How rejection letters from AI recruiters shape recruiter's prestige and experienced respect

Abstract

Introduction:

Organizations increasingly rely on artificial intelligence (AI) in recruitment, yet little is known about how AI systems should communicate unfavorable outcomes such as rejection. The present research examined how type of rejection letters of an AI recruiter shape applicants' perceptions of the recruiter and their experienced respect.

Methods:

Two experiments were conducted. In Study 1, the type of AI-generated rejection letter (considerate vs. typified) and applicants' employment status (unemployed vs. control) were experimentally manipulated. Study 2 examined the same rejection-letter manipulation while using participants' actual employment status. Across both studies, mechanistic dehumanization, perceived recruiter prestige, and experienced respect were assessed.

Results:

In Study 1, the effects of type of AI rejection letter emerged only among participants positioned as unemployed, suggesting that unemployment status shaped sensitivity to AI-enabled recruitment. In Study 2, considerate versus typified rejection letters directly influenced mechanistic dehumanization. Mechanistic dehumanization, in turn, was negatively associated with perceived recruiter prestige and experienced respect. Unexpectedly, participants' actual employment status did not moderate the effects of rejection-letter type. Exploratory analyses, however, suggested that attitudes toward unemployment influenced the interaction between rejection-letter type and employment status in predicting the outcome variables.

Discussion:

The findings demonstrate that how AI recruiters communicate rejection can shape applicants' perceptions of the recruiter and their experience of respect.

Introduction

Organizations increasingly rely on artificial intelligence (AI), algorithms, and automated decision systems to perform tasks that were historically carried out by humans (Fousiani et al., 2026a; Van Quaquebeke and Gerpott, 2024). In contemporary workplaces, AI tools support or replace human judgment in areas such as evaluation and decision-making, promising efficiency through speed, consistency, and scalable information processing (Fahim et al., 2025; Starke et al., 2022). These developments reshape how organizational decisions are made and experienced by employees and external stakeholders (Lee, 2018; Starke et al., 2022).

Recruitment and personnel selection is a domain in which algorithmic tools have expanded rapidly. Traditional algorithmic systems are commonly used to screen résumés, rank candidates or evaluate interview performance (Acikgoz et al., 2020; Langer et al., 2019; Wesche and Sonderegger, 2021), while more recent generative AI systems, including applications based on large language models, can produce natural-language content and assist with recruitment communications—such as drafting job advertisements, interview correspondence, feedback, and rejection messages (Andrieux et al., 2024). Traditional algorithmic systems often operate alongside generative AI tools (or human decision makers) in recruitment processes. While organizations have embraced AI-enabled recruitment tools, applicants often respond ambivalently or even negatively to AI in hiring, commonly perceiving AI-driven recruitment as impersonal, “cold,” and dehumanizing (Fritts and Cabrera, 2021; Mirowska and Mesnet, 2022; Pyle et al., 2024). This matters because recruitment is not merely a matching procedure but also a social encounter in which applicants seek recognition, fair treatment, and respect (Bankins et al., 2022; Fousiani et al., 2026a).

Negative reactions persist even though AI is sometimes introduced to reduce human bias. More broadly, people often prefer human over AI (or algorithmic) judgment, a tendency reflected in research on algorithm aversion (Dietvorst et al., 2015; Mirowska and Mesnet, 2022), although acceptance can vary depending on the context (Fousiani et al., 2024; Logg et al., 2019). These findings suggest that applicant responses to AI recruitment are conditional and may depend on how AI-driven decisions are designed and communicated (see Fousiani et al., 2026a).

A central issue concerns how negative outcomes are communicated. Rejected applicants rarely receive meaningful explanations, despite typically wanting to know why they were unsuccessful (Olckers et al., 2022). Classic applicant reaction research further shows that organizational communication affects responses to selection outcomes (Gilliland et al., 2001; Hausknecht et al., 2004). In line with this, rejection message design—whether typified (e.g., generic and minimal) or considerate (i.e., informative and personalized)—shapes job seekers' reactions (Cortini et al., 2019; Eveleth and Eveleth, 2021; Thominet, 2020). These concerns are particularly relevant when rejection letters are generated by AI, as automation may increase standardization and therefore perceived impersonality (see Fousiani et al., 2026a).

In the present research, we use the term AI recruiters (or AI-enabled recruitment system) to refer to a computational system described as performing recruitment-related tasks that would otherwise be undertaken by a human recruiter, including reviewing application information, evaluating the applicant's interview, and communicating the selection outcome.

Moreover, we focus on applicants' reactions to rejection decisions communicated by AI recruiters via rejection emails (Barattucci et al., 2025). Although AI recruiters may communicate with applicants through various channels (e.g., chatbots or messaging platforms), the present research focuses on AI-generated rejection emails, which is a rather common and realistic form of recruitment communication. We examine how the completeness of AI-generated rejection emails—from typified and standardized communication to more considerate and personalized communication—shapes applicants' perceptions of the AI recruiter and applicants' experienced respect in the recruitment process. Drawing on Thominet's genre analysis, typified rejection letters consist mainly of mandatory moves, such as thanking the applicant, signaling a decision, refusing the job, and closing with best wishes. In contrast, more considerate and elaborated letters add customizing and personalizing elements, especially personalization and explanation, which may help applicants make sense of the outcome and feel more acknowledged (Thominet, 2020).

Building on anthropomorphism theory (Epley et al., 2007; Fousiani et al., 2026a; Ochmann et al., 2024), we argue that the content and style used in the communication of rejection letters shapes whether applicants construe the AI recruiter as a dehumanized tool (machine-like, cold and unempathic agent; Bastian and Haslam, 2010; Fousiani et al., 2016; 2019; Haslam, 2006) or as a legitimate, less machine-like recruiter. We propose that considerate and elaborated rejection letters (personalized with customized explanation) reduce dehumanization of the AI recruiter relative to typified, minimal letters, and in turn increase perceived recruiter prestige (i.e., see the AI recruiter as knowledgeable, competent, and experienced; Henrich and Gil-White, 2001; Kakkar et al., 2020) as well as applicants' experienced respect (van Quaquebeke and Eckloff, 2010). Crucially, we test applicants' employment status as a boundary condition, expecting the hypothesized effects to be stronger among unemployed applicants who may experience greater vulnerability and dependence on recruiter decisions (Nørskov et al., 2022; Pyle et al., 2024).

Overall, this study makes three main theoretical contributions. First, it advances research on AI decision-making in recruitment processes by moving beyond broad judgments of fairness and justice (Bankins et al., 2022; Fousiani et al., 2026a) to examine a high-stakes and psychologically consequential event: rejection. By focusing on how rejected applicants evaluate AI recruiters and the experience itself, our study highlights perceived respect and impressions of the AI recruiter as central outcomes that may shape how applicants interpret and respond to negative selection decisions of AI agents (Bankins et al., 2022; Pyle et al., 2024).

Second, the study shifts attention away from the well-documented advantage that human recruiters typically hold over AI recruiters in shaping applicants' reactions (Dietvorst et al., 2015; Mirowska and Mesnet, 2022), and instead examines when and how AI recruiters may be experienced in less dehumanizing terms. Drawing on insights from anthropomorphism research in AI-enabled hiring, and integrating them with classic work on applicant reactions, we theorize that the content (i.e., completeness and consideration) of rejection communication is a key condition that can make AI-driven decisions feel less mechanical and more socially meaningful (Gilliland et al., 2001; Hausknecht et al., 2004; Ochmann et al., 2024). By identifying considerate and elaborated rejection letters as a concrete design feature that can shape applicants' perceptions of the recruiter and the experienced interaction, the study positions rejection communication as an important (yet understudied) antecedent of AI recruitment systems acceptance.

Third, the study forwards a better understanding of when AI recruiters are perceived as dehumanizing (or more human-like) by demonstrating that acceptance of AI recruitment depends not only on the content or quality of AI-mediated communication but also on the applicant's employment status. By theorizing and testing applicant employment status (and the heightened vulnerability that may accompany being unemployed; Pyle et al., 2024) as a boundary condition, we clarify why standardized, minimal AI communication may be experienced as particularly harmful for some applicants, whereas more considerate and complete communication can support more respectful and psychologically acceptable recruitment encounters (Fritts and Cabrera, 2021; Mirowska and Mesnet, 2022).

Rejection communication as a consequential feature of AI-mediated recruitment

Rejected applicants typically report receiving limited explanation through typified letters, despite a strong desire to understand why they were not selected (Olckers et al., 2022). Typified rejection letters are characterized by standardized, conventional “moves” that fulfill minimal communicative requirements, such as thanking the applicant, signaling that a decision has been made, communicating refusal, and closing with best wishes (Thominet, 2020). These letters provide clarity about the outcome but offer limited informational and relational content. In contrast, more considerate and elaborated rejection letters extend beyond these mandatory moves by including personalization and explanation, elements that acknowledge the applicant more directly and provide decision rationale that can facilitate sensemaking (Thominet, 2020). In AI-mediated recruitment, where communication can easily become standardized and repetitive, this distinction between typified vs. considerate rejection letters may be particularly consequential for how applicants perceive the decision-maker and the interaction itself (Starke et al., 2022).

In this study, we propose that the level of consideration and completeness of AI-generated rejection communication shapes applicants' perceptions of the AI recruiter by influencing mechanistic dehumanization of the recruiter. Mechanistic dehumanization refers to perceiving a target as lacking human nature—cold, rigid, and emotionless—resembling an object or automaton rather than a socially responsive agent (Haslam, 2006; see also Bastian and Haslam, 2010; Roupa et al., 2025; Sakalaki et al., 2017). When applied to AI recruiters, mechanistic dehumanization captures the degree to which applicants construe the recruiter as a mindless system incapable of understanding, care, or socially appropriate responsiveness, particularly in contexts where interpersonal sensitivity is expected (Bankins et al., 2022; Lee, 2018).

