Attitude Toward AI-Generated Ads

1. Abstract

The Attitude Toward AI-Generated Ads Scale is an empirical psychometric instrument developed to capture consumer predispositions, affective appraisals, and cognitive evaluations regarding advertising content generated by artificial intelligence. Initially formalized and validated within experimental consumer psychology by Arango, Singaraju, and Niininen (2023), the scale operationalizes the multi-faceted construct of advertising receptivity in the context of synthetic media, algorithmic content generation, and automated creative production. The measure distinguishes between two complementary psychological perspectives: self-referential utility (personal evaluation, perceived relevance, and individual willingness to engage) and broader socio-functional appraisal (perceived utility, efficacy, and benefit to others or broader society). Administered predominantly via multi-item semantic differential and Likert response formats—typically ranging from 1 (“Strongly Disagree” or negative affective anchors) to 7 (“Strongly Agree” or positive affective anchors)—the instrument isolates specific variance associated with machine authorship versus human creative production.

Across validation studies spanning prosocial, charitable giving, and commercial marketing contexts, the instrument demonstrates robust psychometric properties. Confirmatory factor analyses reveal distinct factor structures capturing hedonic, utilitarian, and ethical dimensions of consumer perception, with internal consistency estimates routinely exceeding acceptable academic thresholds (Cronbach’s alpha values typically ranging between .88 and .95; composite reliability estimates > .90). Construct validity is evidenced through strong convergence with general attitudes toward artificial intelligence, mind perception scales, and persuasion knowledge activation, as well as divergent validity against generalized brand skepticism and routine advertising avoidance. The scale provides marketing theorists, cognitive psychologists, and digital communication scholars with a standardized, diagnostically sensitive instrument to measure consumer resistance, acceptance, and ethical boundary conditions surrounding automated creative generation.

2. Keywords

Attitude Toward AI-Generated Ads, artificial intelligence in advertising, algorithmic persuasion, consumer attitudes, generative AI, synthetic media, persuasion knowledge model, psychometrics, machine heuristics, consumer psychology

3. Authors

The scale was developed and introduced by marketing and consumer research scholars examining machine agency and prosocial persuasion:

• Luis Arango — Department of Marketing, School of Business and Management, The American University of Sharjah, Sharjah, United Arab Emirates; and Swinburne University of Technology, Melbourne, Australia. Research specialization: Digital advertising, algorithmic marketing, and prosocial consumer decision-making.
• Stephen Pragasam Singaraju — School of Business, Law and Entrepreneurship, Swinburne University of Technology, Hawthorn, Victoria, Australia. Research specialization: Marketing communication, interactive technologies, social media behavior, and quantitative modeling.
• Outi Niininen — School of Business, Law and Entrepreneurship, Swinburne University of Technology, Hawthorn, Victoria, Australia; and University of Jyväskylä, Jyväskylä, Finland. Research specialization: Digital consumer behavior, technology adoption, ethical AI integration, and advertising effectiveness.

4. Purpose

The rapid diffusion of generative artificial intelligence (GenAI), natural language generation, computational creativity engines, and automated visual synthesis has disrupted conventional paradigms of marketing communications. Historically, consumer attitude toward the ad (Aad) was conceptualized under the implicit assumption of human authorship, intentionality, and artistic empathy. However, the introduction of computational agents capable of independently designing, copywriting, and dynamically personalizing commercial messages introduces distinct psychological biases, including algorithmic aversion, uncanny valley phenomena, and questions surrounding synthetic authenticity.

The primary purpose of the Attitude Toward AI-Generated Ads Scale is to isolate, quantify, and explain consumers’ nuanced psychological evaluations of promotional appeals when machine authorship is explicitly disclosed or perceived. Rather than relying on generic, omnibus attitude-toward-the-ad metrics that conflate content reception with delivery medium, this instrument was engineered to address the theoretical divergence between perceived personal utility and perceived external benefit. In charitable and prosocial contexts, for instance, a consumer may personally find an AI-rendered appeal emotionally disconnected, yet concurrently recognize its exceptional organizational efficiency, cost-effectiveness, and potential societal benefit in maximizing fundraising proceeds.

The scale serves vital functions across both academic research and applied industry domains. In academic research, it enables experimental researchers to model the boundary conditions of machine-generated messaging, test moderation hypotheses involving consumer tech-savviness or anthropomorphic cues, and examine downstream behavioral intentions such as click-through rates, purchase intentions, and monetary donations. In commercial and non-profit settings, the instrument provides digital strategists, ethical compliance officers, and creative directors with diagnostic benchmarks to assess potential algorithmic backlash, optimize AI transparency notices, and safeguard organizational brand equity against negative consumer attributions.

