Gender stereotypes are popular overgeneralized beliefs regarding supposedly typical traits of each gender (Eagly and Wood, 2012). Traditionally, traits that are related to agency (e.g., ambitious, assertive, competent, dominant, independent) are stereotypically associated with men (Bakan, 1966; Abele et al., 2008; Hentschel et al., 2019). Traits that are related to communion (e.g., caring, emotional, friendly, gentle, understanding) are stereotypically associated with women (Bakan, 1966; Abele et al., 2008; Hentschel et al., 2019). Negative consequences of gender stereotypes, for example, on professional and educational opportunities for women, or backlash effects on nonconforming individuals, have been demonstrated in many studies (e.g., Carr and Steele, 2010; Appel et al., 2011; von Hippel et al., 2011).

Research demonstrates that gender stereotyping applies not only to humans but also to robots (Eyssel and Hegel, 2012) and other non-human agents (Forlizzi et al., 2007). Even disembodied chatbots (Brahnam and De Angeli, 2012; Chaves and Gerosa, 2020) and computer voices (Nass et al., 1997) are perceived and categorized according to gender stereotypes that are traditionally attributed to men and women. Furthermore, it has been found that people tend to apply gender stereotypes if the field of application is traditionally associated with one gender and, in particular, if the chatbot does not act in accordance with its expected gender role (McDonnell and Baxter, 2019). Female-featured chatbots are more likely than male featured chatbots to be attributed with negative stereotypes (e.g., low competence) and to be the recipients of both implicit and explicit sexual language (Brahnam and De Angeli, 2012). Recent reports discuss anecdotal evidence about consumers who use sexually abusive language when addressing “female” speech assistants such as Siri and Alexa (Curry and Rieser, 2018; West et al., 2019). A study by Weisman and Janardhanan (2020) indicated that the use of female-sounding voice assistants—unlike male-sounding voice assistants—had both short- and long-term implications for the treatment of female subordinates in workplace settings: After using a female-featured voice assistant, help from female subordinates was expected to be given more quickly, they were penalized more harshly if they made mistakes, and were spoken to more impersonally than male subordinates. Further research is needed to better understand differential effects of AI design factors that influence the application of gender stereotypes to voice assistants, particularly due to the lack of prior research drawing a connection between the BPNT and gender stereotypes in technology design,

As outlined above, emerging evidence supports the relevance of research on impacts of gendered technology designs. Strategies that help to avoid the reproduction of societal stereotypes and gender biases in and through AI have received increased attention in recent years (cf. West et al., 2019). To promote gender equality and fairness in AI, suggestions include that datasets from which AI systems learn must be examined for inherent gender bias, AI literacy should be fostered especially among females, and teams developing AI systems should become more diverse and inclusive (cf. Jobin et al., 2019). Given that the stereotypical pairing of female gender markers (female names, voices) and passive, servant roles of AI-based speech assistants has been widely problematized (cf. West et al., 2019), counter-stereotypical designs of AI assistants could complement these strategies, since confrontation with stereotype-incongruent information has the potential to weaken a person’s access to stereotypic associations (e.g., Dasgupta and Asgari, 2004; Finnegan et al., 2015). Such non-conformity with stereotypes may be expressed in technology design, for example, by a female-featured speech assistant that is high in agentic traits. To the best of our knowledge, no previous study has yet examined user responses to stereotype-conforming vs. non-conforming design characteristics of gendered AI assistants in a controlled and randomized study, nor related them to individual differences in BPN Fulfillment based on user gender.

Supporting female agency and dismantling disadvantages that result from prevailing gender stereotypes have always been integral to the feminist movement (e.g., Farrell, 1995; Leaper and Arias, 2011). Feminist identity has been found to be associated with gender role atypicality and criticism of stereotypical depictions of the genders (van Breen et al., 2017). Individuals who identify with feminism also showed a greater likeability of describing themselves with agentic attributes (Saunders and Kashubeck-West, 2006) and to include agentic themes when narrating about their lives (Boytos et al., 2020). Drawing on these findings, in the context of the current study, we take the liberty of referring to a high-agency female AI assistant as the “feminist” condition in our experiment.

With regard to human preferences of voice gender, it should be noted that, starting in infancy, there seems to be a preference for the female voice (Standley and Madsen, 1990). Stereotypically, females are labeled as warm, tender and sensitive, whereas males are more likely to be described as dominant, assertive and forceful (Brems and Johnson, 1990; Guo et al., 2020). These stereotypical perceptions also align with the perception of female speech (Wiley and Eskilson, 1985; Ko et al., 2006), which is generally attributed with greater likeability (Krahé et al., 2021) and kindness (Ko et al., 2006) than male speech. Such socially favorable perceptions of female-sounding voices, in addition to early habituation to the mother’s voice, have been suggested as major drivers behind frequently found preferences for female speech. Alongside societal changes such as the growing role of fathers in early parenting or more balanced gender distributions in leadership positions, it could be assumed that disparities in evaluations of female and male voices might decrease over time. Today, however, customer preferences for female voices are still frequently put forward as an argument for the selection of female communicators in social interaction domains. This is reflected, for example, in a disproportionate share of female workers in call centers (Fernandez and Sosa, 2005), but also in the female-sounding default setting of many contemporary voice assistants.