This reasoning aligns with anthropomorphism theory, which conceptualizes anthropomorphism as the tendency to attribute humanlike qualities (e.g., intentions, feelings, social motives) to non-human agents (Epley et al., 2007). Importantly, anthropomorphism does not imply that recipients literally believe the agent is human; rather, humanlike attributions arise when an agent's cues allow people to interpret it as socially meaningful and responsive (see also Fousiani et al., 2026a). In AI recruitment contexts, communication cues are central to these interpretations because applicants often have no direct interaction with a human decision-maker and must infer the “quality” of the decision agent from the message itself (Ochmann et al., 2024).

Downstream implications: prestige perceptions and experienced respect

Mechanistic dehumanization of the AI recruiter should carry important implications for how applicants evaluate both the decision agent (AI recruiter) and their own treatment in the interaction. When an AI recruiter is construed as cold, rigid, and emotionally incapable, applicants are likely to infer that the agent is not equipped to exercise socially appropriate judgment in a high-stakes interpersonal context such as personnel selection (Bankins et al., 2022; Haslam, 2006). This inference is consequential because applicants use cues about the decision-maker's quality and legitimacy to interpret the meaning of selection outcomes and the trustworthiness of the process (Lee, 2018; Starke et al., 2022).

Accordingly, we argue that mechanistic dehumanization of the AI recruiter should undermine perceived prestige of the recruiter. Prestige reflects the extent to which a target is granted esteem and deference because it is seen as knowledgeable, skilled, and competent (Henrich and Gil-White, 2001). In selection contexts, such perceptions are central because they indicate whether applicants view the recruiter as a credible evaluator whose decisions reflect expertise and professional standards. By reducing the recruiter to an impersonal and mindless system, mechanistic dehumanization may erode perceptions of professional credibility and judgment, thereby lowering perceived prestige. This logic is consistent with evidence that applicants often experience algorithmic hiring as reductionistic and less legitimate, which can translate into more negative evaluations of the decision-making process and the organizational actor behind it (Koch-Bayram and Kaibel, 2024; Newman et al., 2020).

Indirect effects of rejection communication through mechanistic dehumanization

Applicant employment status as a moderator: heightened stakes and sensitivity to AI-mediated rejection

Applicants do not enter selection processes with equal levels of security. For unemployed individuals, recruitment outcomes typically carry higher instrumental and psychological stakes, because obtaining employment is closely tied to economic security, social integration, and wellbeing (McKee-Ryan et al., 2005; Pohlan, 2019). Meta-analytic evidence shows that unemployment is associated with lower psychological and physical wellbeing (McKee-Ryan et al., 2005), and research further suggests that job loss can undermine subjective feelings of social inclusion and life satisfaction (Gedikli et al., 2023; Pohlan, 2019). Importantly, unemployment is also widely conceptualized as a stigmatized and marginal social position in labor markets, as unemployed applicants can be viewed as less desirable and may anticipate biased evaluations due to their employment history (Ho et al., 2011; Krug and Prechsl, 2020). Such stigma-based dynamics can heighten vulnerability by making applicants more sensitive to the communication of a rejection during selection encounters (Kristal et al., 2022; O'Donnell et al., 2015).

Overview of the current research

We tested our hypotheses across two scenario-based experiments in which participants imagined applying for a position and subsequently receiving a rejection from an AI recruiter. Prior to the main studies, we conducted a pilot study to develop and validate the rejection letters used across the experiments. Following Thominet's (2020) genre framework, we varied whether the rejection letter contained only mandatory components (typified communication) or additionally included personalization and explanation (considerate communication). Thus, the pilot study served exclusively to establish the validity of the experimental manipulation, whereas the two subsequent studies tested the proposed theoretical model.

Study 1 was a time-lagged, two-wave study. At Time 1, we assessed participants' demographic characteristics and control variables. The experimental component took place at Time 2 and used a 2 (applicant employment status: unemployed vs. control) × 2 (rejection letter: typified vs. considerate) between-subjects design. Employment status was manipulated by presenting the applicant as having a 2-year unemployment gap, and we assessed mechanistic dehumanization of the recruiter, perceived recruiter prestige, and applicants' experienced respect. Study 1, therefore, provided an initial experimental test of our theoretical model by examining whether making unemployment salient as an applicant characteristic shaped reactions to different types of AI-generated rejection letters. In Study 2 (preregistered on the Open Science Framework; https://osf.io/tnz6p?view_only=a7b5b37ab34545b2914b00b319bb13d6; please see Supplementary material S1 for some deviations from preregistration), rather than manipulating employment status, we recruited both employed participants and unemployed job seekers and again experimentally manipulated rejection-letter type. We also assessed, exploratorily, negative attitudes toward unemployed people (Gurr and Jungbauer-Gans, 2013) to examine whether such attitudes further shape reactions to AI-mediated rejection. Accordingly, Study 2 served as a conceptual replication of Study 1 while increasing ecological validity by examining the hypothesized model among participants with actual employment statuses rather than experimentally assigned ones. This design also allowed us to explore whether attitudes toward unemployed people further qualified applicants' reactions to AI-mediated rejection. Figure 1 presents the overview of the research design and illustrates the purpose of the pilot study, Study 1, and Study 2. All participants provided their active informed consent prior to participation.

Figure 1

Pilot study

Participants

We recruited 80 participants (60% women) residing in the United Kingdom, ranging in age from 22 to 67 years (M = 38.00, SD = 9.11). All participants were employed in paid jobs at the time of the study.

Experimental procedure and design

Participants were recruited via Prolific and completed an online scenario-based experiment. The study took place in October 2025. All study materials were presented in English. The current study underwent a fast-track ethics review and was granted an exemption from full ethical review by the Ethics Committee of the University of Groningen, prior to data collection. All participants provided active informed consent before starting the survey. Participants were asked to imagine themselves applying for a Project Coordinator position at a fictitious organization (IntraCOM UK). This job position was selected because it represents a broadly recognizable, age-neutral, and gender-neutral white-collar role that does not require highly specialized technical expertise. This made the scenario broadly applicable to participants from diverse backgrounds and is consistent with previous vignette-based research using the same job role (e.g., Fousiani et al., 2026b). Participants first read a job advertisement describing the role, key responsibilities, and required qualifications. They were instructed to immerse themselves in the scenario and respond as if the experience were real and personally relevant. After reading the vacancy, participants were told that they submitted an application and completed an interview. A few days later, they received a formal rejection letter from IntraCOM UK and proceeded to answer questions about their perceptions and reactions.

The study employed a between-subjects design in which the content of the rejection letter was experimentally manipulated. Following Thominet (2020), we varied rejection-letter content along a continuum from more typified (i.e., containing only mandatory components) to less typified/more considerate. According to Thominet, the mandatory parts of a rejection letter include elements such as an appreciation move, declaration move, job-refusal move, and closing/best wishes; together, these constitute a more typified rejection letter. By contrast, more considerate (and less typified) letters additionally include components such as personalization and explanation. Accordingly, participants were randomly assigned to one of four rejection-letter conditions varying whether the letter included personalization and/or explanation: (a) a full letter (mandatory parts plus personalization and explanation), (b) no personalization (mandatory parts and explanation only), (c) no explanation (mandatory parts and personalization only), and (d) no personalization and no explanation (a typified letter including mandatory parts only). All letters communicated the same decision outcome (rejection) and included the same context (i.e., that the application and interview had been reviewed), while varying only in the presence vs. absence of (a) personalized/goodwill language and (b) an explanation/justification for the decision. Apart from these manipulated elements, the wording, tone, and overall structure of the rejection message were held constant across conditions. For the complete scenarios, see Supplemental material S2.

Measures

Experience of the rejection letter as considerate or typified

Participants evaluated the letter's content using three items: “To what extent does the rejection letter sound impersonal or typified/generic?,” and “To what extent does the rejection letter sound considerate (e.g., personalized and customized)?” “To what extent does the letter provide sufficient explanation for the rejection?” Items were rated on 7-point scales (1 = Not at all, 7 = Very much).

Results

To validate the rejection-letter manipulations, we conducted one-way analyses of variance (ANOVAs) in IBM SPSS Statistics (Version 30; IBM Corp., Armonk, NY, USA), comparing the four rejection-letter conditions on (a) perceived typified communication, (b) perceived considerate communication, and (c) provision of sufficient explanation.

Impersonal/typified communication

Perceptions of the letter as impersonal/typified differed significantly across conditions, F_{(3, 76)} = 8.22, p < 0.001, η² = 0.25. Descriptively, the typified letter (no personalization and no explanation) was rated as the most impersonal/typified (M = 5.79, SD = 1.08), whereas the full letter was rated as the least impersonal/typified (M = 3.38, SD = 1.80). The intermediate conditions showed mid-range ratings (no explanation: M = 4.40, SD = 1.79; no personalization: M = 4.70, SD = 1.34).

Considerate communication

Perceptions of the letter as considerate also differed significantly across conditions, F_{(3, 76)} = 6.92, p < 0.001, η² = 0.22. The full letter was rated as the most considerate (M = 5.10, SD = 1.45), whereas the typified letter (no personalization and no explanation) was rated as the least considerate (M = 3.16, SD = 1.50). The intermediate conditions again fell in between (no explanation: M = 4.15, SD = 1.18; no personalization: M = 4.05, SD = 1.23).

Provision of sufficient explanation

Perceptions of whether the rejection letter provided sufficient explanation differed significantly across conditions, F_{(3, 76)} = 9.71, p < 0.001, η² = 0.28. The full letter was perceived as providing the most sufficient explanation (M = 5.67, SD = 1.20), whereas the typified letter (no personalization and no explanation) was perceived as providing the least explanation (M = 3.42, SD = 1.61). As expected, the intermediate conditions fell between these extremes: the no-personalization condition (which retained explanation but excluded personalization) received relatively high explanation ratings, yet lower than the full letter (M = 4.55, SD = 1.47), whereas the no-explanation condition was rated lower (M = 3.50, SD = 1.79).