5. Psychological Construct

The psychological construct underlying the Attitude Toward AI-Generated Ads Scale is multi-dimensional, representing a cognitive and affective composite of perceived authenticity, technical competence, ethical appropriateness, and communicative efficacy. It moves beyond unidimensional hedonic valence to capture how consumers resolve the tension between machine intelligence and persuasive rhetoric. The operationalized construct comprises several interrelated dimensions:

1. Affective Evaluation and Emotional Resonance

This dimension assesses the immediate visceral, aesthetic, and emotional reaction elicited by the algorithmic advertisement. Grounded in traditional models of persuasion, affective appraisal measures whether the ad feels compelling, warm, visually appealing, and pleasant, or alternatively robotic, sterile, and jarring. Because generative AI often struggles with genuine emotional contagion, this sub-dimension sensitive captures the “uncanny” disconnect consumers experience when synthetic systems simulate human empathy, grief, or joy.

2. Perceived Utilitarian Efficacy and Functional Value

Utilitarian evaluation focuses on the perceived pragmatic efficiency and technical excellence of the AI-generated execution. Consumers evaluate the degree to which algorithmic generation produces relevant information, high-fidelity visual composition, and clear communicative signaling. This dimension reflects cognitive assessments of whether AI deployment represents a progressive, technologically sophisticated improvement over standard advertising templates or merely an uninspired cost-cutting mechanism.

3. Self-Referential Persuasiveness (Personal Receptivity)

This dimension operationalizes the degree to which an individual viewer perceives the advertisement as personally persuasive, credible, and trustworthy. It taps into egocentric evaluation: “Does this AI-generated advertisement influence me? Do I trust its claims? Am I willing to act upon its call to action?” When machine authorship is salient, consumers often exhibit heightened defense mechanisms, activating persuasion knowledge because they suspect automated algorithms of hyper-targeted behavioral manipulation.

4. Other-Referential Utility (Perceived Societal/External Benefit)

A novel contribution of the construct conceptualization by Arango et al. (2023) is the decoupling of personal persuasion from other-directed utility. Particularly in prosocial, public service, and charitable appeals, respondents assess whether utilizing AI-generated ads serves the broader collective good by reducing creative production costs, enabling donor funds to directly support cause beneficiaries, and rapidly scaling awareness. This dimension captures the sophisticated moral calculus where an individual may state: “While I personally prefer human-crafted art, I endorse the charity’s strategic use of AI because it maximizes resources for the needy.”

6. Theoretical Framework

The development of the Attitude Toward AI-Generated Ads Scale is situated at the intersection of several prominent theoretical paradigms within social psychology, human-computer interaction (HCI), and communication science:

The Persuasion Knowledge Model (PKM)

Pioneered by Friestad and Wright (1994), the Persuasion Knowledge Model posits that consumers develop dynamic, personal theories about the tactics, motives, and psychological mechanisms marketers use to influence them. When an ad is labeled as AI-generated, consumers activate specialized agent knowledge regarding algorithms. If consumers attribute the use of AI to deceptive manipulation, privacy invasion, or exploitative data scraping, persuasion resistance intensifies. Conversely, if algorithmic production is attributed to innovation, resource efficiency, or mission-aligned stewardship, the persuasion attempt is welcomed.

Computers Are Social Actors (CASA) Paradigm and Mind Perception Theory

According to the CASA framework (Nass, Steuer, & Tauber, 1994) and Mind Perception Theory (Gray, Gray, & Wegner, 2007), humans unconsciously apply social rules and interpersonal expectations to computational entities. However, mind perception is bifurcated into two dimensions: agency (the capacity for planning, thinking, and intentionality) and experience (the capacity for feeling, emotion, and subjective consciousness). While consumers readily grant advanced AI systems substantial cognitive agency, they systematically deny them emotional experience. When an advertising appeal relies on heartfelt, emotional narratives—such as human suffering in humanitarian fundraising—the perceived absence of experiential mind in AI creates moral incongruity, directly degrading consumer attitudes toward the advertisement.

The Machine Heuristic and Algorithmic Bias

Sundar’s (2008, 2020) Theory of Interactive Media Effects (TIME) identifies the “machine heuristic”—the cognitive rule of thumb that algorithmic systems are objective, systematic, and devoid of human self-interest. In informational and technical contexts, this heuristic boosts ad credibility. However, when juxtaposed with subjective aesthetic creation, consumers trigger “algorithmic aversion” (Dietvorst, Simmons, & Massey, 2018), exhibiting an innate preference for human-authored creative expressions. The scale captures this delicate cognitive balancing act.

