Sakilé K. Camara
California State University Northridge
Abdul Rehmen
University of Education Jauharabad Campus
Suggested Citation:
Camara, S. K., & Rehmen, A. (2026). Student reactions to AI-generated feedback in simulated employment interview sessions. Utah Journal of Communication, 4(1), 1218. https://doi.org/10.5281/zenodo.20741630
Abstract
Artificial intelligence (AI) is increasingly used to evaluate communication performance during employment interviews, yet little is known about how candidates perceive algorithmic feedback. This pilot study examined student reactions to AI-generated interview evaluations produced by the ™Bravofolio Interview Plus platform. Using a convergent parallel mixed methods design, data were collected from 34 mock interview sessions completed by 15 undergraduate students. Findings revealed substantial disagreement with AI-generated confidence scores and concerns regarding transparency, trust, and accuracy. Qualitative analysis identified issues related to eye-contact tracking, environmental limitations, filler-word penalization, and the absence of human interaction. Recommendations for improving AI interview assessment systems are discussed.
Keywords: artificial intelligence, communication confidence, employment interviews, algorithmic assessment
Communication confidence is a central construct in employment interviewing (Tompsett et al., 2017) and hiring managers interpret observable behaviors (i.e., eye contact, vocal steadiness, message organization, posture, and environmental professionalism) as signals of readiness for workplace participation (Argyle, 2013). In high-stakes interview settings, even subtle variations in delivery, gaze stability, or structural coherence can influence evaluative judgments and hiring outcomes (Martín-Raugh et al., 2023).
The mechanisms through which these judgments are formed are rapidly changing. Over the past decade, employment interviewing has shifted from exclusively human-mediated encounters to hybrid and fully automated systems in which artificial intelligence (AI) evaluates recorded responses. Research suggests that as many as 86% of employers reported using technology-driven job interviews by 2022, with many adopting automated video interviews as a preferred screening method (Jaser, et al., 2022). As (AI) continues to automate business functions (Barker, 2025), communication studies programs face pressure to rethink what communication competence means and how to prepare students for algorithmically mediated evaluation contexts.
This raises a central empirical question: Can algorithms evaluate communication confidence in ways that align with human self-perception? While machine-learning systems can quantify signals such as filler frequency, gaze alignment, and vocal intensity, confidence as a psychological state is internal. Divergence between algorithmic outputs and human experience may signal construct validity concerns, interpretive gaps, or design limitations in AI-mediated systems. The present pilot addresses this question by examining student reactions to AI-generated feedback from ™Bravofolio’s mock interview feature, which evaluates performance across eight communication dimensions: (1) facial engagement units, (2) eye stability, (3) vocal delivery, (4) gestures, (5) posture, (6) space and framing, (7) background and lighting, and (8) message structure, and provides confidence-related feedback derived from these signals. Students then assessed whether AI feedback aligned with their own perceptions of performance.
By analyzing patterns of convergence, misalignment, and user-identified inaccuracies, this pilot contributes to emerging scholarship on algorithmic evaluation, communicative competence, and the human perception of confidence in digitally mediated interviews.
Literature Review
Mock Interviewing and Career Readiness
Simulated mock interviewing is widely used in university career centers to prepare students for the workforce and enhance self-confidence (Dey & Cruzvergara, 2014; Hansen et al., 2009; Rowell & Mihuta, 2016). Research documents a consistent positive relationship between structured interview practice and increased student confidence (Quinlan & Renner, 2005; Tross & Maurer, 2008) consistent with Bandura’s (1997) self-efficacy theory, which identifies mastery experiences as a primary source of confidence development. Recent studies confirm that practice-based career interventions enhance both perceived employability and confidence (Jackson & Wilton, 2017; Okay-Somerville & Scholarios, 2017). Hudak et al. (2019) found statistically significant increases in self-reported confidence following technology-supported interview practice in introductory communication courses, suggesting that the design of AI feedback tools must preserve rather than undermine self-efficacy.
Automated video interview platforms modernize this tradition by providing immediate, multimodal feedback (Campion et al., 2016). These systems leverage computer vision and natural language processing to analyze facial expressions, vocal tone, physical posture, and gaze patterns producing composite behavioral profiles of applicants. However, as these platforms transition from experimental to enterprise tools, concerns have emerged regarding how candidates experience and trust algorithmic evaluation.