Discussion

Overall, the pilot results indicate that the four letters differed significantly on the intended manipulation targets (impersonal/typified vs. considerate communication). Based on this pattern, in the main studies we operationalized the considerate rejection-letter condition using the full letter (mandatory components plus personalization and explanation) and operationalized the typified condition using the short letter containing only the mandatory components (i.e., no personalization and no explanation).

Study 1

Method

Participants

Participants were 197 adults ranging in age from 22 to 68 years (M = 44.97, SD = 11.45). The sample included 104 men (52.8%), 92 women (46.7%), and one participant who preferred not to disclose gender (0.5%). Most participants reported British nationality (94.4%), followed by Irish (2.0%), American (1.0%), and other nationalities (2.5%). Regarding education, 16.2% reported a high school diploma or equivalent, 17.3% some college without a degree, 4.1% an associate degree, 42.1% a bachelor's degree, 15.7% a master's degree, 3.6% a doctoral degree, and 1.0% a professional degree. Most participants reported having a paid job (91.4%), while 8.6% reported not having a paid job. Participants' occupational backgrounds were diverse, with the largest group reporting management/professional occupations (43.1%), followed by service (16.2%), sales/office (15.7%), government (12.7%), production/transportation/material moving (5.1%), and smaller proportions in other categories. All participants indicated English as their mother language.

Experimental procedure and manipulations

Participants were recruited via Prolific and completed an online scenario-based experiment. The study took place in October 2025. All study materials were presented in English. The study underwent a fast-track ethics review and was granted an exemption from full ethical review by the Ethics Committee of the University of BLINDED FOR REVIEW, prior to data collection. All participants provided active informed consent. To prevent participants' fatigue from a long questionnaire, we measured participants' demographic characteristics and control variables at Time 1, which took place one week before the experimental component of the study. At Time 2, the study employed a × 2 (applicant employment status: unemployed vs. control) by 2 (type of rejection letter: typified vs. considerate) between-subjects design (see also Supplementary material S3 for further information about the research design, including a human recruiter condition as well, for exploratory purposes). Participants were asked to imagine themselves applying for a Project Coordinator position at a fictitious organization. Participants first read a job advertisement describing the role, key responsibilities, and required qualifications. They were instructed to immerse themselves in the scenario and respond as if the experience were real. All participants were informed that the recruitment process was conducted by an AI recruiter responsible for reviewing applications, evaluating the interview, and communicating the hiring decision. Thus, the AI was presented as being actively involved across both the evaluation and communication stages of the process. No particular AI architecture, such as a large language model, was used in the vignette.¹

Experimental manipulations

Applicant employment status manipulation

Participants were randomly assigned to one of two applicant-status conditions. In the unemployed condition, participants were told that their application materials (CV and/or application profile) contained a 2-year period of unemployment, presented as a noticeable gap in their employment history. This gap was described as occurring prior to the current application, thereby signaling a marginalized status in the labor market. In the control condition, the application materials contained no information suggesting unemployment or an employment gap (i.e., the applicant's employment status was not made salient).

Rejection letter manipulation

Before receiving the hiring decision, participants were asked to imagine that they had applied for the position, were motivated to obtain the job, and had become enthusiastic about the role and the organization after reviewing the vacancy and completing the application process. Participants were instructed to respond as if the application experience were real and personally relevant.

Participants then received a rejection letter (considerate vs. typified) from the AI recruiter, similar to the one in the Pilot Study. The considerate rejection-letter used the full letter (mandatory components plus personalization and explanation), while the typified letter consisted of the short letter containing only the mandatory components (i.e., no personalization and no explanation). Apart from these manipulated elements, the wording, tone, and overall format of the rejection message were held constant across conditions. For the complete vignettes see Supplemental material S2.

Measures (Time 2)

Mechanistic dehumanization of the recruiter

Mechanistic dehumanization was measured with six items assessing the extent to which the recruiter was perceived as machine-like, cold, and lacking human qualities. The scale was based on measures already used in prior studies (Bastian and Haslam, 2010; Fousiani et al., 2016, 2019; Haslam, 2006). Participants rated the recruiter using statements reflecting mechanistic perceptions (e.g., cold/robotic, object-like), alongside reverse-coded items capturing warmth and emotional capacity. Responses were recorded on a 7-point scale (1 = Strongly disagree, 7 = Strongly agree), with higher scores indicating greater mechanistic dehumanization. Reliability of the scale was acceptable α = 0.74.

Perceived respect toward the applicant

Perceived respect was assessed with 10 items capturing the extent to which participants felt treated with respect and consideration (e.g., recognition of effort, feeling treated as a legitimate candidate). An example item was: “Despite the negative outcome, the recruiter acknowledged my skills and qualifications.” Items were adapted from the Respectful Leadership Scale (van Quaquebeke and Eckloff, 2010) for the recruiter–applicant context and were rated on a 7-point scale (1= Strongly disagree, 7 = Strongly agree) and averaged, with higher scores indicating higher perceived respect. Reliability of the scale was acceptable (α = 0.97).

Recruiter prestige

Perceived prestige of the AI recruiter was measured with three items assessing professionalism, knowledgeability, and credibility (e.g., the recruiter came across as knowledgeable and experienced). The measure was adapted from Wilhelmy et al. (2016, 2017). Items were rated on a 7-point scale (1= Strongly disagree, 7 = Strongly agree) and averaged, with higher scores indicating greater perceived prestige. Reliability of the scale was α = 0.92.

Control variables (Time 1)

We controlled for participants' general attitudes toward AI, anthropocentric decision-making beliefs, and organizational conspiracy beliefs because these three individual-difference variables can shape people's baseline willingness to trust, accept, or negatively interpret AI-driven decisions, independently of the rejection-letter manipulation. First, more positive attitudes toward AI are linked to greater acceptance of algorithmic advice/decisions, whereas negative attitudes often translate into algorithm aversion and harsher evaluations of algorithmic output (Millet et al., 2023). Second, stronger anthropocentric decision-making beliefs (i.e., seeing decision-making as uniquely human) should bias participants against AI acting in a “human-typical” domain like hiring, making AI decisions feel less legitimate and more unsettling, thereby affecting dehumanization and downstream reactions regardless of letter style. Third, organizational conspiracy beliefs reflect a general tendency to suspect hidden agendas and malevolent intent from powerful actors (e.g., organizations/management; Fousiani et al., 2025), which can spill over into interpreting AI-based hiring as opaque, self-serving, or untrustworthy, potentially amplifying negative reactions even when communication is relatively respectful. These control variables were measured as follows:

General attitudes toward AI

These were assessed using Grassini's (2023) four-item scale. The items capture perceived benefits and openness toward AI (e.g., “I believe that AI can or will improve my life”). Responses were given on a 10-point scale ranging from 1 = Not at all to 10 = Completely agree. Items were averaged, with higher scores indicating more positive attitudes toward AI (α = 0.91).

Anthropocentric decision-making beliefs

These beliefs were measured with five items adapted from anthropocentric creativity beliefs introduced by Millet et al. (2023). The items capture the extent to which participants endorse decision making as a uniquely human capacity (e.g., “Decision making is a uniquely human characteristic/ ability”). Responses were recorded on a 7-point Likert scale (1 = Strongly disagree, 7 = Strongly agree), and items were averaged, with higher scores indicating stronger anthropocentric decision-making beliefs (α = 0.89).

Organizational conspiracy beliefs

These beliefs were assessed with the organizational conspiracy beliefs scale developed by Van Prooijen and De Vries (2016). Participants indicated their agreement with nine items describing conspiratorial perceptions of management and supervisors in their organization (e.g., “In the organization where I work, our management has a hidden agenda”). Responses were given on a 7-point Likert scale (1 = Strongly disagree, 7 = Strongly agree), and items were averaged such that higher scores reflected stronger organizational conspiracy beliefs (α = 0.92).

Results

Correlations between study variables, means and standard deviations are presented in Table 1.

Analytic strategy

We used Hayes' PROCESS macro with 5,000 bootstrap samples and 95% confidence intervals. Rejection-letter type was coded 1 = considerate and 2 = typified, and employment status was coded 1 = unemployed and 2 = control. All models included general attitudes toward AI, anthropocentric decision-making beliefs, and organizational conspiracy beliefs as covariates.

Effects of letter type on mechanistic dehumanization of the recruiter and in turn, on recruiter prestige

To test the hypothesized effects, we ran a mediation analysis using model 4 in PROCESS macro. Letter type was entered as the independent variable, mechanistic dehumanization was entered as the mediator, and recruiter prestige as the outcome variable. General attitudes toward AI, anthropocentric decision-making beliefs, and organizational conspiracy beliefs were entered as covariates. As shown in Table 2, contrary to Hypothesis 1, letter type did not significantly predict mechanistic dehumanization of the recruiter. Mechanistic dehumanization significantly predicted lower recruiter prestige, supporting Hypothesis 2a. However, neither the direct effect of letter type on recruiter prestige nor the indirect effect was significant. These results do not provide support for Hypothesis 3a.

Effects of letter type on mechanistic dehumanization of the recruiter and, in turn, on applicant experienced respect

We then tested the same mediation model with experienced respect as the outcome (PROCESS Model 4). Mechanistic dehumanization significantly predicted lower experienced respect, supporting Hypothesis 2b. However, neither the direct effect of letter type on experienced respect, nor the indirect effect via mechanistic dehumanization, was significant. Therefore, these results did not support Hypothesis 3b (see Table 2 for the relevant statistics).

Applicant employment status as a moderator of the effect of letter type on mechanistic dehumanization

Next, we tested whether applicant employment status (unemployed vs. control) moderated the effect of letter type on mechanistic dehumanization and in turn, predicted each of the outcome variables (PROCESS Model 8). As shown in Table 3, the letter type × employment status interaction was significant. Supporting Hypothesis 4, the conditional effect of letter type on mechanistic dehumanization was significant among unemployed applicants but not in the control condition. Given the coding of letter type (1 = considerate, 2 = typified), these results indicate that typified (vs. considerate) rejection letters elicited higher mechanistic dehumanization among unemployed applicants, whereas letter type did not influence mechanistic dehumanization in the control condition (see Table 3 for the relevant statistics; Figure 2).

