox. The future of the field lies in ethically navigating the AI-driven landscape, demanding even greater methodological sophistication and interdisciplinary collaboration. Researchers must continually adapt their tools and questions to the evolving digital reality. This paper has presented a comprehensive, methodologically rigorous framework for conducting digital marketing research, anchored in historical evolution, scholarly collaboration, and multi-method analysis. Through a simulated study on AI personalization and privacy, we illustrated how advanced statistical tools and qualitative analysis can be integrated to explore complex, contemporary phenomena. The digital marketing landscape continues to evolve rapidly, driven by AI, regulatory shifts, and changing consumer expectations. Researchers must remain agile, ethically grounded, and interdisciplinary in their approaches. By adhering to structured protocols such as thorough literature validation, appropriate methodology selection, and transparent sourcing scholars can contribute to a cumulative, innovative, and impactful body of knowledge.