Smart Skills for Smart Cities: Developing and Validating an AI Soft Skills Scale in the Framework of the SDGs

Abstract

Artificial intelligence (AI) soft skills have become increasingly vital in today’s technology-driven world, as they support decision-making systems, strengthen collaboration among stakeholders, and enable individuals to adapt to rapidly changing environments—factors that are fundamental for achieving the sustainability goals of smart cities. Even though AI soft skills are becoming more important, no scale specifically designed to identify and evaluate individuals’ AI soft skills has been found in the existing literature. Therefore, this paper aimed to develop a reliable and valid scale to identify the AI soft skills of individuals. A sample of 685 individuals who were employed in AI-active sectors, with a minimum of a bachelor’s degree, and at least one year of AI-related work experience, participated in the study. A sequential exploratory mixed-methods research design was utilized. Exploratory factor analysis (EFA) identified a five-factor structure that accounted for 67.37% of the total variation, including persuasion, collaboration, adaptability, emotional intelligence, and creativity. Factor loadings ranged from 0.621 to 0.893, and communalities ranged from 0.587 to 0.875. Confirmatory factor analysis (CFA) supported this structure, with strong model fit indices (GFI = 0.940, AGFI = 0.947, NFI = 0.949, PNFI = 0.833, PGFI = 0.823, TLI = 0.972, IFI = 0.975, CFI = 0.975, RMSEA = 0.052, SRMR = 0.035). Internal consistency for each factor was high, with Cronbach’s alpha values of dimensions ranging from 0.804 to 0.875, with a value of 0.921 for the overall scale. Convergent and discriminant validity analyses further confirmed the construct’s robustness. The finalized AI soft skills (AISS) scale, consisting of 24 items, offers a psychometrically valid and reliable tool for assessing essential AI soft skills in professional contexts. Ultimately, this developed scale enables the determination of the social and cognitive skills needed in the human-centered and participatory governance structures of smart cities, supporting the achievement of specific Sustainable Development Goals such as SDG 4, SDG 8, and SDG 11, and contributes to the design of policies and training programs to eliminate the deficiencies of individuals in these areas. Thus, it becomes possible to create qualified human resources that support sustainable development in smart cities, and for these individuals to take an active part in the labor market.

1. Introduction

A smart city is a holistic urban management model that simultaneously improves resource efficiency, service quality and environmental sustainability through the integration of innovative technology-based systems at every stage of urban operations, such as transportation, energy, waste, public services, etc. [1]. In this approach, technology, especially artificial intelligence, one of the most popular scientific fields in recent years, is the main determining factor that will provide instant and accurate decision-making in businesses and effective control of the operational processes of businesses. Therefore, artificial intelligence has the potential to make significant contributions to the design and development of smart cities of the future in terms of cognition and technology [2]. Soft skills of AI, such as communication, problem solving, and adaptation, play a critical role in decision support systems, interaction between stakeholders, and adaptation processes to changing environmental conditions in achieving the sustainability goals of smart cities. In this context, the socio-technical integration of artificial-intelligence-supported systems enables the effective realization of sustainable development goals [3].

In today’s world, where digital transformation is rapidly advancing, it has become inevitable—particularly for businesses—to adapt to technological innovations in order to maintain their competitive power and ensure sustainable growth [4,5]. Artificial intelligence technologies play a wide range of critical roles, from the automation of operational processes to decision support systems, in many sectors such as education [6], finance [7], manufacturing [8] marketing [9] healthcare [10], politics [11], and human resources [12], and smart city applications, contributing to the achievement of Sustainable Development Goals (SDGs) by fostering innovation, efficiency, and sustainability.

According to Hostinger’s report [13], as of 2025, 78% of businesses globally have implemented AI, which represents a notable growth compared to previous years. In fact, approximately 280 million of the 359 million businesses worldwide utilize AI in some capacity. Also, Hostinger reported that, on average, firms are currently adopting AI in three distinct functions, demonstrating a significant increase since the beginning of 2024. It is pointed out that AI is utilized by business functions such as marketing and sales (36%), service operations (33%), for product and/or service development (31%), human resources (20%), and manufacturing (12%). Also, it was mentioned that around 89% of small companies use AI tools to handle daily operations like communication (generating e-mails), data analysis, and the production of marketing material. Currently, the effect seems to be favorable: more than 60% of owners of small enterprises that use AI report increases in worker productivity and job satisfaction. Additionally, it was noted that many AI-using enterprises report cost reductions, particularly in service operations. AI helps save expenditure in this area, according to about 49% of respondents. Notable savings are also being seen in software engineering (41%), supply chain and inventory (43%), and other activities. Also, according to Stanford’s 2025 AI Index Report [14], 78% of organizations reported using AI in 2024, up from 55% the previous year. PwC’s 2025 Global AI Jobs Barometer [15] shows that workers with AI skills earn, on average, wages that are 56% higher, and that wage growth in AI-exposed sectors is nearly twice as fast as in others. Moreover, a recent analysis by Bone et al. [16] found that AI skills command a 23% wage premium, exceeding that associated with academic degrees in many high-demand roles.

On the other hand, under today’s conditions, where the competitive environment is intensifying, making strategic and effective decisions is becoming increasingly complex for businesses, particularly in the context of smart city development and innovation. At this point, fully benefiting from the data analysis and forecasting capabilities offered by artificial intelligence technologies has become an indispensable necessity for individuals and institutions to gain a competitive advantage. However, this involves not only using artificial intelligence, but managing it consciously, accurately, and effectively—that is, having strong AI skills directly affects individuals’ success in forecasting the future and in decision-making processes [5,17]. In this context, individuals’ skills in understanding, applying, and optimizing AI tools enhance their capacity to extract meaning from complex data and support strategic decisions, helping businesses to make more accurate predictions for the future.

AI skill development programs lead to greater increases in wages for participating employees [5]. When employees finish AI upskilling programs, companies often increase the size of wage increments more frequently [18]. AI upskilling programs offer significant benefits to workers of all income levels, improving their job stability and opening up new career opportunities [19]. Considering that unemployment and job loss are problems faced by people all around the world today, and that AI is increasingly influencing operations across all sectors, including smart city initiatives, it becomes even clearer that employees need to put effort into improving their AI skills.

On the other hand, according to the United States (U.S.)-based report by Amazon Web Services and Access Partnership [20], acquiring artificial intelligence skills can increase salaries and accelerate career development. Employers in the United States acknowledge the considerable impact of AI on efficiency, productivity, and decision-making, and are therefore willing to offer higher salaries to employees who possess AI-related skills and knowledge. In particular, the survey found that salary advantages are expected to extend across departments, even if companies are ready to pay IT employees with AI skills 47% more on average. U.S. employers stated they would be ready to pay more for workers with AI expertise in the following areas: finance (42%), company administration (41%), legal, regulatory, and compliance (37%), human resources (35%), and sales and marketing (43%). In general, around 80% of workers in the U.S. who responded to the study said they would be interested in learning AI techniques to further their professions. All these findings reveal that AI has a critical role in the workplace, and that the demand is rising not only for specialized AI professionals but also for employees with a more general understanding of AI. Among the U.S.-based employers surveyed who recognize the value AI can bring, 73% view recruiting talent with AI expertise as a top priority. However, nearly three-quarters (75%) report that they are unable to find the talent they need. The rapid digitalization process in recent years has made it essential not only to improve technical infrastructure, but also to ensure that employees possess the knowledge and skills necessary to adapt to and effectively use these new technologies. Increasing global competition in the labor market and shifting job dynamics require individuals to acquire new skills in order to find employment and advance in their careers.

In a globalized economy, having AI-related technological skills, such as soft skills, along with the ability to use them effectively, significantly enhances job opportunities and provides a major advantage in career development. Therefore, it has become a critical necessity for today’s workforce to be equipped not only with technical knowledge, but also with AI-specific soft skills to achieve sustainable success in the business world, particularly in emerging areas such as smart city development. Smart cities are being shaped by the integration of various digital technologies, including the Internet of Things, big data, and cloud computing, as well as artificial intelligence. In this multifaceted digitalization process, the importance of the soft skills individuals require is increasing. AI can facilitate access to systems for individuals with low digital literacy by providing user-friendly interfaces and personalized guidance during the digitalization processes of smart cities and businesses [21]. However, for these technologies to be used effectively and inclusively, soft skills such as empathy, communication, and adaptation also need to be developed. Thus, AI solutions can address not only technical but also social needs, preventing the exclusion of vulnerable groups in both smart cities and digitalizing businesses. A recent survey conducted by Access Partnership and Amazon Web Services (AWS) revealed that 73% of employers prioritize hiring talent with AI skills [22].