Figure 2

Conditional indirect effects of letter type on recruiter prestige via mechanistic dehumanization

Mechanistic dehumanization significantly and negatively predicted recruiter prestige. The direct letter type × employment status interaction predicting recruiter prestige was not significant. However, supporting Hypothesis 5a, the conditional indirect effect of letter type on recruiter prestige via mechanistic dehumanization was significant among unemployed applicants, but not in the control condition. The index of moderated mediation was significant, indicating that the indirect effect differed significantly across employment-status conditions (see Table 3 for the relevant statistics).

Conditional indirect effects of letter type on experienced respect via mechanistic dehumanization

Mechanistic dehumanization significantly and negatively predicted experienced respect. Employment status also significantly predicted experienced respect. The direct letter type × employment status interaction predicting experienced respect was not significant. However, supporting Hypothesis 5b, the conditional indirect effect of letter type on experienced respect via mechanistic dehumanization was significant among unemployed applicants, but not in the control condition. The index of moderated mediation was significant, indicating that the indirect effect differed significantly across employment-status conditions (see Table 3 for the relevant statistics).

Discussion

Contrary to Hypothesis 1, we did not find a significant main effect of rejection-letter type on mechanistic dehumanization of the AI recruiter. However, in line with Hypotheses 2a and 2b, mechanistic dehumanization of the recruiter was negatively related to perceived recruiter prestige and perceived respect of the applicant, providing support for Hypotheses 2a and 2b, respectively. Nevertheless, opposite to Hypothesis 3, the indirect effects of letter type on the outcome variables via mechanistic dehumanization were not significant, meaning that Hypothesis 3 was not supported. Crucially, however, and in line with Hypotheses 4 and 5, the results revealed a significant moderation effect: applicant employment status (unemployed vs. control) moderated the effect of rejection-letter type on mechanistic dehumanization of the AI recruiter, which in turn predicted negatively both perceived recruiter prestige and experienced respect. Specifically, among applicants described as unemployed, considerate (vs. typified) rejection letters led to lower mechanistic dehumanization of the AI recruiter, which was associated with higher perceived prestige of the recruiter and greater experienced respect. In contrast, these effects did not emerge among applicants in the control condition, for whom the type of rejection letter did not significantly influence perceptions of the AI recruiter or experienced respect.

Study 2 aimed to test our hypotheses using a different design: rather than experimentally manipulating applicant employment status, we recruited participants who were either currently employed in a paid job or unemployed and actively seeking employment. This approach allowed us to examine the effects of rejection-letter type among participants for whom unemployment constituted a real and personally relevant status, thereby increasing the ecological validity of the study.

Study 2

Method

Participants

Participants were 232 adults ranging in age from 22 to 70 years (M = 41.01, SD = 11.28). The sample included 103 men (44.4%) and 129 women (55.6%). Most participants reported British nationality (73.3%), followed by American (20.7%), Irish (1.3%), and other nationalities (4.7%). Regarding education, 0.9% reported less than a high school degree, 15.9% a high school diploma or equivalent, 20.3% some college without a degree, 3.4% an associate degree, 38.4% a bachelor's degree, 18.1% a master's degree, 1.7% a doctoral degree, and 1.3% a professional degree. Most participants reported being employed (working in a paid job; n = 150, 64.7%), whereas 82 participants (35.3%) reported being unemployed and actively seeking a job.

Experimental procedure and manipulations

Like Study 1, participants were recruited via Prolific and completed an online scenario-based study in English. The study took place in November 2025. All study materials were presented in English. The study underwent a fast-track ethics review (see Study 1), and all participants provided active informed consent. Given the limited pool of unemployed job-seeking participants on Prolific, we implemented Study 2 as a single-wave design (in contrast to the two-wave design of Study 1) to minimize attrition and reduce the risk of an underpowered unemployed subgroup. To limit participant fatigue and keep the questionnaire length manageable, we did not include the control variables assessed at Time 1 in Study 1.

After providing informed consent and completing basic demographics (including current employment status: employed in a paid job vs. unemployed and actively seeking a job), participants were instructed to immerse themselves in a hiring scenario and respond as if the experience were real. Specifically, and similar to Study 1, participants read a job advertisement for a Project Coordinator position at a fictitious organization (IntraCOM UK). Again, participants were informed that the selection process was conducted by an AI recruiter with no human professionals involved (see also Supplementary material S3 for further information about the research design, including a human recruiter condition as well, for exploratory purposes).

The study employed a between-subjects design in which the type of rejection letter was experimentally manipulated (typified vs. considerate) similar to Study 1.

Measures

We measured mechanistic dehumanization of the recruiter, perceived respect toward the applicant, and recruiter prestige with the same measures that we used in Study 1. Reliabilities were α = 0.94, α = 0.97, and α = 0.82, respectively.

Further exploratory measures

Attitudes toward unemployment

We assessed negative attitudes toward unemployed people using Gurr and Jungbauer-Gans's (2013) prejudice toward the unemployed scale. The measure uses a percentage-estimate format based on Brigham's (1971) “percentage approach” which is widely used in social-psychological research to capture stereotypical beliefs about social groups. Participants indicated, on five items, how many out of 100 unemployed people they believed (a) are really seeking a job, (b) have higher unemployment benefits than their income would be, (c) simply do not want to work, (d) are satisfied with the financial support they receive, and (e) are illegally employed (responses ranged from 0 to 100). Items were scored so that higher values reflected more negative attitudes toward unemployed people. Internal consistency in the present study was acceptable (α = 0.75).

Results

Correlations between study variables, means and standard deviations are presented in Table 1.

Analytic strategy

We used Hayes' PROCESS macro with 5,000 bootstrap samples and 95% confidence intervals. Rejection-letter type was coded 1 = considerate and 2 = typified. We report effects from PROCESS Model 4 (mediation) and Model 8 (moderated mediation).

Effects of letter type on mechanistic dehumanization of the recruiter and, in turn, on recruiter prestige

To test the hypothesized effects, we ran a mediation analysis using Model 4, where letter type was entered as the independent variable, mechanistic dehumanization was entered as the mediator, and recruiter prestige as the outcome variable. As shown in Table 4, letter type significantly predicted mechanistic dehumanization of the recruiter, such that typified (vs. considerate) letters were associated with higher mechanistic dehumanization. These results provide support for Hypothesis 1.

Importantly, mechanistic dehumanization significantly predicted lower recruiter prestige, supporting Hypothesis 2a. The total effect of letter type on recruiter prestige was significant, whereas the direct effect was not. Crucially, the indirect effect of letter type on recruiter prestige via mechanistic dehumanization was significant, supporting Hypothesis 3a (Table 4).

Effects of letter type on mechanistic dehumanization of the recruiter and, in turn, on experienced respect

We then tested the same mediation model with experienced respect as the outcome (PROCESS Model 4). Mechanistic dehumanization significantly predicted lower experienced respect, supporting Hypothesis 2b. In addition, both the total and the direct effect of letter type on experienced respect were significant. Most importantly, the indirect effect via mechanistic dehumanization was significant, supporting Hypothesis 3b (see Table 4 for the relevant statistics).

Applicant employment status as a moderator of the effect of letter type on mechanistic dehumanization

Next, we tested whether employment status moderated the effect of letter type on mechanistic dehumanization (PROCESS Model 8). The letter type × employment status (coded as 1= employed, 2 = unemployed and actively seeking a job) interaction was not significant, indicating that employment status did not moderate the effect of letter type on mechanistic dehumanization. Thus, Hypothesis 4 was not supported (Table 5).

Conditional indirect effects of letter type on recruiter prestige via mechanistic dehumanization

The index of moderated mediation with recruiter prestige as outcome was not significant, indicating that the size of the indirect effect did not differ significantly across employment-status groups. Therefore, Hypothesis 5a was not supported (see Table 5 for the relevant statistics).

Conditional indirect effects of letter type on experienced respect via mechanistic dehumanization

Similarly, the index of moderated mediation with experienced respect as outcome variable was not significant, indicating that the indirect effect did not differ significantly by employment status. Therefore, Hypothesis 5b was not supported (see Table 5 for the relevant statistics).

Exploratory analyses with attitudes toward unemployment as an additional moderator

Exploratorily, we examined whether the effect of letter type on mechanistic dehumanization (and in turn on outcome variables) depended jointly on employment status and participant attitudes toward unemployment. To test this model, we run regression analysis in PROCESS (Model 12) predicting mechanistic dehumanization, and in turn each of the outcome variables. Letter type was entered as the independent variable, employment status as the first moderator and attitudes toward unemployment (mean-centered) was entered as the second moderator. As shown in in Table 6, the results showed a significant three-way interaction (letter type × employment status × attitudes toward unemployment), indicating that the association between letter type and mechanistic dehumanization varied across combinations of employment status and attitudes toward unemployment.

Probing the interaction revealed the following pattern: among employed participants, typified (vs. considerate) letters were associated with higher mechanistic dehumanization at the mean level of attitudes toward unemployment and at higher levels (+1 SD) that is when participants held either neutral or strong negative attitudes toward unemployment, whereas this conditional effect was not significant at lower levels (−1 SD). Conversely, among unemployed participants, typified (vs. considerate) letters were associated with higher mechanistic dehumanization especially at lower levels of negative attitudes toward unemployment (−1 SD), that is when participants held least negative attitudes toward unemployment, and moderately at the mean levels of attitudes toward unemployment. However, at higher levels of negative attitudes toward unemployment (+1 SD), letter type was not significantly related to mechanistic dehumanization among unemployed participants (see Table 6 for the relevant statistics; Figure 3).