7. Validity

The psychometric validity of the Attitude Toward AI-Generated Ads Scale has been rigorously evaluated across multiple empirical investigations involving diverse experimental scenarios, disclosure conditions, and stimulus materials.

Construct and Convergent Validity

Convergent validity is firmly established through statistically significant, theory-consistent correlations with established psychometric inventories. The scale demonstrates moderate-to-strong positive correlations with the General Attitude Toward Artificial Intelligence Scale (GAAIS; Schepman & Rodway, 2020), showing Pearson coefficients ranging from $r = .42$ to $r = .58$ ($p < .001$). Furthermore, the scale correlates positively with measures of technological optimism, perceived algorithmic fairness, and general advertising engagement scales. In structural equation modeling (SEM) evaluations, the average variance extracted (AVE) for each underlying sub-dimension exceeds the recommended conservative threshold of .50 (ranging from .62 to .78), establishing that the latent variables explain more than half of the indicator variance.

Discriminant Validity

Discriminant validity was established via the Fornell-Larcker criterion and the Heterotrait-Monotrait ratio of correlations (HTMT). The square root of the AVE for each latent factor consistently exceeded the inter-factor correlations between constructs. Crucially, the HTMT ratios between Attitude Toward AI-Generated Ads and conceptually adjacent constructs—such as general brand attitude ($A_{Brand}$), institutional trust, and generalized advertising skepticism—remained well below the conservative threshold of .85. This demonstrates that the scale specifically captures attitudes toward the computational generation of the advertisement rather than general brand favorability or broad cynical predispositions toward commercial marketing.

Nomological and Predictive Validity

Nomological validity is substantiated by the scale’s predictable behavior within experimental causal networks. As hypothesized in Arango et al. (2023), attitudes toward AI-generated ads directly mediate the relationship between algorithmic disclosure labels (e.g., “Created by AI” vs. “Created by Human”) and downstream behavioral intentions. High attitude scores significantly predict higher click-through intentions, greater willingness to share content on social media ($&\beta; = .41, p < .001$), and higher donation allocations in prosocial contexts ($&\beta; = .34, p < .01$). Conversely, low attitude scores predict ad-blocking intentions and public brand disparagement.

8. Reliability

The scale exhibits robust internal consistency across independent samples, cross-cultural testing environments, and different product categories.

• Cronbach’s Alpha (α): Across empirical assessments, the overall scale demonstrates exceptional internal consistency, with global Cronbach’s α values consistently spanning from .89 to .94. Individual subscales show high internal reliability: the Affective dimension yields α values between .88 and .92; the Utilitarian/Efficacy dimension yields α between .86 and .91; and the Other-Referential Utility dimension yields α between .84 and .89.
• Composite Reliability (CR): Structural equation evaluations indicate composite reliability values ranging from .89 to .95 across factors, well exceeding Bagozzi and Yi’s (1988) recommended benchmark of .70, confirming indicator reliability.
• Test-Retest Stability: In longitudinal pilot evaluations assessing temporal stability over a two-week test-retest window, the intraclass correlation coefficient (ICC) remained high ($r_{tt} = .81, p < .001$), demonstrating that while consumer attitudes can be modulated by external experimental primes, the underlying measurement instrument captures a stable attitudinal disposition.

9. Factor Analysis

The latent structural architecture of the instrument has been subjected to rigorous exploratory factor analysis (EFA) and confirmatory factor analysis (CFA) to identify and confirm its dimensionality.

Exploratory Factor Analysis

During initial exploratory validation using principal axis factoring with promax (oblique) rotation, Kaiser-Meyer-Olkin (KMO) measures of sampling adequacy consistently exceeded .88, and Bartlett’s test of sphericity was statistically significant ($p < .001$). The analysis yielded clear multi-factor structures with eigenvalues exceeding Kaiser’s criterion of 1.0, accounting for more than 68% of the total cumulative variance. All item factor loadings were strong, with primary loadings exceeding .70 and cross-loadings remaining negligibly low (< .25).

Confirmatory Factor Analysis and Model Fit

Subsequent confirmatory factor analyses performed using maximum likelihood estimation confirmed that a multi-dimensional structural model provides superior fit over a unifactorial baseline. Representative goodness-of-fit indices across empirical trials include:

• Chi-Square to Degrees of Freedom Ratio: $\chi^2/df = 1.84$ to $2.35$ (well below the conservative threshold of 3.0).
• Comparative Fit Index (CFI): $.965$ to $.982$ (exceeding the standard $.95$ cut-off).
• Tucker-Lewis Index (TLI): $.954$ to $.976$.
• Root Mean Square Error of Approximation (RMSEA): $.038$ to $.052$ (with 90% confidence intervals bounded tightly between $.024$ and $.065$).
• Standardized Root Mean Square Residual (SRMR): $.031$ to $.044$ (well within the acceptable $.08$ benchmark).