Nonverbal Communication in Employment Interviews
A substantial body of research affirms that nonverbal behaviors are critical to interview outcomes. A meta-analysis spanning more than 70 years found that professional appearance, eye contact, and head movement were among the strongest predictors of interview ratings (Martín-Raugh et al., 2023), consistent with Argyle’s (2013) observation that bodily communication serves as a primary channel for forming interpersonal impressions. Gifford et al. (1985) demonstrated that social skill was more accurately inferred through nonverbal cues than verbal content alone, yet the mapping between behavioral signals and internal states is far from direct. This gap is amplified in digitally mediated contexts where camera quality, environmental lighting, and physical framing introduce variables that affect the legibility of nonverbal signals (Bailenson, 2021).
Automated Video Interviews and Algorithmic Assessment
As automated video interview tools have entered mainstream recruitment, documented concerns have grown regarding algorithmic transparency and candidate experience (Suen & Hung, 2023; Jaser et al., 2022). Survey evidence suggests that only 26% of job candidates trust AI to evaluate them fairly (Langer et al., 2017), reflecting concerns about opacity, lack of contextual understanding, and perceived dehumanization. Research grounded in organizational justice theory indicates that applicants evaluate interview fairness not only on outcomes but on the transparency and relational quality for the assessment itself (Xu et al., 2025; Chan, 2000). When AI systems fail to provide comprehensible explanations for scores, candidates tend to perceive the evaluation as arbitrary and procedurally unjust (Suen & Hung, 2023).
AI-driven interview tools have drawn scrutiny for potential bias. Chen (2023) and others have documented that algorithmic bias in recruitment may manifest along lines of race, gender, and socioeconomic status, often rooted in unrepresentative training data. The U.S. Equal Employment Opportunity Commission (EEOC) has launched initiatives to monitor AI bias in employment screening, recognizing that algorithms measuring speech patterns and facial expressions risk disadvantaging individuals with disabilities, non-native accents, or neurodiverse communication styles (Ajunwa, 2023).
Zoom Fatigue and the Camera-Screen Offset Problem
One structural constraint in AI-mediated interviewing is the physical separation between the camera lens and the screen display (Suen & Hung, 2025). In face-to-face interaction, mutual gaze is natural; in digital environments, candidate’s who look at the screen to read prompts appear to be looking downward from the camera’s perspective, potentially triggering penalty scoring from gaze-tracking algorithms. Bailenson (2021) articulated the broader psychological cost of this constraint arguing that forced continuous webcam gaze violates social norms regarding interpersonal distance. A large-scale study of 9,787 participants confirmed that managing nonverbal cues in video-based communication predicted Zoom fatigue, with women disproportionately affected (Fauville et al., 2021).
Cognitive Load Theory and Interview Performance
Cognitive load theory (CSL; Sweller, 1988) provides a useful framework for understanding candidate behavior in AI-evaluated interviews. CLT distinguishes between intrinsic load (task complexity), extraneous load (demands imposed by poor task design), and germane load (effort directed toward schema construction). When responding to complex interview questions, candidates allocate substantial cognitive resources to semantic processing and verbal organization, leaving fewer resources for impression management behaviors such as maintaining posture, adjusting vocal tone, and sustaining camera gaze. AI interview systems, however, evaluate behavioral data streams without contextualizing them against the cognitive demands of the question. This misalignment between AI assessment logic and human cognitive architecture is a central source of the divergence examined in the present study. These considerations led to the following research questions:
RQ1: To what extent do student validators agree or disagree with AI assessment of interviewing performance?
RQ2: How can AI interviewing platforms be refined to better align with the user experience?
Method
Research Design
This study employed a convergent parallel mixed methods design (Creswell & Plano Clark, 2018), collecting quantitative Likert-scale data and qualitative open-ended responses concurrently, analyzing them separately, and integrating findings at the interpretation stage. This design was selected because the dual nature of the data, numerical ratings and narrative feedback, required approaches capable of capturing both the distributional patterns experiential texture of student perception, enabling cross-validation and triangulation (Creswell & Plano Clark, 2018). The research questions were not designed to generalize to broader populations, but rather to inform evidence-based refinement of an AI evaluation tool through structured user validation.