Figure 3

Conditional indirect effects on recruiter prestige via mechanistic dehumanization

When examining recruiter prestige as the outcome, mechanistic dehumanization was negatively related to recruiter prestige, indicating that viewing the AI recruiter as more mechanistically dehumanized was associated with lower perceived prestige. The three-way interaction predicting recruiter prestige directly (letter type × employment status × attitudes toward unemployment) was not significant, nor were the conditional direct effects of letter type on recruiter prestige. This indicates that differences in recruiter prestige did not stem directly from rejection-letter type across combinations of employment status and attitudes toward unemployment but rather operated indirectly via mechanistic dehumanization.

Turning to the conditional indirect effects, the results revealed a significant moderated moderated mediation, as indicated by a significant index of moderated moderated mediation (see Table 6). This pattern shows that the indirect effect of rejection-letter type on recruiter prestige via mechanistic dehumanization depended jointly on applicants' employment status and their attitudes toward unemployment. Probing these conditional indirect effects clarified the nature of this interaction. Among employed participants, the indirect effect was significant at the mean and high (+1 SD) levels of attitudes toward unemployment, that is, when applicants held neutral or strong negative attitudes toward unemployment. At these levels, receiving a typified (vs. considerate) rejection letter increased mechanistic dehumanization of the AI recruiter, which in turn reduced perceived recruiter prestige. In contrast, among employed participants with low (−1 SD) attitudes toward unemployment, the indirect effect was not significant (Table 6).

Among unemployed participants, a different pattern emerged. The indirect effect was significant at low (−1 SD) levels of attitudes toward unemployment (that is when participants held least negative attitudes toward unemployment), such that typified (vs. considerate) rejection letters elicited higher mechanistic dehumanization, which subsequently reduced recruiter prestige. However, at the mean and high (+1 SD) levels of attitudes toward unemployment, the indirect effect was not significant, suggesting that for unemployed applicants who held neutral or more negative attitudes toward unemployment, rejection-letter type did not significantly shape recruiter prestige through mechanistic dehumanization (see Table 6 for the relevant statistics).

Conditional indirect effects on experienced respect via mechanistic dehumanization

The conditional indirect effects revealed a significant moderated moderated mediation, as evidenced by a significant index of moderated moderated mediation (see Table 6 for the relevant statistics). This indicates that the indirect effect of rejection-letter type on experienced respect through mechanistic dehumanization varied as a function of both employment status and attitudes toward unemployment.

Among employed participants, the indirect effect was significant at the mean and high (+1 SD) levels of attitudes toward unemployment. At these levels, typified (vs. considerate) rejection letters increased mechanistic dehumanization of the AI recruiter, which in turn reduced applicants' experienced respect. The indirect effect was not significant at low (−1 SD) levels of attitudes toward unemployment, suggesting that employed applicants with less negative attitudes toward unemployment were less responsive to differences in rejection-letter type in terms of experienced respect.

Among unemployed applicants, the indirect effect was significant at low (−1 SD) levels of attitudes toward unemployment. In this subgroup, typified rejection letters led to higher mechanistic dehumanization and, consequently, lower experienced respect. However, at the mean and high (+1 SD) levels of attitudes toward unemployment, the indirect effect was not significant, indicating that unemployed applicants who held neutral or more negative attitudes toward unemployment were less affected by rejection-letter type via mechanistic dehumanization.

Discussion

Study 2 tested hypotheses using a sample in which employment status reflected participants' real-life situation (rather than being manipulated). Consistent with Hypothesis 1, receiving a typified (vs. considerate) rejection letter increased mechanistic dehumanization of the AI recruiter. In line with Hypotheses 2a and 2b, higher mechanistic dehumanization, in turn, predicted lower perceived prestige of the AI recruiter and lower perceived respect toward the applicant. Supporting Hypothesis 3, the effect of rejection-letter type on both outcomes operated indirectly through mechanistic dehumanization, indicating that applicants' evaluations of the recruiter and perceived respect were shaped primarily by the extent to which the AI recruiter was perceived as cold, machine-like, and lacking human qualities.

Contrary to Hypotheses 4 and 5, however, employment status (employed vs. unemployed job seekers) did not significantly moderate the effect of letter type on mechanistic dehumanization or the corresponding indirect effects on recruiter prestige and perceived respect. However, exploratory analyses with attitudes toward unemployment as an additional moderator revealed a more nuanced pattern: among employed participants, typified letters increased mechanistic dehumanization primarily when they held average to more negative attitudes toward unemployed people, suggesting that stigmatizing beliefs may heighten sensitivity to impersonal AI communication even when one is not personally vulnerable. In contrast, among unemployed participants, the negative effects of typified letters were evident only among those with relatively non-stigmatizing attitudes toward unemployment; for unemployed individuals who themselves endorsed more negative views of unemployment, rejection-letter type no longer significantly influenced perceptions of the AI recruiter. These results are discussed in detail in the General Discussion.

General discussion

The present research examined how the type of rejection letters of an AI recruiter shapes applicants' perceptions of the recruiter and their own experience of respect in the recruitment process. Drawing on theorizing on mechanistic dehumanization (Haslam, 2006), anthropomorphism (Epley et al., 2007), and applicant reactions to selection procedures (Gilliland et al., 2001; Hausknecht et al., 2004), across two studies, we tested whether considerate (vs. typified) rejection letters reduce mechanistic dehumanization of AI recruiters and, in turn, increase perceived recruiter prestige (i.e., perceived recruiter experience, professionalism, knowledgeability) and applicant's experienced respect. We further examined applicant employment status as a boundary condition, given the increased vulnerability and dependence on recruitment outcomes typically associated with unemployment (Kristal et al., 2022; Krug and Prechsl, 2020; O'Donnell et al., 2015).

Across both studies, the findings provide a nuanced picture of when and for whom AI-mediated rejection communication matters. In Study 1, we did not find support for the hypothesized main effects of rejection-letter type on mechanistic dehumanization (Hypothesis 1), recruiter prestige or experienced respect (Hypotheses 2a and 2b), nor for the proposed mediation through mechanistic dehumanization (Hypothesis 3). However, consistent with Hypotheses 4 and 5, employment status of the applicant moderated these relationships. Specifically, among applicants described as unemployed (vs. control), considerate (vs. typified) AI rejection letters reduced mechanistic dehumanization of the AI recruiter, which in turn predicted higher perceived recruiter prestige and greater experienced respect in the recruitment process. Thus, in Study 1, the effects of AI rejection communication were conditional on applicants occupying a marginalized labor-market position (i.e., being unemployed).

In Study 2, using participants' actual employment status rather than a manipulation, the results partially diverged from Study 1. Here, rejection-letter type had a direct effect on mechanistic dehumanization, such that typified (vs. considerate) letters increased perceptions of the AI recruiter as cold and machine-like, supporting Hypothesis 1. Mechanistic dehumanization, in turn, negatively predicted recruiter prestige and experienced respect, supporting Hypotheses 2a and 2b. In line with Hypothesis 3, these effects operated indirectly through mechanistic dehumanization. However, contrary to Hypotheses 4 and 5, employment status alone did not moderate these effects. Nevertheless, exploratory analyses revealed that the moderation pattern emerged when applicants' attitudes toward unemployment were taken into account: the hypothesized conditional effects of employment status were significant and in the expected direction, primarily among participants who did not hold strongly negative attitudes toward unemployed people. Together, these findings indicate that AI-mediated rejection communication affects applicants' evaluations primarily through perceptions of mechanistic dehumanization, but that the strength of these effects depend on both applicants' employment position (Study 1) and their broader beliefs about unemployment (Study 2).

One plausible reason why the employment-status moderation (Hypotheses 4–5) was significant in Study 1 but non-significant (and only emerged conditional on attitudes) in Study 2 is that (un)employment status was operationalised differently across the two studies. In Study 1, unemployment was explicitly highlighted in the vignette (vs. a control vignette where employment status was not mentioned), which likely made it a salient identity cue (i.e., “in this situation, I am being positioned as unemployed”). This heightened salience may have made participants more responsive to rejection-letter cues (e.g., personalization and explanation), producing a clearer pattern or results. In Study 2, employment status was measured as a real-life characteristic (employed vs. unemployed and actively seeking work), meaning that “unemployed” likely bundled together additional sources of heterogeneity (e.g., differences in stigma internalization, perceived controllability of unemployment, and normative beliefs about unemployed people; Ho et al., 2011; Krug and Prechsl, 2020). In that context, it is theoretically consistent that employment status alone was not sufficient to reproduce the moderation that was suggested in hypotheses 4 and 5 (see also Sterud et al., 2025). Instead, the pattern emerged only among those who did not hold strongly negative attitudes toward unemployed people, because evaluative beliefs about unemployment can shape whether considerate cues are read as meaningful respect (vs. discounted or interpreted through a stigmatizing lens; see also Truxillo et al., 2002). Finally, the studies also differed in design features that can affect moderation detection: Study 2 was a single-wave study and did not include the control variables measured in the first wave of Study 1 (e.g., general attitudes toward AI, anthropocentric decision-making beliefs, organizational conspiracy beliefs). Omitting these controls may have left more unmodeled individual-difference variance in responses to AI decisions, making it harder for the employment-status moderator to appear as a standalone boundary condition and pushing the conditionality toward variables more proximal to unemployment-related interpretation (i.e., attitudes toward unemployment). Table 7 provides a concise overview of the proposed hypotheses and the extent to which they were supported across the two studies.