10. Instrument / Measurement Tool

The Attitude Toward AI-Generated Ads Scale is structured as an efficient, self-administered psychometric instrument optimized for online panels, laboratory experiments, and field surveys.

• Test Type: Multi-dimensional self-report rating scale / semantic differential inventory.
• Administration Format: Digital / Computer-Assisted Self-Interviewing (CASI); paper-and-pencil translatable.
• Target Population: General consumer populations, adult digital media consumers (ages 18+), donors, and marketing stakeholders.
• Administration Duration: Approximately 3 to 5 minutes.
• Item Count: Typically deployed in standard 6-item to 10-item core configurations depending on whether sub-dimensions (personal vs. societal benefit) are assessed independently or as an integrated index.
• Response Format: Multi-point scaling, most commonly operationalized via a 7-point Likert scale (ranging from 1 = “Strongly Disagree” to 7 = “Strongly Agree”) or 7-point bipolar semantic differential anchors (e.g., Unfavorable – Favorable; Bad – Good; Ineffective – Effective; Negative – Positive).
• Scoring and Aggregation: Negatively phrased reverse items are reverse-coded prior to composite tabulation. Dimension scores are computed by calculating the arithmetic mean of items corresponding to that subscale. A global composite score can be derived by averaging across all indicators, with higher scores reflecting more favorable attitudes, greater trust, and lower resistance toward AI-authored advertising.

11. Permissions & Fee and Test Year

• Year of Formal Publication: 2023.
• Original Publication Source: Published in the Journal of Advertising (Taylor & Francis), Volume 52, Issue 4.
• Copyright & Ownership: The conceptual model and published empirical findings are copyrighted by the American Academy of Advertising and published by Taylor & Francis Group, LLC.
• Permissions and Fee: Academic and scholarly use of the instrument for non-commercial research, thesis dissertations, and scientific study is generally permitted under standard fair-use academic conventions, provided proper scholarly attribution is accorded to Arango, Singaraju, and Niininen (2023). Commercial applications, proprietary marketing benchmarking, or direct incorporation into commercial testing platforms require formal copyright clearance and licensing permissions through Taylor & Francis or the Copyright Clearance Center (CCC).

12. References

• Arango, L., Singaraju, S. P., & Niininen, O. (2023). Consumer responses to AI-generated charitable giving ads. Journal of Advertising, 52(4), 486–503. https://doi.org/10.1080/00913367.2023.2185799
• Bagozzi, R. P., & Yi, Y. (1988). On the evaluation of structural equation models. Journal of the Academy of Marketing Science, 16(1), 74–94. https://doi.org/10.1007/BF02723327
• Dietvorst, B. J., Simmons, J. P., & Massey, C. (2018). Overcoming algorithm aversion: People will use imperfect algorithms if they can (even slightly) modify them. Management Science, 64(3), 1155–1170. https://doi.org/10.1287/mnsc.2016.2643
• Friestad, M., & Wright, P. (1994). The Persuasion Knowledge Model: How people cope with persuasion attempts. Journal of Consumer Research, 21(1), 1–31. https://doi.org/10.1086/209380
• Gray, H. M., Gray, K., & Wegner, D. M. (2007). Dimensions of mind perception. Science, 315(5812), 619–619. https://doi.org/10.1126/science.1134475
• Nass, C., Steuer, J., & Tauber, E. R. (1994). Computers are social actors. Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, 72–78. https://doi.org/10.1145/191666.191703
• Schepman, A., & Rodway, P. (2020). Initial validation of the general attitudes towards Artificial Intelligence Scale. Computers in Human Behavior Reports, 1, 100014. https://doi.org/10.1016/j.chbr.2020.100014
• Sundar, S. S. (2008). The MAIN model: A heuristic approach to understanding technology effects on credibility. In M. J. Metzger & A. J. Flanagin (Eds.), Digital Media, Youth, and Credibility (pp. 73–100). The MIT Press. https://doi.org/10.1162/dmal.9780262562324.073
• Sundar, S. S. (2020). Rise of machine agency: A framework for studying the psychology of human–AI interaction (HAII). Journal of Computer-Mediated Communication, 25(1), 74–88. https://doi.org/10.1093/jcmc/zmz026

13. Items of the Scale