Context and Setting
The study used an anonymized dataset from the ™Bravofolio “Interview Plus,” which integrates two AI tools. First, MediaPipe, Google’s machine learning framework, analyzes video recordings to detect and quantify facial expressions, body posture, eye gaze, and head positioning. Second, OpenAI’s NLP API processes these measurements to generate evidence-based communication confidence scores. This hybrid approach enables scalable analysis of communication competencies traditionally required trained human evaluators.
The methodological workflow (Figure 1) begins with an Accessibility Status check, proceeds through Target Industry and Question Difficulty selection, an optional Warmup, Calibration, and Live Recording phases, and concludes with simultaneous Self-Rating and data Input. The final Evaluation Output stage integrates algorithmic scoring with self-assessment data to generate post-performance feedback across communication dimensions (Figure 2).
Figure 1
Interview Plus Process Flow
Figure 2
Results Provided to Users
Participants
Participants were undergraduate students enrolled in one section of a Communication Internship course during Spring 2026 at a mid-size university. The course primarily enrolls junior and seniors completing applied professional communication internships. Thirty-five students were enrolled in the course at the time of implementation.
Mock interview sessions are required as a course component for monitoring career-readiness as a learning outcome. No incentives were provided for completing course-required components. A 25% legacy product discount was offered in connection with platform access, independent of participation in optional recalibration feedback. Students completed an agreement authorizing de-identified data use for research and AI recalibration purposes. In accordance with institutional policy, this activity qualified for exemption from Institutional Review Board review under the category of normal educational practices.
Data Collection Instruments
Two instruments embedded within the Interview Plus platform. The first was a five-item quantitative feedback survey administered after each session, measured on a seven-point Likert scale (1=Strongly Disagree, 7=Strongly Agree). Items assessed: (Q1) perceived accuracy of the AI confidence score, (Q2) future trust in the AI system, (Q3) perceived utility of the feedback for skill improvement, (Q4) transparency and comprehensibility of the score, and (Q5) willingness to engage with the technology again.
The second instrument was a single open-ended qualitative prompt (Q6): “What did the AI miss or capture well?” This item was designed to elicit narrative feedback regarding specific algorithmic behaviors, inaccuracies, and experiential reactions. Responses were analyzed thematically to identify recurring patterns of algorithmic failure and user perception.
Data Analysis Procedures
Quantitative data were analyzed using descriptive statistics, including frequency distributions across the seven-point scale. Responses were grouped into disagreement ratings (1-3), neutral (rating 4), and agreement ratings (5-7) categories to facilitate pattern identification. Cross-tabulation by question difficulty (Easy, Medium, and Hard) examined whether task complexity moderated patterns of algorithmic trust and satisfaction.
Qualitative data were analyzed using inductive thematic coding, consistent with qualitative content analysis procedures (Creswell & Creswell, 2017). Two analytical passes were conducted: an initial open coding phase and a focused coding phase to collapse and define recurring themes. Emergent themes were validated against the quantitative distribution data to identify convergence and divergence, consistent with the integration logic of the convergent parallel design (Creswell & Plano Clark, 2018). Four primary themes emerged: (1) optical constraints related to camera-screen offset; (2) computer vision vulnerabilities related to lighting, and framing; (3) linguistic constraints related to filler word penalization; and (4) human alienation reflecting relational and emotional dimensions of automated assessment.
Trustworthiness and Limitations
Several considerations bear on the trustworthiness of these findings. The pilot dataset (N = 34 sessions, 15 unique users) limits statistical power and generalizability. The sample was drawn from a single course section at one institution. For the applied goal of AI recalibration, however, the depth and specificity of the qualitative data provided actionable diagnostic insights.
Potential response bias is acknowledged, as participants completed surveys as a required course activity. Social desirability pressures may have influenced some responses, though the directional skew toward strong disagreement suggests authentic reporting. Researcher positionality is also noted: The first author is the developer of the Interview Plus platform, and the other is an AI Prompt engineer hired by the first author’s company. This potential conflict of interest is mitigated by the commitment to reporting all feedback, including negative responses, without selective omission.
Results and Findings
The analysis covers 34 interview practice sessions generated by 15 unique users, distributed across three difficulty tiers: Easy (n = 14; 41.2%), Medium (n = 12; 35.3%), and Hard (n = 8; 23.5%). Quantitative Likert-scale data and qualitative open-ended responses are examined jointly to address RQ1, with qualitative themes informing the recalibration discussion addressing RQ2.