Theoretical implications

First and foremost, the present research extends the literature on AI in recruitment by shifting attention from AI-supported selection decisions to AI-mediated recruitment communication. Whereas previous research has primarily examined applicants' reactions to algorithmic screening, evaluation, and hiring decisions (e.g., Acikgoz et al., 2020; Langer et al., 2019; Mirowska and Mesnet, 2022; Pyle et al., 2024), as well as the role of explanations and justifications in improving reactions to AI-supported decisions (e.g., Köchling and Wehner, 2023; Olckers et al., 2022), our findings demonstrate that the way in which AI communicates those decisions also shapes applicant reactions. In doing so, the present research complements emerging work on generative AI in human resource management (e.g., Andrieux et al., 2024) by identifying mechanistic dehumanization and experienced respect as key psychological mechanisms through which respectful and personalized AI-generated rejection messages influence applicants' evaluations of the recruitment process. Together, these findings suggest that effective AI-mediated recruitment communication depends not only on what is communicated but also on how it is communicated. More broadly, our findings should be interpreted within the rapidly evolving landscape of generative AI in HRM. As large language models increasingly enable AI systems to produce natural, personalized, and context-sensitive recruitment communications, applicants' reactions may become shaped less by AI involvement per se than by the quality of the communication itself. Our findings suggest that even in this evolving context, respectful communication remains critical for reducing mechanistic dehumanization and fostering experienced respect.

Moreover, our findings advance mechanistic dehumanization theory (Haslam, 2006) by demonstrating that perceptions of mechanistic dehumanization are not only elicited by the presence of AI in recruitment but are shaped by how AI systems communicate unfavorable outcomes (i.e., rejection). Prior work has largely treated AI as inherently dehumanizing due to its lack of emotional capacity, flexibility, and intentionality (e.g., Fousiani et al., 2019; Haslam, 2006). Our results refine this view by showing that mechanistic dehumanization is sensitive to interactional cues embedded in organizational communication, such as personalization and explanation, even when the decision-maker is explicitly described as an AI system. This suggests that mechanistic dehumanization should be conceptualized not as a fixed response to automation, but as a relational judgment shaped by communicative design features.

Additionally, the findings contribute to anthropomorphism theory (Epley et al., 2007) by illustrating that human-like attributions to AI recruiters are contingent upon contextual cues that signal social responsiveness and moral consideration (Fousiani et al., 2026a). Whereas much prior work has focused on dispositional tendencies to anthropomorphize AI (e.g., individual differences in anthropocentric beliefs; Millet et al., 2023), the present study demonstrates that organizational communication can actively attenuate or exacerbate mechanistic dehumanization of AI agents. By linking rejection-letter design to perceptions of warmth, flexibility, and social consideration (Haslam, 2006), the findings underscore that anthropomorphism processes are shaped by organizational practices and communication systems rather than solely by user characteristics.

Furthermore, the present research extends applicant reactions theory (Gilliland et al., 2001; Hausknecht et al., 2004) into the domain of AI-enabled recruitment. Traditional models of applicant reactions emphasize interpersonal treatment and informational justice as key predictors of fairness perceptions, organizational attraction, and acceptance of selection outcomes (Konradt et al., 2017). Our findings show that these relational concerns remain highly relevant when decisions are communicated by AI systems. Importantly, the results demonstrate that applicants still infer intentionality, moral regard, and social valuation (i.e., human nature traits; Haslam, 2006) from AI-mediated communication, and that mechanistic dehumanization serves as a psychological bridge linking procedural cues (type of rejection letter) to outcome evaluations (prestige of the recruiter and experienced respect). This challenges implicit assumptions that AI decisions are processed as purely technical or impersonal events (Fritts and Cabrera, 2021) and instead positions AI recruiters as socially evaluated organizational actors (Fousiani et al., 2026a).

Finally, the findings across Studies 1 and 2 highlight the importance of applicant vulnerability and stigma in shaping responses to AI decision-making (Ho et al., 2011; Krug and Prechsl, 2020). The moderation effects observed in Study 1 among unemployed applicants align with stigma and social identity perspectives suggesting that marginalized groups are especially sensitive to cues of dignity, recognition, and respect (McKee-Ryan et al., 2005; Pohlan, 2019). Study 2 further refines this insight by showing that employment status alone (i.e., being unemployed or employed) is insufficient to explain variability in responses; rather, broader attitudes toward unemployment influence whether applicants interpret AI rejection as dehumanizing. This suggests that reactions to AI decision-makers are shaped by both structural position (employment status) and personal beliefs, offering a more nuanced, socially embedded understanding of AI perception. In a similar vein, our findings also contribute to the growing literature suggesting that algorithm aversion is not universal. Recent research indicates that individuals who have experienced higher levels of everyday discrimination may respond more positively to AI-supported recruitment decisions (Kr6ll et al., 2026), highlighting perceived discrimination as an important boundary condition that future research should examine.

Practical implications

The findings offer several actionable implications for organizations using AI in recruitment. First, organizations should recognize that rejection communication is not a neutral administrative step but a psychologically consequential interaction, especially when mediated by AI. Using highly standardized, typified rejection letters may inadvertently signal coldness and indifference, increasing dehumanization of the AI recruiter. Second, incorporating relatively simple design features, such as personalization and explanation, can reduce mechanistic dehumanization and support perceptions of professionalism (i.e., prestige) and experienced respect. These features are particularly important when communicating with applicants who may be more vulnerable, such as unemployed job seekers. Third, organizations should be cautious about assuming that AI communication affects all applicants uniformly. Applicants' beliefs and prior attitudes toward unemployment can shape how AI decisions are interpreted, suggesting that “one-size-fits-all” communication strategies may be insufficient in AI-enabled recruitment systems. Fourth, AI-mediated rejection communication may also have broader implications for organizations. Rejected applicants remain important organizational stakeholders and the way rejection decisions are communicated can influence perceptions of organizational fairness, respect, and professionalism. Consequently, organizations that rely on impersonal or highly standardized AI-generated rejection messages may inadvertently undermine their employer image, whereas AI systems that communicate in a more respectful and personalized manner may help organizations balance efficiency with a more positive applicant experience. However, although our findings suggest that more personalized and explanatory rejection letters of AI recruiters can reduce mechanistic dehumanization and improve applicants' evaluations (e.g., perceived respect and recruiter prestige), these benefits should not be taken to imply that AI recruitment systems can, or should, replace human recruiters. Even when AI communication is crafted to sound considerate, applicants may still experience the process as fundamentally impersonal, particularly because AI systems cannot genuinely engage in perspective-taking, moral judgment, or responsive dialogue when a decision is contested. In practice, organizations should therefore treat AI as a decision-support tool rather than a full substitute for human responsibility in rejection communication: applicants should have access to human-led channels for clarification and follow-up, opportunities to request feedback or correction of errors, and transparent information about how decisions were reached. In other words, “human-like” wording can improve the tone of AI rejection letters but it cannot replace the relational and accountability functions that human recruiters uniquely provide. Importantly, organizations should ensure that AI-enabled recruitment systems are deployed within appropriate governance frameworks. Current regulatory frameworks (e.g., the EU GDPR, the EU AI Act, and UK guidance on responsible AI in recruitment) increasingly require transparency regarding the use of AI, human oversight of decisions, and mechanisms through which applicants can request clarification or challenge decisions. Designing AI-generated rejection communications that are respectful, transparent, and accompanied by accessible human follow-up may therefore support not only applicant experiences but also responsible organizational AI governance.

Limitations

Several limitations of the present research should be acknowledged. First, although both studies used carefully constructed and realistic vignettes, they relied on scenario-based designs, which may underestimate the emotional intensity of real rejection experiences. Actual job rejection often carries higher stakes, prolonged uncertainty, and cumulative stress, particularly for unemployed applicants (Ho et al., 2011; McKee-Ryan et al., 2005). As a result, the observed effects may represent conservative estimates of how mechanistic dehumanization unfolds in real recruitment contexts. Second, key psychological processes were measured simultaneously, at the same time point, limiting strong causal claims about indirect effects (Podsakoff et al., 2003). Although the hypothesized ordering of variables is theoretically grounded, longitudinal or experimental manipulation of mechanistic dehumanization itself would further strengthen causal inference. Nevertheless, it should be noted that in exploratory analyses (in both studies) in which the proposed outcome variables (recruiter prestige and experienced respect) were specified as mediators and the proposed mediator (mechanistic dehumanization) was specified as the dependent variable, none of the conditional direct or indirect effects were significant. This pattern provides additional support for the theoretically proposed ordering of variables, suggesting that mechanistic dehumanization is more plausibly positioned as an antecedent rather than a consequence of prestige and respect perceptions. Third, an important limitation concerns inconsistencies in control variables across studies. Study 1 included measures of, and controlled for, general attitudes toward AI, anthropocentric decision-making beliefs, and organizational conspiracy beliefs, whereas Study 2 did not include these controls. Although this decision was driven by design considerations (e.g., due to survey length constraints), it limits direct comparability across studies and leaves open the possibility that unmeasured individual differences influenced responses in Study 2. Fourth, we did not measure participants' AI literacy, familiarity with generative AI, or their interpretation of the technical architecture underlying the “AI recruiter” label. Although the scenario explicitly described the role and functions assigned to the AI, participants may have differed in whether they imagined a rule-based system, a predictive algorithm, a generative AI system, or a combination of these technologies. Such interpretations could shape perceived agency, competence, and responsibility of AI-enabled recruiting systems. Future research should therefore measure and control for participants' AI literacy and familiarity. Fifth, the present studies focused exclusively on AI-generated rejection emails. Although email remains a common medium for communicating recruitment decisions, AI-enabled recruitment increasingly involves conversational interfaces such as chatbots and virtual assistants. Accordingly, our findings cannot generalize across different communication channels, such as interactive AI conversations or messaging platforms.

Finally, we should note that, across studies, the experimental vignette described an AI system as autonomously reviewing applications, evaluating interviews, and communicating hiring decisions. Although this design facilitated experimental control, it does not necessarily reflect current organizational practice, where AI-assisted recruitment is increasingly subject to requirements for transparency, meaningful human oversight, and opportunities for human review. Future research should therefore examine applicant reactions under recruitment scenarios that more closely reflect contemporary regulatory and organizational frameworks.