Consequently, employees already working in companies, individuals seeking career advancement, or those about to enter the workforce are required to develop AI soft skills such as persuasion, collaboration, adaptability, emotional intelligence, and creativity. AI soft skills encompass non-technical abilities that are important for interacting with artificial intelligence. Moreover, the measurement and evaluation of these skills have become increasingly important. In today’s context, it has become essential for individuals to develop their AI soft skills. As part of broader efforts in skill development, identifying these specific skills is a critical first step toward achieving this goal. AI soft skills gain increasing importance; however, the existing literature lacks a scale for the identification and assessment of individuals’ AI soft skills. While existing studies have developed scales to measure some cognitive [23,24] and affective dimensions [25,26,27] related to AI, a valid and practical measurement tool that can comprehensively assess industry-specific AI soft skills is lacking. This study aims to fill this gap by developing an original scale to analyze the skill levels of individuals actively using AI technologies in various sectors. Thus, it aims to both contribute theoretically to the literature and expand practical measurement capabilities. For this purpose, the main aim of this study is to develop a reliable and valid scale to identify and evaluate the AI soft skills of professionals using AI technologies across different sectors. To achieve this goal, the following research question was sought:

• Is the developed AI soft skills scale, along with its identified factors, valid and reliable in the context of smart cities and digitalizing businesses?

In doing so, it seeks to contribute both to identifying areas for individual development and to enhancing organizations’ training and development strategies. Moreover, this study developed an original scale, the AISS scale, aimed at identifying the AI soft skills individuals need to possess in the era of artificial intelligence the world, aligning with the SDGs by promoting inclusive and future-ready human skills. The developed unique scale, especially aligned with SDG 4, aims to contribute to educational processes in the context of digital literacy and 21st-century skills. Despite increasing emphasis on AI-related skills, there is a lack of comprehensive and practical measurement tools that evaluate sector-specific AI soft skills, which this study addresses by developing a new scale to fill this gap. Additionally, in line with SDG 8 and SDG 11, it can be considered as a tool to determine the skills of the human resources required for building AI-supported sustainable cities. Moreover, in the scale development process, the SDGs and smart cities concepts served as guiding frameworks. Specifically, the items and dimensions of the developed AISS scale were formulated by aligning with the skills required to support inclusive education (SDG 4), sustainable economic growth and decent work (SDG 8), as well as the development of sustainable, AI-enabled urban environments (SDG 11). This ensured that the scale not only measures relevant AI soft skills but also reflects the broader sustainability and smart city agendas, thereby grounding the instrument in real-world applications and policy goals. In this context, the study aims to both increase the quality of education and contribute to sustainable digital transformation.

The Introduction section explains the importance of AI soft skills, their relationship with smart cities and the SDGs, the current lack of scales in the field, and the development of an original, valid, and reliable scale (AISS) to address this gap. The Materials and Methods section details the study design, study group, and scale development process. The Results section describes the comprehensive analyses and all findings. The Discussion section includes a comparative analysis with similar scales in the literature, highlighting that existing studies focus on cognitive and affective dimensions, while highlighting the lack of an industry-specific AI soft skills scale. The Conclusions section details the theoretical and practical contributions and limitations of the developed scale to measure AI soft skills, as well as recommendations for future research.

2. Materials and Methods

2.1. Design

In this study, a sequential exploratory mixed-methods research design was utilized, involving the following steps: (i) generation of items in the scale, (ii) assessment of experts’ opinions, (iii) administration of the questionnaire, and (iv) validation of the scale [28,29]. Qualitative data were collected first through expert interviews and content validity assessments to inform item generation and refinement. Subsequently, quantitative data were gathered via pilot testing and large-scale survey administration for psychometric evaluation of the scale. Both qualitative and quantitative data were gathered to refine the measurement model, confirm the factorial structure, assess the psychometric properties, and ensure the validity and reliability of the scale. The scale development process consisted of a series of systematic and interrelated stages to ensure conceptual clarity, validity, reliability, and overall robustness of the scale. First, a comprehensive literature review was conducted to strengthen the theoretical infrastructure of soft skills related to artificial intelligence, and interviews were conducted with 10 field experts. The scale item development process was grounded in an integrated theoretical framework combining AI skills and soft skills identified as critical for success in AI-related work environments. A comprehensive literature review and expert interviews led to the selection of five key dimensions: persuasion, collaboration, adaptability, emotional intelligence, and creativity. Although the literature identifies a broad range of AI soft skills such as communication, critical thinking, ethical judgment, learning agility, and digital literacy, the present study focuses specifically on persuasion, collaboration, adaptability, emotional intelligence, and creativity. Based on the relevant literature review, it was determined that the most important AI soft skills are persuasion [30], collaboration [31,32], adaptability [33,34], emotional intelligence [31,35,36], and creativity [33,37]. These skills help balance technology and human interaction when working with AI-supported systems, support decision-making processes, and contribute to the development of sustainable business models. Furthermore, these skills align with SDG 4 by promoting inclusive and equitable quality education and lifelong learning opportunities; support SDG 8 by enhancing workforce capabilities and fostering sustainable economic growth; and advance SDG 11 by enabling human-centered, adaptive, and innovative solutions critical for the development of smart and sustainable urban environments. Therefore, these AI soft skills were selected as the main factors for this study. Based on these areas, an item pool was created consisting of statements that reflect the behavioral implications of these AI soft skills in professional settings. For example, items related to persuasion target the ability to advocate for AI systems within the organization, while items related to adaptability reflect individuals’ ability to adapt to changing AI datasets or technologies. Items related to emotional intelligence were drawn from emotional skill theories and adapted to ethical decisions and user experience in AI applications. Then, these items were evaluated by experts in the field in terms of content validity. AI engineers and researchers, HR and competency experts, project managers, IT experts, and domain-specific academics (in computer information systems and AI engineering) who possess practical knowledge of AI technologies and experience in sector-specific competency assessment were consulted for their expert opinions. The experts were based in both academia and the sector, with affiliations in Cyprus. Experts were identified through purposive sampling, based on their academic publications, professional roles, and experience in AI-related projects. Each expert was contacted individually via professional networks (institutional affiliations) and invited to participate. Those who agreed received an information sheet and the draft scale via e-mail. The review process was conducted individually, not as a group discussion, in order to allow for independent assessment without peer influence. Experts were asked to evaluate each item’s clarity, relevance, and alignment with the construct definitions, and to suggest any modifications. Their feedback was collected through structured content evaluation forms. Then, a pilot application was conducted with the preliminary form of the scale, and item analysis was performed based on the obtained data. The finalized scale was applied to a large sample, and the obtained data were subjected to various statistical analyses. These analyses included normality tests, explanatory factor analysis (EFA), and confirmatory factor analysis (CFA) to test the construct validity, convergent and discriminant validity analyses, criterion-related validity, reliability analyses, and item analyses. The “AI Self-Efficacy Scale”, which was developed by Wang and Chuang [25], was used as a reference measure to evaluate the criterion validity of the AI soft skills scale developed in this study. The “AI Self-Efficacy Scale” is a well-established tool that assesses individuals’ self-efficacy in using AI technologies. It was chosen as the criterion because of its strong theoretical alignment with the construct that is being measured. Both scales were administered to the same participant group, and the correlation between total scores was analyzed. If a positive high correlation is found, it will show that the developed scale accurately reflects AI soft skills in line with an established criterion.