Quantitative Distribution of Student Sentiment
Table 1 presents the aggregate distribution of student responses across five evaluation dimensions. The overall pattern shows a pronounced skew toward disagreement, particularly on accuracy, trustworthiness, and transparency.
Table 1:
Distribution of Responses Across Core Evaluation Metrics
| Assessment Metric | SD | D | SwD | Neutral | SwA | A | SA |
| Q1: AI score accuracy | 12 | 6 | 5 | 5 | 2 | 3 | 1 |
| Q2: Future trust in AI | 13 | 2 | 5 | 6 | 4 | 3 | 1 |
| Q3: Skill improvement utility | 8 | 3 | 2 | 6 | 9 | 5 | 1 |
| Q4: Score transparency | 11 | 3 | 6 | 6 | 4 | 3 | 1 |
| Q5: Willingness to re-engage | 8 | 1 | 0 | 12 | 5 | 4 | 1 |
Note. SD = Strongly Disagree; D = Disagree; SwD = Somewhat Disagree; N = Neutral; SwA = Somewhat Agree; A = Agree; SA = Strongly Agree. Disagree = ratings 1–3; Neutral = rating 4; Agree = ratings 5-7: n=33 (one missing response).
For the primary alignment metric (Q1), approximately 67.6% of responses (23 of 34) fell into disagreement categories. Disaggregated by difficulty (Table 2), this pattern was most pronounced in Medium sessions (75.0%, mean = 2.25), compared to Easy (64.3%, mean = 3.21) and Hard sessions (62.5%, mean =2.75), suggesting that mid-range task demands may produce the sharpest mismatch between algorithmic and self-reported performance, a pattern further examined below.
Trust in the platform’s predictive validity (Q2) followed a similar trajectory: 58.8% of respondents expressed distrust overall, with Medium sessions again producing the highest rate (75.0%, mean = 2.58) compared to Easy (42.9%, mean = 3.50) and Hard (62.5%, mean = 2.62). These patterns are consistent with research documenting fragile trust in automated interview systems (Langer et al., 2019).
Table 2:
Distribution of Responses by Question Difficulty Level (Easy, Medium, Hard)
| Assessment Metric | Easy (n=14) | M | Medium (n=12) | M | Hard (n=8) | M | ||||||
| D% | N% | A% | Mean | D% | N% | A% | Mean | D% | N% | A% | Mean | |
| Q1: AI score accuracy | 64% | 14% | 21% | 3.21 | 75% | 17% | 8% | 2.25 | 63% | 13% | 25% | 2.75 |
| Q2: Future trust in AI | 43% | 29% | 29% | 3.50 | 75% | 8% | 17% | 2.58 | 63% | 13% | 25% | 2.62 |
| Q3: Skill improvement utility | 21% | 21% | 57% | 4.43 | 58% | 17% | 25% | 3.00 | 50% | 13% | 38% | 3.12 |
| Q4: Score transparency | 29% | 21% | 50% | 4.21 | 67% | 17% | 17% | 2.83 | 63% | 13% | 25% | 2.88 |
| Q5: Willingness to re-engage | 15% | 46% | 38% | 4.23 | 50% | 17% | 33% | 3.08 | 38% | 38% | 25% | 3.25 |
Note. D% = percentage disagreeing (ratings 1–3); N% = neutral (rating 4); A% = agreeing (ratings 5–7); M = mean on 7-point scale. Bold values indicate the highest disagreement and agreement rates within each row. Q5 Easy: n = 13 (one missing response).
Score transparency (Q4) showed the most acute dissatisfaction in Medium sessions (66.7% disagreeing, mean = 2.83), reflecting the “black box” problem in AI-identified in AI-driven recruitment (Chen, 2023).
The Medium Difficulty Paradox
A theoretically notable pattern is the consistently highest rejection of AI scoring in Medium, rather than Har, sessions across nearly all metrics. CLT offers a partial explanation: Under Hard conditions, candidates may attribute degraded nonverbal performance to question difficulty, providing a cognitive account for the AI’s negative scores that softens rejection. Medium difficulty questions, by contrast, impose sufficient cognitive demand to degrade nonverbal performance but not enough for candidates to attribute that degradation to task complexity. The result is a failure mode in which neither the candidate nor the algorithm can account for the misalignment, producing frustration and minimum trust. This interpretation should be treated as tentative given the small sample, but it warrants further investigation in larger-scale studies.