Future research directions

The present findings open several promising avenues for future research. Most importantly, future studies should move beyond vignette-based designs and examine AI-recruiters' rejection communication in real organizational settings. A particularly important next step would be a field experiment in collaboration with organizations that use AI-enabled recruitment systems. For example, rejected applicants could be randomly assigned to receive either typified or considerate AI-generated rejection letters following actual selection processes. Subsequent surveys could assess perceived dehumanization, organizational prestige, willingness to reapply, and employer reputation over time. Such a design would provide high ecological validity and allow researchers to examine longer-term consequences for employer branding and applicant pools. Second, future research should more systematically investigate individual difference moderators, including stigma sensitivity, and prior experiences with unemployment or AI-enabled decision-making. Integrating these variables across studies would help disentangle when reactions to AI recruiters are driven by situational cues vs. stable belief systems. Third, future work could explore whether similar dehumanization processes emerge in other AI-mediated organizational decisions, such as performance evaluations, promotion decisions, or termination notifications. Extending the framework beyond recruitment would clarify whether mechanistic dehumanization represents a general response to AI authority or a phenomenon uniquely amplified in high-stakes entry contexts (i.e., recruitment). Finally, future research could examine design interventions aimed at reducing mechanistic dehumanization, such as adaptive personalization, transparency about AI limitations, or hybrid AI–human communication models (Molenaar, 2022). Understanding which design features preserve dignity without creating false impressions of human involvement would be particularly valuable for ethical AI deployment.

Conclusion

Across two studies, the present research shows that the way AI recruiters communicate rejection decisions shapes applicants' experienced respect and the prestige they attribute to the recruiter, through mechanistic dehumanization of the recruiter. These effects are particularly consequential for applicants in vulnerable positions (e.g., unemployed; Study 1) and are further shaped by broader beliefs about unemployment (Study 2). Taken together, the findings highlight rejection communication as a critical design feature of AI-enabled recruitment systems and underscore the importance of considerate (rather than typified/standardized) communication in automated organizational decision-making.

References

• 1

  AcikgozY.DavisonK. H.CompagnoneM.LaskeM. (2020). Justice perceptions of artificial intelligence in selection. Int. J. Sel. Assess.28, 399–416. doi: 10.1111/ijsa.12306

  • CrossRef
  • Google Scholar
  • View reference in article

• 2

  AndrieuxP.JohnsonR. D.SarabadaniJ.Van SlykeC. (2024). Ethical considerations of generative AI-enabled human resource management. Organ. Dyn.53:101032. The article directly discusses generative AI affordances in HRM. doi: 10.1016/j.orgdyn.2024.101032

  • CrossRef
  • Google Scholar
  • View reference in article

• 3

  BankinsS.FormosaP.GriepY.RichardsD. (2022). AI decision making with dignity? contrasting workers' justice perceptions of human and AI decision making in a human resource management context. Inf. Syst. Front.24, 857–875. doi: 10.1007/s10796-021-10223-8

  • CrossRef
  • Google Scholar
  • View reference in article

• 4

  BarattucciM.RussoA.GrobelnyJ.SantisiG.RamaciT. (2025). Candidates' reactions to job application rejections at different phases of the recruitment process: the impact of employability and communication delays on perceived fairness and recruitment selection outcomes. J. Manag. Organ.31, 2341–2359. doi: 10.1017/jmo.2025.11

  • CrossRef
  • Google Scholar
  • View reference in article

• 5

  BastianB.HaslamN. (2010). Excluded from humanity: the dehumanizing effects of social ostracism. J. Exp. Soc. Psychol.46, 107–113. doi: 10.1016/j.jesp.2009.06.022

  • CrossRef
  • Google Scholar
  • View reference in article

• 6

  BrighamJ. C. (1971). Ethnic stereotypes. Psychol. Bull.76, 15–38. doi: 10.1037/h0031446

  • CrossRef
  • Google Scholar
  • View reference in article

• 7

  CortiniM.ConversoD.GalantiT.Di FioreT.Di DomenicoA.FantinelliS. (2019). Gratitude at work works! A mix-method study on different dimensions of gratitude, job satisfaction, and job performance. Sustainability11:3902. doi: 10.3390/su11143902

  • CrossRef
  • Google Scholar
  • View reference in article

• 8

  DietvorstB. J.SimmonsJ. P.MasseyC. (2015). Algorithm aversion: people erroneously avoid algorithms after seeing them err. J. Exp. Psychol.144, 114–126. doi: 10.1037/xge0000033

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 9

  EpleyN.WaytzA.CacioppoJ. T. (2007). On seeing human: a three-factor theory of anthropomorphism. Psychol. Rev. 114, 864–886. doi: 10.1037/0033-295X.114.4.864

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 10

  EvelethD. M.EvelethH. (2021). Job-seeker reactions to rejection emails: Int. J. Technol. Human Interact.17, 1–15. doi: 10.4018/IJTHI.2021070101

  • CrossRef
  • Google Scholar
  • View reference in article

• 11

  FahimY. A.HasaniI. W.KabbaS.RagabW. M. (2025). Artificial intelligence in healthcare and medicine: clinical applications, therapeutic advances, and future perspectives. Eur. J. Med. Res.30:848. doi: 10.1186/s40001-025-03196-w

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 12

  FousianiK.CremerL. R.ÇetinM.RaederS. (2026a). When algorithmic managers fail to fulfill their promises: the role of anthropomorphism in shaping justice perceptions. PLOS ONE21:e0340860. doi: 10.1371/journal.pone.0340860

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 13

  FousianiK.DimitropoulouP.MichaelidesM. P.Van PetegemS. (2016). Perceived parenting and adolescent cyber-bullying: examining the intervening role of autonomy and relatedness need satisfaction, empathic concern and recognition of humanness. J. Child Fam. Stud.25, 2120–2129. doi: 10.1007/s10826-016-0401-1

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 14

  FousianiK.MichaelidesM.DimitropoulouP. (2019). The effects of ethnic group membership on bullying at school: when do observers dehumanize bullies?J. Soc. Psychol.159, 431–442. doi: 10.1080/00224545.2018.1505709

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 15

  FousianiK.MichelakisG.MinnighP. A.De JongeK. M. M. (2024). Competitive organizational climate and artificial intelligence (AI) acceptance: the moderating role of leaders' power construal. Front. Psychol.15:1359164. doi: 10.3389/fpsyg.2024.1359164

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 16

  FousianiK.XuS.Armenta GutierrezB. M.SypesC. (2026b). Too old to telework? Age but not gender shapes hiring biases across telework and office settings. PLoS ONE21:e0340366. doi: 10.1371/journal.pone.0340366

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 17

  FousianiK.XuS.van ProoijenJ.-W. (2025). Leaders' power construal influences malevolent creativity: the mediating role of organizational conspiracy beliefs. J. Occup. Organ. Psychol.98:e70005. doi: 10.1111/joop.70005

  • CrossRef
  • Google Scholar
  • View reference in article

• 18

  FrittsM.CabreraF. (2021). AI recruitment algorithms and the dehumanization problem. Ethics Inf. Technol.23, 791–801. doi: 10.1007/s10676-021-09615-w

  • CrossRef
  • Google Scholar
  • View reference in article

• 19

  GedikliC.MiragliaM.ConnollyS.BryanM.WatsonD. (2023). The relationship between unemployment and wellbeing: an updated meta-analysis of longitudinal evidence. Eur. J. Work Organ. Psychol.32, 128–144. doi: 10.1080/1359432X.2022.2106855

  • CrossRef
  • Google Scholar
  • View reference in article

• 20

  GillilandS. W.GrothM.Baker IvR. C.DewA. E.PollyL. M.LangdonJ. C. (2001). Improving applicants' reactions to rejection letters: an application of fairness theory. Pers. Psychol.54, 669–703. doi: 10.1111/j.1744-6570.2001.tb00227.x

  • CrossRef
  • Google Scholar
  • View reference in article

• 21

  GrassiniS. (2023). Development and validation of the AI attitude scale (AIAS-4): a brief measure of general attitude toward artificial intelligence. Front. Psychol.14:1191628. doi: 10.3389/fpsyg.2023.1191628

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 22

  GurrT.Jungbauer-GansM. (2013). Stigma consciousness among the unemployed and prejudices against them: development of two scales for the 7th wave of the panel study “Labour market and social security (PASS)”. J. Labour Mark. Res.46, 335–351. doi: 10.1007/s12651-013-0144-z

  • CrossRef
  • Google Scholar
  • View reference in article

• 23

  HaslamN. (2006). Dehumanization: an integrative review. Pers. Soc. Psychol. Rev.10, 252–264. doi: 10.1207/s15327957pspr1003_4

  • CrossRef
  • Google Scholar
  • View reference in article

• 24

  HausknechtJ. P.DayD. V.ThomasS. C. (2004). Applicant reactions to selection procedures: an updated model and meta-analysis. Pers. Psychol.57, 639–683. doi: 10.1111/j.1744-6570.2004.00003.x

  • CrossRef
  • Google Scholar
  • View reference in article

• 25

  HenrichJ.Gil-WhiteF. J. (2001). The evolution of prestige: freely conferred deference as a mechanism for enhancing the benefits of cultural transmission. Evol. Hum. Behav.22, 165–196. doi: 10.1016/S1090-5138(00)00071-4

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 26

  HoV. T.WongS.-S.LeeC. H. (2011). A tale of passion: linking job passion and cognitive engagement to employee work performance. J. Manag. Stud.48, 26–47. doi: 10.1111/j.1467-6486.2009.00878.x

  • CrossRef
  • Google Scholar
  • View reference in article

• 27

  KakkarH.SivanathanN.GobelM. S. (2020). Fall from grace: the role of dominance and prestige in the punishment of high-status actors. Acad. Manag. J.63, 530–553. doi: 10.5465/amj.2017.0729

  • CrossRef
  • Google Scholar
  • View reference in article

• 28

  Koch-BayramI. F.KaibelC. (2024). Algorithms in personnel selection, applicants' attributions about organizations' intents and organizational attractiveness: an experimental study. Hum. Resour. Manag. J.34, 733–752. doi: 10.1111/1748-8583.12528