2.2. Study Group

A total of 685 individuals voluntarily took part in the study. All participants were in Cyprus, where AI technologies are increasingly integrated across multiple sectors. Participant selection criteria in this AI soft skills development study included individuals who (1) are currently employed in sectors where AI technologies are actively utilized, (2) have at least a bachelor’s degree, and (3) have a minimum of one year of hands-on experience using AI technologies in their professional roles. Exclusion criteria included individuals who were not employed in AI-related sectors, did not hold at least a bachelor’s degree, and had not used AI technology for more than a year in a professional context. In this study, a purposive sampling method was used to choose the sample from different sectors where AI technologies are widely used, namely Finance and Banking, Information Technology and Software, Digital Marketing and Advertising, Engineering, Management, Human Resources, and Project Development. In this study, sectors where AI technologies are actively utilized refer to those in which professionals engage in the direct and continuous application of AI-driven tools and systems such as machine learning, automation, predictive analytics, and AI-supported decision-making in their routine workflows. On the other hand, the dataset contained no missing values. To identify multivariate outliers, the Mahalanobis Distance (MD) measure was applied. Observations exceeding the critical χ² threshold of 46.963 were considered outliers (unusual observations) and removed. Consequently, 5 cases (representing 0.73% of the total sample) were excluded, which proves the dataset’s overall quality and reliability for subsequent analysis. Following their removal, a total of 680 individuals were retained for inclusion in the analysis. In line with previous research on scale development, it is a common procedure to assess data through both exploratory factor analysis (EFA) and confirmatory factor analysis (CFA) by dividing the sample into 2 independent subsets [38,39]. This methodological approach is widely recommended to avoid overfitting and to ensure that the factor structure identified through EFA is not merely a result of sample-specific variance [38,39]. By performing EFA and CFA on independent groups, the robustness, generalizability, and replicability of the factor structure are strengthened. This approach provides the initial model to be explored with one subset (EFA) and independently confirmed with the other (CFA). Accordingly, the total sample of 680 participants was randomly divided into 2 equal groups. The first group (n1 = 340) was used to perform EFA, which assisted in the initial structure and refinement of the scale. The second group (n2 = 340) was employed for CFA to confirm and validate the proposed factor structure derived from the EFA.

Although the possibility of self-reporting bias cannot be completely excluded because participants completed the survey on their own, anonymity and confidentiality were given great importance in our study in order to reduce this bias. During the data collection process, care was taken to ensure that the questions were clear and understandable, and participants were encouraged to give honest answers, assuring them that their answers would remain confidential.

As seen in

Table 1. Frequency distributions of the demographic variables on CFA, EFA, and the total samples.

Variable	Category	EFA Sample (n1 = 340)		CFA Sample (n2 =340)		Total Sample (n = 680)
		F	%	F	%	F	%
Gender	Female	148	43.5	139	40.9	287	42.2
	Male	192	56.5	201	59.1	393	57.8
Age	24–29	71	20.9	80	23.5	151	22.2
	30–34	66	19.4	64	18.8	130	19.1
	35–38	64	18.8	60	17.6	124	18.2
	40–44	82	24.1	70	20.6	152	22.4
	45–49	57	16.8	66	19.4	123	18.1
Education Level	Bachelor’s	201	59.1	187	55.0	388	57.1
	Master’s	95	27.9	95	27.9	190	27.9
	PhD	44	12.9	58	17.1	102	15.0
AI Usage Start Year	Before 2020	8	2.4	11	3.2	19	2.8
	Since 2020	60	17.6	71	20.9	131	19.3
	Since 2021	75	22.1	55	16.2	130	19.1
	Since 2022	64	18.8	72	21.2	136	20
	Since 2023	96	28.2	103	30.3	199	29.3
	Since 2024	37	10.9	28	8.2	65	9.6
Sector	Finance and Banking	89	26.2	103	30.3	192	28.2
	Information Technology and Software	78	22.9	73	21.5	151	22.2
	Digital Marketing and Advertising	65	19.1	76	20.6	141	20.7
	Engineering	38	11.2	30	8.8	68	10.0
	Management, Human Resources, and Project Development	70	20.6	58	17.1	128	18.8

, based on the total sample (n = 680), 57.8% of the participants are male and 42.2% are female. While participants were given “Male,” “Female,” and “Other” gender options, none selected “Other”; thus, the gender distribution reflects only male and female responses. When the age distribution is examined, the densest groups are 40–44 years old, comprising 22.4% of participants, and 24–29 years old, comprising 22.2%. Overall, 57.1% of the participants have a bachelor’s degree, 27.9% have a master’s degree, and 15.0% have a doctorate. The use of AI is most concentrated in participants who started in 2023 (29.3%), followed by users in 2022 (20.0%). The study was conducted in Cyprus, where rates of higher education are notably high, and nearly all working-age professionals hold at least a bachelor’s degree. The observed age range (24–49) and educational levels naturally reflect the inclusion criteria and demographic context. Most individuals complete a bachelor’s degree by the age of 22 and enter the workforce shortly thereafter. Consequently, the youngest professionals who naturally met the inclusion criteria were aged 24. On the upper end, AI-related tasks are more prevalent among younger and mid-career professionals who are more adaptive to technological innovation, which explains why the participant age range did not extend beyond 49. In terms of sectoral distribution, the highest participation was from the finance and banking sector with 28.2%, followed by the IT and software sector with 22.2%. Involving participants from various occupational backgrounds provides the broader applicability and external relevance of the study’s findings, while also acting as a mitigating factor against potential bias associated with self-reported data. The distributions between the EFA and CFA samples are largely similar. In both groups, the gender distribution is in favor of men (56.5% and 59.1%), and the level of education is predominantly composed of bachelor’s degree graduates. Similar rates were observed in terms of artificial intelligence usage by year and sector, which shows that the samples were created homogeneously. This study includes professionals working in the fields of finance and banking (Bank IT Specialist, Financial Analyst, Data Analyst, Bank Manager, Risk Management Specialist), information technology and software (Software Developer, Network Specialist, Data Engineer), digital marketing and advertising (Digital Marketing Specialist, Social Media Manager, Advertiser), engineering, and management, human resources, and project development (Administrator, Businessperson, Project Manager, Human Resources Specialist).

Two weeks following the initial administration of the preliminary survey, the AI soft skills scale was reapplied to 180 individuals randomly chosen within the sample, which evaluates the temporal stability and internal consistency of a measurement instrument over time. This reapplication was essential to verify whether the scale consistently captures the same constructs when administered under similar conditions across two different points in time, thereby enhancing the instrument’s overall reliability. The random selection method reduced selection bias by ensuring that participants had equal chances and increased the wider applicability of test–retest reliability. Specifically, a stratified random sampling method was implemented through a computer-aided algorithm that ensured proportionate representation across key demographic variables such as age, gender, and employment sector. The algorithm assigned a random number to each eligible participant within the original dataset, and a predefined number of participants was then selected from each stratum based on proportional allocation. This procedure not only ensured impartiality but also increased representativeness across subgroups, improving the external validity of the test–retest process. The sample size of 180 was determined under psychometric literature, which suggests that a sample size of at least 100 is sufficient to produce stable and reliable test–retest correlation estimates [40]. The two-week interval allowed for measuring the consistency of responses by reducing the probability that participants would remember their previous responses. Additionally, to minimize the risk of copying or recall bias which could create a false sense of reliability, the survey was re-administered with a re-randomized item order, and participants were unaware they would be retested. This approach is supported by test–retest reliability best practices, ensuring that response consistency reflects true stability rather than memory. This period was found to be appropriate for reliability assessment without significant changes in the participants’ conditions. This method is compatible with accepted practices in the literature and strengthens the reliability of the study [41,42].

2.3. Expert Reviews and Content Validity

During the qualitative phase, a draft 37-items questionnaire was reviewed by a panel of 10 field experts. A panel size of 6 to 10 experts is recommended for content validity analysis [43]. The Content Validity Index (CVI) and Content Validity Ratio (CVR) were computed during the quantitative phase. CVR was determined using Lawshe’s method, where items rated as “essential” by experts must exceed a threshold CVR of 0.78 for 9 experts [44]. For CVI, each item was assessed on a 4-point scale for relevance, clarity, simplicity, and ambiguity [45,46]. Based on expert feedback, a total of 7 items were excluded from the original 37-item pool, as seen in

Table 2. Measures for content validity of the developed AISS scale.