Algorithmic Output Versus Human Self-Efficacy
A recurring pattern in the data is incongruence between the AI’s behavioral scores and candidates’ self-reported confidence. One student who reported feeling “Very confident” across three sessions received AI scores of “Low,” “Needs Improvement,” and “Major Improvements Needed.” A similar pattern appeared with another student who reported high confidence but received consistently low AI ratings, ultimately dismissing the tool. Overconfident User Gap, defined as sessions where users reported high confidence while receiving low AI scores was most prevalent in “Medium” sessions (41.75), compared to Easy (28.6%) and Hard (25.0%0, further supporting the “Medium” difficulty interpretation.
This dynamic is theoretically consistent with Bandura’s (2001) social cognitive theory: when evaluative feedback contradicts a learner’s existing sense of competence without adequate explanation, it risks eroding self-efficacy and reducing motivation to engage. Conversely, alignment between algorithmic and human assessment occurred most frequently when candidates reported low confidence, raising the possibility that apparent satisfaction with AI output may in some cases reflect confirmation rather than objective algorithmic accuracy. This observation warrants caution given the pilot’s scope.
Deconstruction of Algorithmic Sensing Failures
Optical Constraints. Participants consistently reported that despite looking directly at their camera, the system scored them as lacking eye contact. This failure reflects the camera-screen offset problem: when candidates look at the screen to read prompts, the camera captures their gaze as pointing downward. The AI’s geometric gaze-tracking penalizes natural screen-reading as disengagement (Bailenson, 20210.
Computer Vision Vulnerabilities. Participants reported that the AI missed hand gestures and incorrectly penalized posture. One student hypothesized that dark clothing against a poorly illuminated background made it difficult for the system to track her body, consistent with documented optical tracking vulnerabilities (Ajunwa, 2023). Spatial constraints of a standard webcam’s field of view also inhibit natural gesture expression, as candidates must sit close to screens to ensure facial capture, truncating arms from the frame.
Linguistic Constraints. Several participants described bewilderment at the penalization of filler words. Decades of psycholinguistic research indicate that “um” “uh” and “like” are functional communicative devices, signals of cognitive processing, turn-taking management, and upcoming syntactic complexity, not defects in professional communication (Fox Tree, 2003, 2007). Similarly, users expressed confusion at vocal tone feedback, noting disconnects between their perceived projection and the algorithm’s acoustic assessment. The algorithm analyzes raw acoustic metrics but lacks the semantic comprehension to differentiate expressive delivery from erratic speech.
Human Alienation. Beyond technical issues, the data reveal emotional resistance to the automated format. One student observed that the algorithm fail to account for her being an actual person with real feelings, noting that recording oneself differs fundamentally from speaking to a human face to face. This reflects the relational deficit inherent in automated video assessments: the traditional interview is bidirectional social interaction providing micro-affirmations (nods, smiles) that allow candidates to self-regulate. Automated platforms strip away this dynamic, consistent with organizational justice research linking perceived process dignity to evaluation fairness (Suen & Hung, 2023; Jaser et al., 2022).
Discussion
The qualitative data reveal a consistent pattern of false negative assessments in which rigid tracking parameters penalized natural human behavior or environmental limitations. Discrepancies were primarily identified in eye stability, gestures and framing, and vocal delivery. The recalibration process addressed three issues in response to RQ2.
First, the eye contact algorithm was recalibrated to accept a downward pitch up to 20 degrees, mapping this coordinate range as Maintained Eye Contact to eliminate punitive scoring for natural reading behavior. Second, a Dynamic Exclusion protocol was implemented so that when a dimension such as Hand Visibility fails to meet the minimum tracking confidence threshold, the data point is marked as NULL and excluded from the final average, ensuring evaluation only on behaviors the system can confidently observe. Third, a weighted vocal override heuristic was added so that when a student demonstrates high volume stability alongside an optimal speech rate, the filler word penalty is actively deprioritized, appropriately weighting fluency, pacing, and vocal tone as primary confidence indicators.
These recalibrations represent a first step toward aligning algorithmic design with human communicative experience. Larger-scale validation will be necessary to assess whether these adjustments produce meaningful improvements in alignment across diverse user populations.
Conclusion
This pilot exposes critical friction points between AI-generated interview assessment and human communicative experience. While automated platforms offer scalability and immediacy for career readiness development, the data suggest meaningful gaps between algorithmic scoring and student self-perception, particularly regarding eye contact tracking, gesture recognition, filler word penalization, and the relational dimensions of performance. These findings are exploratory and context-specific, but they point toward a broader challenge: AI systems that evaluate communication confidence must account for the environmental, cognitive, and relational variability inherent in human interaction.