  • CrossRef
  • Google Scholar
  • View reference in article

• 29

  KöchlingA.WehnerM. C. (2023). Better explaining the benefits why AI? Analyzing the impact of explaining the benefits of AI-supported selection on applicant responses. Int. J. Sel. Assess.31, 45–62. doi: 10.1111/ijsa.12412

  • CrossRef
  • Google Scholar
  • View reference in article

• 30

  KonradtU.GarbersY.BögeM.ErdoganB.BauerT. N. (2017). Antecedents and consequences of fairness perceptions in personnel selection: a 3-year longitudinal study. Group Organ. Manag.42, 113–146. doi: 10.1177/1059601115617665

  • CrossRef
  • Google Scholar
  • View reference in article

• 31

  KristalA. S.NicksL.GloorJ. L.HauserO. P. (2022). Reducing discrimination against job seekers with and without employment gaps. Nat. Hum. Behav.7, 211–218. doi: 10.1038/s41562-022-01485-6

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 32

  KrugG.PrechslS. (2020). The role of social integration in the adverse effect of unemployment on mental health – Testing the causal pathway and buffering hypotheses using panel data. Soc. Sci. Res.86:102379. doi: 10.1016/j.ssresearch.2019.102379

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 33

  KröllH.FleißJ.EderM.LexE.ThalmannS. (2026). AI risk perceptions and the effect of AI Act safeguards: results from a representative population sample. Comput. Law Secur. Rev.64:106386. doi: 10.1016/j.clsr.2026.106386

  • CrossRef
  • Google Scholar
  • View reference in article

• 34

  LangerM.KönigC. J.PapathanasiouM. (2019). Highly automated job interviews: acceptance under the influence of stakes. Int. J. Sel. Assess.27, 217–234. doi: 10.1111/ijsa.12246

  • CrossRef
  • Google Scholar
  • View reference in article

• 35

  LeeM. K. (2018). Understanding perception of algorithmic decisions: fairness, trust, and emotion in response to algorithmic management. Big Data Soc.5:2053951718756684. doi: 10.1177/2053951718756684

  • CrossRef
  • Google Scholar
  • View reference in article

• 36

  LoggJ. M.MinsonJ. A.MooreD. A. (2019). Algorithm appreciation: people prefer algorithmic to human judgment. Organ. Behav. Hum. Decis. Process.151, 90–103. doi: 10.1016/j.obhdp.2018.12.005

  • CrossRef
  • Google Scholar
  • View reference in article

• 37

  McKee-RyanF.SongZ.WanbergC. R.KinickiA. J. (2005). Psychological and physical well-being during unemployment: a meta-analytic study. J. Appl. Psychol.90, 53–76. doi: 10.1037/0021-9010.90.1.53

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 38

  MilletK.BuehlerF.DuG.KokkorisM. D. (2023). Defending humankind: anthropocentric bias in the appreciation of AI art. Comput. Human Behav.143:107707. doi: 10.1016/j.chb.2023.107707

  • CrossRef
  • Google Scholar
  • View reference in article

• 39

  MirowskaA.MesnetL. (2022). Preferring the devil you know: potential applicant reactions to artificial intelligence evaluation of interviews. Hum. Resour. Manag. J.32, 364–383. doi: 10.1111/1748-8583.12393

  • CrossRef
  • Google Scholar
  • View reference in article

• 40

  MolenaarI. (2022). Towards hybrid human-AI learning technologies. Eur. J. Educ.57, 632–645. doi: 10.1111/ejed.12527

  • CrossRef
  • Google Scholar
  • View reference in article

• 41

  NewmanD. T.FastN. J.HarmonD. J. (2020). When eliminating bias isn't fair: Algorithmic reductionism and procedural justice in human resource decisions. Organ. Behav. Hum. Decis. Process.160, 149–167. doi: 10.1016/j.obhdp.2020.03.008

  • CrossRef
  • Google Scholar
  • View reference in article

• 42

  NobleS. M.FosterL. L.CraigS. B. (2021). The procedural and interpersonal justice of automated application and resume screening. Int. J. Sel. Assess.29, 139–153. doi: 10.1111/ijsa.12320

  • CrossRef
  • Google Scholar
  • View reference in article

• 43

  NørskovS.DamholdtM. F.UlhøiJ. P.JensenM. B.MathiasenM. K.EssC. M.et al. (2022). Employers' and applicants' fairness perceptions in job interviews: using a teleoperated robot as a fair proxy. Technol. Forecast. Soc. Change179:121641. doi: 10.1016/j.techfore.2022.121641

  • CrossRef
  • Google Scholar
  • View reference in article

• 44

  OchmannJ.MichelsL.TiefenbeckV.MaierC.LaumerS. (2024). Perceived algorithmic fairness: an empirical study of transparency and anthropomorphism in algorithmic recruiting. Inf. Syst. J.34, 384–414. doi: 10.1111/isj.12482

  • CrossRef
  • Google Scholar
  • View reference in article

• 45

  O'DonnellA. T.CorriganF.GallagherS. (2015). The impact of anticipated stigma on psychological and physical health problems in the unemployed group. Front. Psychol.6:1263. doi: 10.3389/fpsyg.2015.01263

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 46

  OlckersM.VidlerA.WalshT. (2022). What type of explanation do rejected job applicants want? Implications for explainable AI. arXiv. [preprint]. arXiv:2205.09649. doi: 10.48550/arXiv.2205.09649

  • CrossRef
  • Google Scholar
  • View reference in article

• 47

  ÖttingS. K.MaierG. W. (2018). The importance of procedural justice in human–machine interactions: intelligent systems as new decision agents in organizations. Comput. Human Behav.89, 27–39. doi: 10.1016/j.chb.2018.07.022

  • CrossRef
  • Google Scholar
  • View reference in article

• 48

  PodsakoffP. M.MacKenzieS. B.LeeJ.-Y.PodsakoffN. P. (2003). Common method biases in behavioral research: a critical review of the literature and recommended remedies. J. Appl. Psychol.88, 879–903. doi: 10.1037/0021-9010.88.5.879

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 49

  PohlanL. (2019). Unemployment and social exclusion. J. Econ. Behav. Organ.164, 273–299. doi: 10.1016/j.jebo.2019.06.006

  • CrossRef
  • Google Scholar
  • View reference in article

• 50

  PyleC.RoemmichK.AndalibiN. (2024). U.S. job-seekers' organizational justice perceptions of emotion AI-enabled interviews. Proc. ACM Hum.-Comput. Interact.8:42. doi: 10.1145/3686993

  • CrossRef
  • Google Scholar
  • View reference in article

• 51

  RoupaA.PatelarouA.GiakoumidakisK.FousianiK.MiliarakiM.StratidakiE.et al. (2025). Measurement of dehumanization, self-dehumanization, and empathy as mediating factors among healthcare professionals. Healthcare13:75. doi: 10.3390/healthcare13010075

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 52

  SakalakiM.RichardsonC.FousianiK. (2017). Is suffering less human? Distressing situations' effects on dehumanizing the self and others. Hell. J. Psychol.14, 39–63.

  • Google Scholar
  • View reference in article

• 53

  StarkeC.BaleisJ.KellerB.MarcinkowskiF. (2022). Fairness perceptions of algorithmic decision-making: a systematic review of the empirical literature. Big Data Soc.9:20539517221115189. doi: 10.1177/20539517221115189

  • CrossRef
  • Google Scholar
  • View reference in article

• 54

  SterudT.LundeL.-K.BergR.ProperK. I.AanesenF. (2025). Mental health effects of unemployment and re-employment: a systematic review and meta-analysis of longitudinal studies. Occup. Environ. Med.82, 343–353. doi: 10.1136/oemed-2025-110194

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 55

  ThominetL. (2020). We've selected a candidate who more closely fits our current needs: a genre analysis of academic job-refusal letters. J. Bus. Tech. Commun.34, 3–37. doi: 10.1177/1050651919874099

  • CrossRef
  • Google Scholar
  • View reference in article

• 56

  TruxilloD. M.BauerT. N.CampionM. A.ParontoM. E. (2002). Selection fairness information and applicant reactions: a longitudinal field study. J. Appl. Psychol.87, 1020–1031. doi: 10.1037/0021-9010.87.6.1020

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 57

  Van ProoijenJ.-W.De VriesR. E. (2016). Organizational conspiracy beliefs: implications for leadership styles and employee outcomes. J. Bus. Psychol.31, 479–491. doi: 10.1007/s10869-015-9428-3

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 58

  van QuaquebekeN.EckloffT. (2010). Defining respectful leadership: s it is, how it can be measured, and another glimpse at what it is related to. J. Bus. Ethics91, 343–358. doi: 10.1007/s10551-009-0087-z

  • CrossRef
  • Google Scholar
  • View reference in article

• 59

  Van QuaquebekeN.GerpottF. H. (2024). Artificial intelligence (AI) and workplace communication: promises, perils, and recommended policy. J. Leadersh. Organ. Stud.31, 375–381. doi: 10.1177/15480518241289644

  • CrossRef
  • Google Scholar
  • View reference in article

• 60

  WescheJ. S.SondereggerA. (2021). Repelled at first sight? Expectations and intentions of job-seekers reading about AI selection in job advertisements. Comput. Human Behav.125:106931. doi: 10.1016/j.chb.2021.106931

  • CrossRef
  • Google Scholar
  • View reference in article

• 61

  WilhelmyA.KleinmannM.KönigC. J.MelchersK. G.TruxilloD. M. (2016). How and why do interviewers try to make impressions on applicants? A qualitative study. J. Appl. Psychol.101, 313–332. doi: 10.1037/apl0000046

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 62

  WilhelmyA.KleinmannM.MelchersK. G.GötzM. (2017). Selling and smooth-talking: effects of interviewer impression management from a signaling perspective. Front. Psychol.8:740. doi: 10.3389/fpsyg.2017.00740

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article