Item	I-CVI (Clarity)	I-CVI (Simplicity)	I-CVI (Ambiguity)	I-CVI (Relevance)	CVR	Decision
I quickly adapt to different use scenarios of AI systems.	0.9	1	0.9	1	1	Retained
2. I synthesize insights from diverse data sources to generate creative outcomes using AI tools.	0.9	0.9	1	0.9	1	Retained
3. I make decisions in AI projects by considering human emotions.	1	1	0.9	0.9	0.8	Retained
4. I am open to learning about artificial intelligence systems.	0	0.4	0	0.3	−0.8	Eliminated
5. I use AI analyses to visualize data in creative and effective ways.	1	1	1	1	1	Retained
6. I ensure coordination with other experts during the development and implementation of AI systems.	1	1	0.9	0.9	1	Retained
7. I persuade decision-makers of the importance of artificial intelligence algorithms.	1	1	1	0.9	1	Retained
8. I create synergy within the team by integrating data from various sources.	1	0.9	1	1	0.8	Retained
9. I prioritize emotional intelligence when evaluating ethical issues in AI projects.	0.9	0.9	1	0.9	0.8	Retained
10. I flexibly adjust AI algorithms to changing datasets or requirements.	0.9	0.9	0.9	0.9	1	Retained
11. I strive to achieve better outcomes by integrating different perspectives in AI projects.	0.9	0.9	1	1	0.8	Retained
12. I understand the emotional needs of AI system users.	1	1	1	0.9	1	Retained
13. I act with empathy and social responsibility when assessing the societal impacts of AI technologies.	1	1	1	0.9	0.8	Retained
14. I adjust my use of AI tools depending on the task at hand.	0	0.3	0.2	0.1	−0.4	Eliminated
15. I develop innovative approaches to AI algorithms to overcome existing limitations.	1	0.9	1	1	0.8	Retained
16. I promote AI solutions among users and stakeholders to gain support.	1	0.9	1	1	1	Retained
17. I adapt to rapid changes and developments in AI technologies.	1	0.9	1	0.9	0.8	Retained
18. I believe collaboration enhances AI project outcomes.	0	0.3	0.1	0.1	0	Eliminated
19. I evaluate the performance of AI systems and make adjustments when necessary.	0.9	1	1	1	1	Retained
20. I adapt to AI projects by following emerging technological trends and best practices.	1	0.9	1	1	0.8	Retained
21. I present strong and logical arguments to advocate for AI systems.	1	0.9	1	1	1	Retained
22. I collaborate harmoniously with team members from various disciplines.	0.9	1	1	1	1	Retained
23. I attend meetings with other team members during AI projects.	0	0	0.2	0	−0.4	Eliminated
24. I challenge traditional thought patterns in AI projects to generate more creative solutions.	1	1	1	1	1	Retained
25. I produce original and creative solutions using AI algorithms.	0.9	1	0.9	1	0.8	Retained
26. I consider people’s emotional reactions when evaluating the impacts of AI applications.	0.9	1	1	1	0.8	Retained
27. I try to keep up with new AI techniques	0.2	0	0	0.3	−1	Eliminated
28. I respond quickly and effectively to unexpected situations in AI projects.	1	0.9	1	1	0.8	Retained
29. I encourage decision-makers to adopt the benefits of AI algorithms.	0.9	0.9	1	1	1	Retained
30. I enjoy working on projects that include AI components.	0.2	0.3	0	0.3	−0.4	Eliminated
31. I encourage decision-makers to consider the potential drawbacks of AI algorithms.	1	1	0.9	1	1	Retained
32. I actively collaborate with team members to overcome obstacles and achieve common goals in AI projects.	1	0.9	1	1	0.8	Retained
33. I ensure effective communication among team members in AI projects.	0.9	1	1	1	1	Retained
34. I try to stay open-minded about changes in AI systems.	0.2	0.2	0	0.2	−0.2	Eliminated
35. I develop innovative projects using AI tools.	0.9	0.9	0.9	1	1	Retained
36. I establish persuasive communication by emphasizing the societal impacts of AI technologies.	1	1	1	0.9	1	Retained
37. I develop empathy-driven solutions to improve the user experience of AI systems.	1	1	1	1	0.8	Retained

. The primary rationale for exclusion included low relevance scores (e.g., Items 4, 14, 18, 23, 27, 30, and 34 all had I-CVI and CVR values below acceptable thresholds), lack of clarity, or redundancy. For instance, Item 4 (“I am open to learning about artificial intelligence systems”) was removed due to its overly general phrasing and low relevance (CVI < 0.50, CVR = −0.8), which did not reflect a measurable soft skill. Similarly, Item 14 (“I adjust my use of AI tools depending on the task at hand”) was excluded due to low simplicity and ambiguity scores, indicating expert confusion about its operational meaning. Item 18 (“I believe collaboration enhances AI project outcomes”) and Item 34 (“I try to stay open-minded about changes in AI systems”) were regarded conceptually imprecise and lacking behavioral specificity. Out of 37 items initially developed, 30 surpassed the minimum thresholds (I-CVI > 0.78, CVR > 0.62) and were deemed suitable for inclusion. The 7 items with low ratings across multiple indicators were excluded from the final version of the scale. The initial average scores for relevance clarity, ambiguity, and simplicity values were computed as 0.819, 0.795, 0.805, and 0.814, accordingly, with an S-CVI/Ave of 0.777. After the exclusion of seven low-performing items, these scores rose significantly to 0.962, 0.953, 0.977, and 0.967, producing a final S-CVI/Ave of 0.949. The mean CVR also increased from 0.654 to 0.913, which indicates content quality. Consequently, an assessment form with 30 items scored on a Likert scale of five points from “Strongly Disagree” (1) to “Strongly Agree” (5) was created. There were no elements on the scale that were scored in reverse.

2.4. Pilot Study

A pilot study was carried out with 150 individuals to assess the clarity, readability, and initial functionality of the 30-item scale. The main objective was to determine the readability, internal consistency, and item-level performance. Mean, standard deviation, median, and interquartile range (IQR: Q3–Q1) were computed as descriptive statistics, as shown in

Table 3. Results of pilot study.

Item	Mean	Median (Q3–Q1)	Std. Deviation (SD)	Corrected Item-Total Correlation	Cronbach’s Alpha When an Item is Removed
I quickly adapt to different use scenarios of AI systems.	3.63	4 (1)	0.89	0.514	0.843
I synthesize insights from diverse data sources to generate creative outcomes using AI tools.	3.50	4 (1)	0.74	0.452	0.839
I make decisions in AI projects by considering human emotions.	3.95	4 (1)	0.74	0.731	0.869
I use AI analyses to visualize data in creative and effective ways.	3.68	4 (1)	0.81	0.636	0.846
I ensure coordination with other experts during the development and implementation of AI systems.	3.74	4 (1)	0.86	0.660	0.871
I persuade decision-makers of the importance of artificial intelligence algorithms.	3.99	4 (1)	0.75	0.458	0.858
I create synergy within the team by integrating data from various sources.	3.86	4 (1)	0.93	0.526	0.831
I prioritize emotional intelligence when evaluating ethical issues in AI projects.	3.54	4 (1)	0.99	0.531	0.867
I flexibly adjust AI algorithms to changing datasets or requirements.	3.41	4 (1)	0.77	0.549	0.848
I strive to achieve better outcomes by integrating different perspectives in AI projects.	3.73	4 (1)	0.73	0.459	0.847
I understand the emotional needs of AI system users.	3.79	4 (1)	0.79	0.682	0.862
I act with empathy and social responsibility when assessing the societal impacts of AI technologies.	3.46	4 (1)	0.91	0.489	0.842
I develop innovative approaches to AI algorithms to overcome existing limitations.	3.55	4 (1)	0.84	0.585	0.854
I promote AI solutions among users and stakeholders to gain support.	3.72	4 (1)	0.75	0.563	0.843
I adapt to rapid changes and developments in AI technologies	3.53	4 (1)	0.86	0.635	0.841
I evaluate the performance of AI systems and make adjustments when necessary.	3.53	4 (1)	0.81	0.523	0.859
I adapt to AI projects by following emerging technological trends and best practices.	3.60	4 (1)	0.96	0.662	0.837
I present strong and logical arguments to advocate for AI systems.	3.90	4 (1)	0.9	0.615	0.861
I collaborate harmoniously with team members from various disciplines.	3.46	4 (1)	0.82	0.588	0.850
I challenge traditional thought patterns in AI projects to generate more creative solutions.	3.70	4 (1)	0.91	0.582	0.878
I produce original and creative solutions using AI algorithms.	3.67	4 (1)	0.74	0.679	0.863
I consider people’s emotional reactions when evaluating the impacts of AI applications.	3.73	4 (1)	0.99	0.647	0.866
I respond quickly and effectively to unexpected situations in AI projects.	3.52	4 (1)	0.84	0.649	0.879
I encourage decision-makers to adopt the benefits of AI algorithms.	3.76	4 (1)	0.81	0.529	0.864
I encourage decision-makers to consider the potential drawbacks of AI algorithms.	3.79	4 (1)	0.84	0.486	0.857
I actively collaborate with team members to overcome obstacles and achieve common goals in AI projects.	3.54	4 (1)	0.89	0.727	0.847
I ensure effective communication among team members in AI projects.	3.48	4 (1)	0.95	0.649	0.865
I develop innovative projects using AI tools.	3.91	4 (1)	0.78	0.465	0.869
I establish persuasive communication by emphasizing the societal impacts of AI technologies.	3.66	4 (1)	0.84	0.637	0.844
I develop empathy-driven solutions to improve the user experience of AI systems.	3.73	4 (1)	0.97	0.485	0.871