Rebuilding trust in automated interview platforms requires intentional redesign centered on accessibility, transparency, and respect for human variability. Systems should allow dynamic calibration of environmental conditions before scoring: implement Explainable AI (XAI) principles with time stamped, behavior-specific feedback; and train models on diverse datasets that reflect cultural and neurodiverse communication styles (Chen, 2023; Ajunwa, 2023). The recalibrations reported here offer one model for iterative, evidence-based platform refinement.
Implications
The technical limitations and user frustrations documented here carry practical implications when these tools are deployed at scale. The algorithm’s inability to account for variables such as dark clothing, eyewear, or physical room constraints means that evaluations may reflect environmental and hardware conditions rather than professional competencies. This introduces a socioeconomic bias into the confidence assessment, as candidates with lower-quality equipment or less-controlled recording environments may be systematically disadvantaged (Springle & Bourdage, 2023).
Strict penalization of filler, non-standard gaze, and postural variation also inherently disadvantages marginalized groups. The EEOC has increasingly scrutinized AI tools, warning that speech patterns and facial expressions can unlawfully screen out individuals with disabilities, speech impediments, or non-native accents (Ajunwa, 2023). As candidates become aware of these heuristics, there is a risk that they abandon authentic communication in favor of algorithmic appeasement, homogenizing behavior in ways that ultimately undermine the predictive validity of the interview itself. The deployment of opaque behavioral-scoring systems thus risks reducing human communicative potential to discrete, decontextualized data points.
References
Ajunwa, I. (2023). Automated video interviewing as the new phrenology. Berkeley Technology Law Journal, 36(3), 1173–1226.
Argyle, M. (2013). Bodily communication. Routledge.
Bailenson, J. N. (2021). Nonverbal overload: A theoretical argument for the causes of Zoom fatigue. Technology, Mind, and Behavior, 2(1), 1–6. https://doi.org/10.1037/tmb0000030
Bandura, A. (1997). Self-efficacy: The exercise of control. W. H. Freeman.
Bandura, A. (2001). Social cognitive theory: An agentic perspective. Annual Review of Psychology, 52(1), 1–26. https://doi.org/10.1146/annurev.psych.52.1.1
Barker, A. (2025). How to improve your communication skills: How to build trust, be heard, and communicate with confidence. Kogan Page.
Campion, M. C., Campion, M. A., Campion, E. D., & Reider, M. H. (2016). Initial investigation into computer scoring of candidate essays for personnel selection. Journal of Applied Psychology, 101(7), 958–975.
Chan, M. (2000). Organizational justice theories and landmark cases. The International Journal of Organizational Analysis, 8(1), 68–88.
Chen, Z. (2023). Ethics and discrimination in artificial intelligence-enabled recruitment practices. Humanities and Social Sciences Communications, 10, Article 567. https://doi.org/10.1057/s41599-023-02079-x
Creswell, J. W., & Creswell, J. D. (2018). Research design: Qualitative, quantitative, and mixed methods approaches (5th ed.). Sage.
Creswell, J. W., & Plano Clark, V. L. (2018). Designing and conducting mixed methods research (3rd ed.). Sage.
Dey, F., & Cruzvergara, C. (2014). Evolution of career services in higher education. New Directions for Student Services, 2014(148), 5–18. https://doi.org/10.1002/ss.20105
Fauville, G., Luo, M., Queiroz, A. C. M., Bailenson, J. N., & Hancock, J. (2021). Nonverbal mechanisms predict Zoom fatigue and explain why women experience higher levels than men. SSRN Electronic Journal. https://doi.org/10.2139/ssrn.3820035
Fox Tree, J. E. (2003). Disfluencies in spoken language. In L. Nadel (Ed.), Encyclopedia of cognitive science (Vol. 1, pp. 983–986). Nature Publishing Group.
Fox Tree, J. E. (2007). Functional spontaneous speech phenomena. Perspectives on Fluency and Fluency Disorders, 17(2), 17–19.