. In addition, corrected item-total correlations and Cronbach’s alpha if the item was removed were computed to determine each item’s contribution to scale reliability. Pilot results showed that item-total correlations ranged from 0.452 to 0.731. Moreover, the Cronbach’s alpha coefficient was found to be 0.902 for all items. However, if one item was removed, then Cronbach’s alpha varied between 0.831 and 0.879, which indicates acceptable internal consistency across items. No major revisions were deemed necessary, and all items were retained for the main study.

These findings confirmed that the scale is prepared for use in primary research. The finalized version of the scale was distributed and administered using an online survey platform, Google Forms. The online survey link was distributed through targeted channels, including LinkedIn professional groups, email lists of AI-related industry networks, and collaborations with HR departments of the selected sectors. Before participating, individuals were informed about the study’s purpose and provided informed consent. The online format allowed the researchers to reach diverse professionals across industries while maintaining data quality and participant anonymity.

2.5. Data Analysis

First, preliminary data screening steps were performed to maintain the integrity and accuracy of the dataset. This process included detecting and addressing any unusual, missing, or extreme values. The dataset contained no missing values. To identify multivariate outliers, the Mahalanobis Distance (MD) measure was applied. Normality assumptions were assessed by examining the absolute values of skewness and kurtosis, with absolute skewness values less than 2 and kurtosis values less than 7 considered indicative of acceptable normal distribution [47,48]. EFA was conducted on the first subset using Principal Axis Factoring (PAF) with oblimin rotation to identify the underlying factor structure [49]. PAF was chosen over Principal Component Analysis (PCA) because it is more suitable for identifying latent variables based on shared variance [50,51,52]. Bartlett’s Test of Sphericity and the Kaiser–Meyer–Olkin (KMO) score were used to assess the feasibility of the data for factor analysis. CFA was performed on second subset using multiple indices including Normed Chi-square (χ²/df), Standardized Root Mean Square Residual (SRMR), Root Mean Square Error of Approximation (RMSEA), Normed Fit Index (NFI), Incremental Fit Index (IFI), Parsimony Goodness-of-Fit Index (PGFI), Parsimony Normed Fit Index (PNFI), Non-Normed Fit Index (NNFI), Comparative Fit Index (CFI), and Incremental Fit Index (IFI), Adjusted Goodness of Fit Index (AGFI), Goodness of Fit Index (GFI) [39,53,54,55,56,57]. The experimental design involved splitting the sample into two independent subsets—one used for exploratory analysis (EFA) and the other for confirmatory analysis (CFA)—to prevent overfitting and to cross-validate the factor structure.

Item analyses were conducted, which involved item-total correlations (≥0.30), Cronbach’s alpha if an item was removed, squared multiple correlations (SMC), and independent samples t-tests comparing the upper and lower 27% groups to evaluate the quality and performance of individual items. Moreover, convergent validity was assessed with Composite Reliability (CR) and Cronbach’s alpha, with values above 0.70 considered acceptable [58]. Average Variance Extracted (AVE) values greater than 0.50 further supported validity [59]. Discriminant validity was tested using Fornell–Larcker criteria and the Heterotrait–Monotrait ratio (HTMT), with HTMT values below 0.85 indicating adequate discriminant validity [59,60]. Reliability analyses included Cronbach’s alpha, Spearman–Brown Split-Half Reliability Coefficient (SBSHRC), and test–retest correlations, with ≥0.70 and split-half coefficients between 0.80 and 0.90 denoting good reliability [61,62]. SPSS 24 was primarily used for data cleaning, descriptive statistics, and preliminary item analysis, while R Studio 4.3.2 was employed for factor analyses (EFA, CFA), calculation of Composite Reliability (CR), Average Variance Extracted (AVE), Fornell–Larcker and HTMT for validity testing, as well as reliability analyses including Spearman–Brown Split-Half and test–retest correlations.

3. Results

3.1. Construct Validity of Artificial Intelligence Soft Skills (AISS) Scale

Construct validity of the AI soft skills (AISS) scale was examined through exploratory factor analysis (EFA) conducted on the development sample (n1 = 340). The KMO value was found to be 0.906, indicating excellent sampling adequacy for factor analysis. Bartlett’s Test of Sphericity was statistically significant (χ²(276) = 5307.783, p < 0.001), confirming that the correlation matrix was not an identity matrix and was therefore appropriate for factor analysis.

Table 4. Description of items, factor loadings, and communality values for the AI soft skills (AISS) scale.

	Item	Factor Loading	Communality
Dimension 1: Persuasion
AISS1	I persuade decision-makers of the importance of artificial intelligence algorithms.	0.743	0.698
AISS2	I present strong and logical arguments to advocate for AI systems.	0.789	0.740
AISS3	I promote AI solutions among users and stakeholders to gain support.	0.680	0.715
AISS4	I encourage decision-makers to adopt the benefits of AI algorithms.	0.677	0.642
AISS5	I encourage decision-makers to consider the potential drawbacks of AI algorithms.	0.893	0.875
AISS6	I establish persuasive communication by emphasizing the societal impacts of AI technologies.	0.756	0.778
Dimension 2: Collaboration
AISS7	I collaborate harmoniously with team members from various disciplines.	0.814	0.735
AISS8	I strive to achieve better outcomes by integrating different perspectives in AI projects.	0.788	0.732
AISS9	I actively collaborate with team members to overcome obstacles and achieve common goals in AI projects.	0.630	0.653
AISS10	I ensure coordination with other experts during the development and implementation of AI systems.	0.621	0.587
AISSl1	I create synergy within the team by integrating data from various sources.	0.808	0.764
Dimension 3: Adaptability
AISS12	I adapt to rapid changes and developments in AI technologies.	0.845	0.781
AISS13	I quickly adapt to different use scenarios of AI systems.	0.822	0.787
AISS14	I flexibly adjust AI algorithms to changing datasets or requirements.	0.769	0.710
AISS15	I respond quickly and effectively to unexpected situations in AI projects.	0.734	0.683
Dimension 4: Emotional Intelligence
AISS16	I understand the emotional needs of AI system users.	0.699	0.616
AISS17	I consider people’s emotional reactions when evaluating the impacts of AI applications.	0.728	0.633
AISS18	I make decisions in AI projects by considering human emotions.	0.774	0.701
AISS19	I develop empathy-driven solutions to improve the user experience of AI systems.	0.834	0.732
AIEAS20	I prioritize emotional intelligence when evaluating ethical issues in AI projects.	0.751	0.685
AIEAS21	I act with empathy and social responsibility when assessing the societal impacts of AI technologies.	0.786	0.743
Dimension 5: Creativity
AISS22	I produce original and creative solutions using AI algorithms.	0.883	0.826
AISS23	I develop innovative projects using AI tools.	0.682	0.649
AISS24	I use AI analyses to visualize data in creative and effective ways	0.710	0.673

shows EFA results with factor loadings and communalities. Items with factor loadings below 0.40 and communalities below 0.40 were excluded from the analysis in accordance with established guidelines [63,64]. A sequential item removal strategy was employed, whereby items were iteratively excluded based on the weakest statistical performance based on factor loading and communalities until a satisfactory factor structure was obtained. In this context, the items “I synthesize insights from diverse data sources to generate creative outcomes using AI tools” (0.123), “I develop innovative approaches to AI algorithms to overcome existing limitations” (0.108), “I evaluate the performance of AI systems and make adjustments when necessary” (0.145), and “I adapt to AI projects by following emerging technological trends and best practices” (0.189) were removed due to low factor loadings. Additionally, the items “I challenge traditional thought patterns in AI projects to generate more creative solutions” and “I ensure effective communication among team members in AI projects” were excluded because of low communalities (0.074 and 0.068, respectively).

Following the removal of these items, the factor structure was determined based on eigenvalues greater than 1 and inspection of the scree plot, which reveals a five-factor structure as seen in

Table 5. Eigenvalues and variance explained for each factor.

Factor	Initial Eigenvalues	% of Variance	Cumulative %
1	11.637	48.49%	48.49%
2	2.231	9.30%	57.79%
3	1.971	8.21%	66.00%
4	1.307	5.45%	71.45%
5	1.193	4.97%	76.42%

Note. This table presents the total variance explained by each factor. Only factors with eigenvalues greater than 1 are listed

and Figure 1. These five factors accounted for a total of 76.42% of the variance. Specifically, Factor 1 explained 48.49% of the variance (eigenvalue = 11.637), followed by Factor 2 (9.30%, eigenvalue = 2.231), Factor 3 (8.21%, eigenvalue = 1.971), Factor 4 (5.45%, eigenvalue = 1.307), and Factor 5 (4.97%, eigenvalue = 1.193). The scree plot indicated a clear inflection point after the fifth factor, supporting the retention of a five-factor structure.

Based on the content of the items, the five dimensions were labeled as follows: persuasion (6 items), collaboration (5 items), adaptability (4 items), emotional intelligence (6 items), and creativity (3 items). Factor loadings varied from 0.621 to 0.893, while communalities were between 0.587 and 0.875. EFA results show that the AISS scale has an interpretable five-factor structure with 24 items. However, through CFA, it is necessary to validate the structure obtained by EFA.

3.2. Item Analysis

The results of the item analysis for the AI soft skills scale are presented in

Table 6. Mean and standard deviation (SD) of items in the AISS scale and comparison results of the t-test for the upper 27% and lower 27%.

Items	Mean (SD)	Median (Q1–Q3)	T	p-Value
Dimension 1: Persuasion
AISS1	3.051 (0.847)	3.000 (3.000–4.000)	15.765	<0.001
AISS2	3.330 (1.011)	4.000 (3.000–4.000)	15.188	<0.001
AISS3	3.892 (0.903)	4.000 (3.000–4.000)	15.356	<0.001
AISS4	3.561 (0.816)	4.000 (3.000–4.000)	18.275	<0.001
AISS5	3.584 (0.936)	4.000 (3.000–4.000)	27.365	<0.001
AISS6	3.357 (0.874)	4.000 (3.000–4.000)	15.465	<0.001
Dimension 2: Collaboration
AISS7	3.153 (1.129)	4.000 (3.000–4.000)	12.474	<0.001
AISS8	2.492 (0.894)	4.000 (3.000–4.000)	20.008	<0.001
AISS9	3.255 (1.110)	4.000 (3.000–4.000)	21.464	<0.001
AISSl10	3.416 (1.070)	4.000 (3.000–4.000)	20.624	<0.001
AISS11	2.960 (0.956)	3.000 (2.000–3.000)	12.061	<0.001
Dimension 3: Adaptability
AISS12	3.355 (0.986)	4.000 (3.000–4.000)	14.275	<0.001
AISS13	3.293 (0.911)	4.000 (3.000–4.000)	13.886	<0.001
AISS14	3.449 (1.019)	4.000 (3.000–4.000)	12.332	<0.001
AISS15	3.390 (0.894)	4.000 (3.000–4.000)	16.415	<0.001
Dimension 4: Emotional Intelligence
AISS16	2.844 (0.853)	3.000 (2.000–3.000)	12.114	<0.001
AISS17	3.372 (0.826)	4.000 (3.000–4.000)	12.127	<0.001
AISS18	3.503 (1.077)	4.000 (3.000–4.000)	15.884	<0.001
AISS19	3.263 (0.876)	4.000 (3.000–4.000)	12.393	<0.001
AISS20	3.535 (1.085)	4.000 (3.000–4.000)	13.950	<0.001
AISS21	3.241 (0.800)	4.000 (3.000–4.000)	16.523	<0.001
Dimension 5: Creativity
AISS22	3.264 (0.825)	4.000 (3.000–4.000)	21.103	<0.001
AISS23	3.180 (0.766)	4.000 (3.000–4.000)	17.178	<0.001
AISS24	3.525 (0.939)	4.000 (3.000–4.000)	15.538	<0.001

and

Table 7. Squared multiple correlation (SMC), corrected item-total correlation (CI-TC), Cronbach’s alpha when the item is removed.

Items	SMC	CI-TC	Cronbach’s Alpha When the Item Is Removed
Dimension 1: Persuasion
AISS1	0.552	0.698	0.917
AISS2	0.623	0.707	0.914
AISS3	0.462	0.614	0.917
AISS4	0.458	0.625	0.919
AISS5	0.797	0.766	0.914
AISS6	0.572	0.681	0.914
Dimension 2: Collaboration
AISS7	0.663	0.670	0.915
AISS8	0.621	0.653	0.916
AISS9	0.397	0.583	0.916
AISS10	0.386	0.541	0.919
AISS11	0.653	0.686	0.918
Dimension 3: Adaptability
AISS12	0.714	0.735	0.913
AISS13	0.676	0.652	0.918
AISS14	0.591	0.634	0.914
AISS15	0.539	0.602	0.914
Dimension 4: Emotional Intelligence
AISS16	0.489	0.542	0.919
AISS17	0.530	0.611	0.917
AISS18	0.599	0.673	0.916
AISS19	0.696	0.672	0.919
AISS20	0.564	0.633	0.912
AISS21	0.618	0.685	0.917
Dimension 5: Creativity
AISS22	0.780	0.793	0.914
AISS23	0.465	0.536	0.918
AISS24	0.712	0.748	0.915

. Table 6 shows the findings of independent samples t-tests comparing the top 27% and bottom 27% groups for all 24 items. The results indicate that each item in the scale significantly differentiates between high and low scorers, with t-values ranging from 12.061 to 27.365, and p-values consistently less than 0.001. These findings support the discrimination power of each item in reflecting individual differences within the construct being measured.

Table 7 demonstrates that each item significantly adds to the scale’s overall consistency, with corrected item-total correlations (CI-TC) ranging from 0.536 to 0.793, which exceeds the cutoff of 0.300 [57]. Similarly, the squared multiple correlations (SMC) ranged from 0.386 to 0.797, further confirming that each item has a strong relationship with the underlying construct. Additionally, Cronbach’s alpha values when each item was removed remained below the total alpha coefficient as 0.921, indicating that none of the items negatively affect internal consistency. The corrected item-total correlations within subdimensions remained above 0.500 in most cases, and the smallest SMC observed (0.386 for AISS10) still met the minimum acceptable level (>0.20), indicating sufficient shared variance with other items in the scale. All items demonstrated acceptable psychometric properties. These results strongly support the internal consistency, reliability, and item-level discrimination capacity of the AISS scale. Each item appears to measure the intended construct effectively and contributes positively to the scale’s validity and reliability. Therefore, it is justified to retain all 24 items in the scale.

3.3. CFA Results

CFA was implemented to evaluate the proposed construct validity using maximum likelihood estimation. The analysis indicated a perfect model fit based on various measures, namely χ²/df = 1.903, and SRMR = 0.035, all falling within excellent fit thresholds. Measures such as CFI (0.975), TLI (0.972), and IFI (0.975) also pointed to a strong model fit, exceeding the recommended cutoff of 0.950 for a perfect fit. GFI (0.940), AGFI (0.947), NFI (0.949), PNFI (0.833), PGFI (0.823), and RMSEA (0.052) fell within the adequate fit interval as seen in

Table 8. CFA fit measures and evaluation criteria.