Gifford, R., Ng, C. F., & Wilkinson, M. (1985). Nonverbal cues in the employment interview: Links between applicant qualities and interviewer judgments. Journal of Applied Psychology, 70(4), 729–736. https://doi.org/10.1037/0021-9010.70.4.729
Hansen, K., Oliphant, G. C., Oliphant, B. J., & Hansen, R. S. (2009). Best practices in preparing students for mock interviews. Business Communication Quarterly, 72(3), 318–327.
Hudak, K., Kile, A., Grodziak, E., & Keptner, E. (2019). Advancing student interview skills: Incorporating virtual interview technology into the basic communication course. International Journal for the Scholarship of Teaching and Learning, 13(1), Article 3. https://doi.org/10.20429/ijsotl.2019.130103
Jackson, D., & Wilton, N. (2017). Perceived employability among undergraduates and the importance of career self-management, work experience and individual characteristics. Higher Education Research & Development, 36(4), 747–762.
Jaser, Z., Petrakaki, D., Starr, R., & Oyarbide-Magaña, E. (2022, January 13). Where automated job interviews fall short. Harvard Business Review. https://hbr.org/2022/01/whereautomated-job-interviews-fall-short
Langer, M., König, C. J., & Krause, K. (2017). Examining digital interviews for personnel selection: Applicant reactions and interviewer ratings. International Journal of Selection and Assessment, 25(4), 371–382. https://doi.org/10.1111/ijsa.12191
Langer, M., König, C. J., & Papathanasiou, M. (2019). Highly automated job interviews: Acceptance under the influence of stakes. International Journal of Selection and Assessment, 27(3), 217–234. https://doi.org/10.1111/ijsa.12246
Lingg, M. (1996). Training for job-skill confidence. Journal of Career Development, 22(4), 261–271. https://doi.org/10.1177/089484539602200404
Martín-Raugh, M. P., Kell, H. J., Randall, J. G., Anguiano-Carrasco, C., & Banfi, C. (2023). Speaking without words: A meta-analysis of over 70 years of research on the power of nonverbal cues in job interviews. Journal of Organizational Behavior, 44(1), 132–156. https://doi.org/10.1002/job.2670
Okay-Somerville, B., & Scholarios, D. (2017). Position, possession or process? Understanding objective and subjective employability during university-to-work transitions. Studies in Higher Education, 42(7), 1275–1291.
Quinlan, K. M., & Renner, P. (2005). Mock job interviews: Improving the self-efficacy and interview performance of college students. Journal of Career Development, 32(1), 75–88.
Rowell, A., & Mihuta, T. (2016). Technology and career preparation: Using virtual interview recordings (VIRs) in an apparel, design, and textiles (ADT) professional seminar course. Journal of Family and Consumer Sciences, 108(2), 27–32.
Springle, A., & Bourdage, J. (2023). Home environment and performance in asynchronous video interviews: Possible sources of adverse impact and candidate reactions. International Journal of Selection and Assessment.
Suen, H. Y., & Hung, K. E. (2023). Building trust in automatic video interviews using various AI interfaces: Tangibility, immediacy, and transparency. Computers in Human Behavior, 143, Article 107713. https://doi.org/10.1016/j.chb.2022.107713
Suen, H. Y., & Hung, K. E. (2025). Comparing job applicant deception in asynchronous vs. synchronous video interviews, with and without AI-assisted assessments. Information Technology & People, 38(2), 963–983.
Sweller, J. (1988). Cognitive load during problem solving: Effects on learning. Cognitive Science, 12(2), 257–285. https://doi.org/10.1016/0364-0213(88)90023-7
Sweller, J., van Merriënboer, J. J. G., & Paas, F. G. W. C. (1998). Cognitive architecture and instructional design. Educational Psychology Review, 10(3), 251–296. https://doi.org/10.1023/A:1022193728205
Tompsett, H., Henderson, K., Byrne, J. M., Mew, E. G., & Tompsett, C. (2017). On the learning journey: What helps and hinders the development of social work students’ core pre-placement skills? Social Work Education, 36(1), 6–25.
Tross, S. A., & Maurer, T. J. (2008). The effect of coaching on interview performance: Practice makes perfect? Journal of Vocational Behavior, 73(3), 398–405.
Xu, Y., Chen, Z., & Dong, M. (2025). Shaping the fairness journey: The roles of AI literacy, explanation, and interpersonal interaction in AI interviews. International Journal of Human-Computer Studies, 205, Article 103629. https://doi.org/10.1016/j.ijhcs.2025.103629