Measure	Value	Perfect Fit Interval	Adequate Fit Interval
χ2	460.526	0 ≤ χ2 ≤ 2df	2df < χ2 ≤ 3df (df = 242)
χ2/df	1.903	0 ≤ χ2/df ≤ 2	2 < χ2/df ≤ 3
GFI	0.940	0.95 ≤ GFI ≤ 1.00	0.90 ≤ GFI < 0.95
AGFI	0.947	0.90 ≤ AGFI ≤ 1.00	0.85 ≤ AGFI < 0.90
NFI	0.949	0.95 ≤ NFI ≤ 1.00	0.90 ≤ NFI < 0.95
PNFI	0.833	0.95 ≤ PNFI ≤ 1.00	0.50 ≤ PNFI ≤ 0.95
PGFI	0.823	0.95 ≤ PGFI ≤ 1.00	0.50 ≤ PGFI ≤ 0.95
TLI (NNFI)	0.972	0.97 ≤ TLI ≤ 1.00	0.95 ≤ TLI < 0.97
IFI	0.975	0.95 ≤ IFI ≤ 1.00	0.90 ≤ IFI < 0.95
CFI	0.975	0.97 ≤ CFI ≤ 1.00	0.95 ≤ CFI < 0.97
RMSEA	0.052	0 ≤ RMSEA ≤ 0.05	0.05 < RMSEA ≤ 0.08
SRMR	0.035	0 ≤ SRMR ≤ 0.05	0.05 < SRMR ≤ 0.10

. Overall, these results provide a satisfactory and powerful basis for the factor structure, which confirms the model’s suitability and construct validity.

3.4. Reliability Analysis Results

The internal consistency of the developed AISS scale was found to be strong, as evidenced by a Cronbach’s alpha coefficient of 0.921 for all 24 items. When the five dimensions of the AISS scale were examined separately, each demonstrated high internal consistency: persuasion (six items, Cronbach’s alpha = 0.875), collaboration (five items, Cronbach’s alpha = 0.829), adaptability (four items, Cronbach’s alpha = 0.857), emotional intelligence (six items, Cronbach’s alpha = 0.870), and creativity (three items, Cronbach’s alpha = 0.804), confirming the contribution of each item to the internal consistency of the scale.

In addition, the SBSHRC for the AISS scale was found to be 0.771, indicating a satisfactory level of reliability across item halves. Pearson’s r and the Intraclass Correlation Coefficient (ICC) were computed using the scores of individuals who took the scale twice during a two-week period in order to evaluate the scale’s test–retest reliability. The results showed excellent stability of the AISS scale over time, with Pearson’s r = 0.972 (p < 0.001) and ICC = 0.983 (p < 0.001), demonstrating the temporal consistency of the scores.

3.5. Convergent Validity Results

It is commonly accepted that each construct’s Average Variance Extracted (AVE) must be at least 0.50, and its Composite Reliability (CR) and Cronbach’s alpha values need to be more than 0.70 to demonstrate convergent validity [58,59]. As presented in

Table 9. Composite reliability, Cronbach’s alpha, and AVE values for convergent validity.

Dimension	Number of Items	Composite Reliability	Cronbach’s Alpha	AVE
Persuasion	6	0.890	0.875	0.577
Collaboration	5	0.855	0.829	0.544
Adaptability	4	0.872	0.857	0.630
Emotional Intelligence	6	0.893	0.870	0.583
Creativity	3	0.848	0.804	0.652

, all five dimensions met these recommended thresholds. Specifically, “persuasion” (CR = 0.890, Cronbach’s alpha = 0.875, AVE = 0.577), “collaboration” (CR = 0.855, Cronbach’s alpha = 0.829, AVE = 0.544), “adaptability” (CR = 0.872, Cronbach’s alpha = 0.857, AVE = 0.630), “emotional intelligence” (CR = 0.893, Cronbach’s alpha = 0.870, AVE = 0.583), and “creativity” (CR = 0.848, Cronbach’s alpha = 0.804, AVE = 0.652) demonstrated sufficient internal consistency and convergent validity.

3.6. Discriminant Validity Results

To assess discriminant validity, both the Fornell–Larcker criterion [56] and Henseler’s Heterotrait–Monotrait ratio (HTMT) [60] were applied. According to the Fornell–Larcker criterion, discriminant validity is considered acceptable if the square root of the Average Variance Extracted (AVE) for each construct is greater than its correlation with other constructs. As shown in

Table 10. Fornell–Larcker criteria results for discriminant validity.

Factor	Fornell–Larcker Criteria
	Persuasion	Collaboration	Adaptability	Emotional Intelligence	Creativity
Persuasion	0.760
Collaboration	0.652	0.738
Adaptability	0.553	0.604	0.794
Emotional Intelligence	0.627	0.656	0.583	0.764
Creativity	0.507	0.520	0.673	0.541	0.807

, the diagonal values representing the square root of AVE for each factor (e.g., persuasion = 0.760, collaboration = 0.738, adaptability = 0.794, emotional intelligence = 0.764, creativity = 0.807) were all greater than the inter-construct correlations, which indicates strong discriminant validity.

Additionally, the HTMT was used as a complementary measure. HTMT values below the recommended threshold of 0.85 suggest that the constructs are empirically distinct [60]. In this study, all HTMT values (ranging from 0.531 to 0.692) were well below this cutoff, further confirming the discriminant validity of the AISS scale (see

Table 11. Henseler’s Heterotrait–Monotrait Ratio (HTMT) criterion for discriminant validity.

Factor	HTMT Criteria
	Persuasion	Collaboration	Adaptability	Emotional Intelligence	Creativity
Persuasion	-
Collaboration	0.678	-
Adaptability	0.576	0.627	-
Emotional Intelligence	0.638	0.643	0.601	-
Creativity	0.531	0.553	0.692	0.560	-

). These results show that the constructs within the scale demonstrate sufficient discriminant validity.

4. Discussion

Artificial intelligence offers digital solutions for attracting and maintaining the customer base [65] and provides a competitive advantage [66] by facilitating the customers’ product [67] and service offerings [68], especially in smart cities. The capacity for AI for improving company productivity is demonstrated by a number of applications, including voice recognition, driverless cars, speech/text producing systems, maintenance forecasting, and trained neural networks to optimize energy use [69]. Also, with AI algorithms, human error is reduced, audience data is efficient, and display advertising is scaled [70]. Moreover, one AI-powered technology that helps in tracking in-store visits and connecting photos to social media accounts is face recognition software [70]. These advanced technologies provide a new degree of personalized user experience by sending each visitor real-time discount offers and greeting messages when combined with AI-powered smart alerts [71]. As a result, in marketing, AI applications have advantages, such as reduced risk, increased speed, improved customer happiness, increased revenue, etc. [72]. Moreover, AI-powered automation, personalized customer experiences, and real-time data-driven decision-making contribute to improved economic and financial performance [73]. Furthermore, AI reduces costs and increases profits [74], improving productivity [75], forecasting [76] strategies on pricing [77], etc. Consequently, the results of most scientific studies have shown that artificial intelligence technologies/tools/systems contribute to increasing businesses’ profit margins by enhancing operational efficiency and thereby reducing costs. It is of great importance for employees to have AI soft skills, especially in smart cities, which will be an important factor in the future. In addition, since customer demands can be analyzed more accurately through AI tools, more targeted and precise marketing strategies are developed, which in turn lead to increased revenue for businesses.

In order to maximize the effectiveness of artificial intelligence technologies and tools in businesses, it has become essential that both employees and decision-making managers possess AI skills. These AI skills are broadly categorized into two main groups: hard skills (knowledge, experience, specialization, professionalism, and the ability to perform job tasks, etc.) and soft skills (collaboration, adaptability, emotional intelligence, etc.). However, there is a growing expectation across sectors that staff should possess advanced AI soft skills [78], and 86% of employers agree on the importance of AI soft skills for success in the workplace [79]. Therefore, it is of great importance across all sectors that individuals possess not only technical knowledge related to AI, but also soft skills [80,81]. AI soft skills contribute to increased productivity and efficiency in the workplace and help reduce working time [82]. A review of the literature shows that various dimensions of AI soft skills have been addressed in previous studies. In this study, persuasion [30], collaboration [31,32], adaptability [33,34], emotional intelligence [31,35,36], and creativity [33,37] have been considered as AI soft skills, as these specific factors are critical and distinctive for businesses to reach their desired profit margins, especially within the context of smart city development.

The first AI soft skill, persuasion, refers to an individual’s ability to influence the attitudes, beliefs, or behaviors of others through logical arguments and emotional engagement. Therefore, persuasion holds strategic importance in strengthening customer relationships, increasing sales, and enabling effective intra-team collaboration within organizations [83]. About SDG 11, persuasion is vital for promoting sustainable practices and community engagement within smart cities. The second soft skill addressed in this study is collaboration (cooperation), which is the ability of individuals to work together effectively by sharing knowledge, skills, and resources in harmony to achieve common goals. Collaboration is of great importance in organizations, as it enhances team efficiency, supports the development of innovative solutions, and helps achieve organizational objectives more effectively [84]. Collaboration supports SDG 4 by fostering cooperative learning environments and SDG 11 through enabling coordinated efforts in sustainable urban development. The third factor, adaptability, is defined as the ability of an individual to respond quickly and effectively to changing conditions, new information, or unexpected situations. This skill is critically important for businesses to maintain competitive strength in the face of rapid technological and structural changes such as those brought by AI [85]. Adaptability aligns with SDG 8 by enhancing workforce resilience and innovation in evolving job markets, and supports SDG 11 by enabling smart cities to respond flexibly to urban challenges. Another key factor, emotional intelligence, describes a person’s capacity to recognize, interpret, and control both their own and other people’s emotions. This competency forms the basis of healthy communication, leadership, and conflict management in work environments, thereby enhancing organizational success and employee satisfaction [86,87]. Emotional intelligence contributes to SDG 4 by supporting inclusive education and social skills development, and to SDG 11 by promoting social cohesion in urban communities. The final soft skill addressed in this study is creativity, which is the ability to generate original ideas, solve problems in unconventional ways, and present innovative approaches. This skill is a fundamental element that supports product, service, and process innovation, enabling businesses to gain a competitive advantage [88]. Creativity fosters lifelong learning and innovation in line with SDG 4, drives sustainable economic growth supporting SDG 8, and encourages innovative solutions crucial for smart cities under SDG 11. In this context, the scale developed in this study includes these AI soft skills, aiming to provide organizations with a tool to identify these skills in their employees or to assess existing strengths within the scope of these abilities. Moreover, the developed artificial intelligence soft skills scale contributes to the socio-technical dimensions of smart city applications, especially with the dimensions of persuasion, cooperation, adaptation, emotional intelligence, and creativity. In this context, the conceptual framework offered by the scale enables the evaluation of the human-centered and interaction-oriented functions of artificial intelligence systems in sustainability-oriented decision-making processes, supporting the development of holistic and sustainable solutions in smart cities.

Consequently, as artificial intelligence technologies are increasingly integrated into the business processes of all sectors today, it has become essential for both current staff and job applicants in organizations to possess AI skills, especially soft skills, to ensure effective and positive digital transformation. However, although there are some existing scales in the literature aiming to measure AI literacy [23], AI self-efficacy [25], and AI attitude [26,27], there is still a lack of a comprehensive and practical scale that can assess sector-specific AI soft skills. Therefore, this study aims to address this gap in the existing literature by developing a new AI skills scale designed to evaluate the AI soft skills of individuals who actively use AI technologies across various sectors. While AI literacy, self-efficacy, and attitudes measure individuals’ knowledge, confidence, and emotional approach to AI, AI soft skills represent the ability to effectively use this technology through social and cognitive skills such as collaboration, empathy, adaptability, and persuasion. In this respect, AI soft skills encompass not only cognitive or affective tendencies but also an individual’s capacity to use AI in a human-centered and functional manner. For businesses, employees equipped with these skills can more effectively manage digital transformation processes and directly contribute to corporate performance by strengthening customer relationships. Moreover, with AI becoming indispensable for all sectors in the future, especially smart cities, this scale will enable individuals to objectively assess their own AI soft skills, while also allowing employers to evaluate candidates’ AI soft skills during recruitment processes based on this standardized framework.

5. Conclusions

The systematic identification and development of AI soft skills have become a strategic imperative for achieving the sustainability goals of smart cities. These skills not only facilitate individuals’ social adaptation, but also enable businesses to access a digitally competent workforce, thereby supporting the sustainability of human-centered technological transformation. At the individual level, the advancement of these skills enhances employability in an increasingly digital and competitive labor market, promoting access to qualified positions and the potential for higher income levels. In particular, individuals with well-developed AI-driven soft skills are emerging as strategic assets in both local and global business transformation processes. For this purpose, this study aimed to develop a scale that can systematically measure the artificial-intelligence-based soft skills of individuals. Following comprehensive validity, reliability, and item analyses, the artificial intelligence soft skills (AISS) scale, with the dimensions of persuasion, collaboration, adaptability, emotional intelligence, and creativity, has been demonstrated to exhibit strong psychometric properties, indicating its suitability for application in future academic and empirical research.

This study’s primary theoretical contribution lies in introducing a measurable framework for AI-driven soft skills, a concept that has gained increasing importance yet remains underexplored in the literature. By operationalizing these skills into a psychometrically sound scale, the research fills a critical gap and lays the groundwork for further conceptual and empirical development in the field. Moreover, the determination and evaluation of AI soft skills with the scale developed in this study will make a significant contribution to the evaluation of human resources for innovative business models, digital transformation, and the infrastructure needs of smart cities. From a managerial perspective, the scale provides a practical tool for organizations to assess, develop, and integrate AI-related soft skills into their workforce strategies, helping improve employee performance, customer satisfaction, and long-term competitiveness. It will also contribute to the determination of individuals’ AI soft skills, which are one of the main elements for the success of smart cities, and their integration into future policies. In addition, this developed scale enables the determination of the social and cognitive skills needed in the human-centered and participatory governance structures of smart cities and contributes to the design of policies and training programs to eliminate the deficiencies of individuals in these areas. Thus, it becomes possible to create qualified human resources that support sustainable development in smart cities and for these individuals to take an active part in the labor market. Furthermore, the developed artificial intelligence soft skills (AISS) scale can be strategically applied by key stakeholders, including municipal authorities, urban planners, policymakers, educational institutions, and private sector organizations involved in smart city development, to systematically identify and develop the AI soft skills necessary for advancing the SDGs. Specifically, municipal governments and policymakers can utilize the scale to assess workforce capabilities and design targeted training programs aligned with SDG 4, SDG 8, and SDG 11. Additionally, educational institutions may integrate the scale into their curricula to prepare future professionals with these essential skills. Moreover, private sector companies can employ the scale to enhance employee skills, fostering innovation and sustainability within smart city ecosystems. By enabling coordinated efforts among diverse stakeholders, the AISS scale serves as a practical tool to operationalize and accelerate the achievement of SDGs within the context of smart cities.

Among the limitations of this study are the restricted scope of the selected AI soft skills dimensions and the confinement of the scale items within a specific conceptual framework. Additionally, the relatively homogeneous structure of the sample used in the scale development process may limit the generalizability of the findings to different demographic or cultural groups. For future research, it is recommended to revalidate the scale across more diverse populations in terms of gender and age, explore additional skill dimensions relevant to evolving AI contexts, and examine the longitudinal impact of these skills on career progression and organizational performance. Also, with the scale developed in the future, a road plan can be prepared by collecting the necessary data and making the necessary analyses to determine the necessary skills of individuals in the labor market transformed by artificial intelligence, to increase workforce qualifications, and to develop human–machine cooperation. Furthermore, as the developed scale was originally constructed in English, its applicability is currently limited to English-speaking populations. To enhance the scale’s cross-cultural relevance and to support the global implementation of smart city initiatives, future research should focus on translating and validating the scale in multiple languages and cultural settings.