For synthetic speech, previous research does not show consistent findings. In a study, both female and male participants preferred male over female synthetic voices and found the former to be more persuasive than the latter (Mullennix et al., 2003). The female synthetic voice was rated “less powerful,” “squeakier,” “softer,” and “faster” than the male voice. In contrast, participants rated the male voice as more positive. Nonetheless, other research in the context of human voices and chatbots found that male and female users differ in their preferences for dominance and persuasiveness. In a recent study (Guo et al., 2020), females had no preference for the gender of a chatbot that was trying to motivate them to pay overdue debt. Male participants, in contrast, were more likely to be persuaded by female chatbots. Furthermore, the study revealed that stereotypically feminine attributes, such as being gentle and warm, positively influenced male customers, while stereotypically male attributes (being forceful and assertive) had negative influence on both male and female customers. These results suggest favorable responses to behavior of chatbots that conform with traditional gender stereotypes. In addition to this, the results are in support of findings from previously outlined studies, proposing an innate human preference for female voices (Standley and Madsen, 1990; Ko et al., 2006; Krahé et al., 2021). Since the findings up to date are ambiguous, the results of the current study will shed further light on voice gender preferences for AI assistants.

Another line of research suggests same-gender preferences for synthesized speech (Lee et al., 2000, 2007) and thus supports the Similarity Attraction Theory by Byrne (1971), which posits that people are generally more attracted to others who are similar, rather than dissimilar, to themselves. These findings could be particularly relevant for the Satisfaction of the Need for Relatedness, as a study has shown that participants felt more psychological closeness when they interacted with a same-gender robot (Eyssel et al., 2012). In a study examining preferences of synthesized speech with children, the same effect was shown: children were asked to sit in front of a computer and listen to passages that would introduce them to different topics which were spoken by a synthetic voice. The topics were divided into stereotypically male and female topics, such as “make up” and “princesses” or “dinosaurs” and “knights.” Conforming to stereotypical portrayals of gender-specific interests, children preferred either the voice gender that matched the content of the topic or the voice gender that matched their own gender (Lee et al., 2007). These findings are also in line with the presumed desirability of same-gender interaction in HCI as proposed by Lee et al. (2000). However, since this evidence is not in line with other findings as outlined above (Mullennix et al., 2003; Guo et al., 2020; Krahé et al., 2021), additional research needs to be conducted to get a clearer picture of causal mechanisms behind differential responses to gendered technology. To date, it has not been investigated whether attraction to similarity—in this case, to a nonhuman agent whose gender a user identifies with—is associated with higher BPN Fulfillment. The current study will provide new insights on whether users have a higher Intention to Use and Relatedness Satisfaction for voice assistants that match their own gender.

With regard to individual differences of users’ preferences for agency levels, we argue that one important factor of influence is in fact existing gender stereotypes. For natural human speech, Carli (1990) demonstrated that male listeners were influenced more by tentatively speaking females, while females were influenced more by female speakers who spoke assertively. For synthetic speech, a study from Taiwan (Chang et al., 2018) revealed that participants exhibited a general preference for an assistive device having an extroverted female synthetic voice, with some individual differences depending on participants’ demographics or personality traits. On the contrary, a study by Nass et al. (1997) found that a female sounding computer voice was perceived as less competent in comparison with a male computer voice, and in addition to this, a female dominant voice was perceived more negatively than a male dominant voice. These findings provide evidence that gender stereotypes are indeed influential for the perception of synthetic speech. In line with gender-stereotypical behavior, female-featured AI assistants may be expected to act less dominant in comparison with male featured ones (Brems and Johnson, 1990; Guo et al., 2020). In reference to widely used systems such as Apple’s Siri and Amazon’s Alexa, there has been increased criticism in recent years over the reproduction of traditional gender stereotypes through the default combination of female gender markers, for example, female names and voices, with low levels of agency, conforming to stereotypical gender roles (Broverman et al., 1972) and outdated portrayals of women as passive servants (West et al., 2019).

Taken together, the majority of evidence either suggests a same-gender preference or a preference for female speech. With regard to agency levels, the small amount of existing research proposes that gender-stereotypical behavior—i.e., a combination of low agency or high communality with feminine features—is preferred by male users. Given the conflicting evidence (Nass et al., 1997; Chang et al., 2018), more research needs to be conducted to get a better understanding of individual differences with regard to gender and agency level preferences of speech assistants. Even though the outlined research does address gender preferences and differences with regard to voice gender and agency levels, none of the studies have focused on BPN Satisfaction in relation to these factors.

The finance coach’s agency level was either high or low. To achieve this manipulation, its services and approach to the customer differed depending on the agency condition. In the low-agency condition the finance coach stated, for example: “If you connect me with your account information, I can help you to manage deposits, planned savings and other tasks,” while the same service would be introduced by “I will autonomously connect with your account information. Without effort on your part, I will take care of deposits, planned savings and other tasks” in the high-agency condition (translated from the original German text). A total of eight phrases were adapted and differed between the conditions to achieve the required variations in agency-level. The full German text and the English translation of the text for both conditions are found in Appendix A.

We manipulated the synthetic voice of the finance coach to indicate male or female gender. The female version was based on a text-to-speech sample from the online platform ttsmp3.com (which offers text-to-speech downloads powered by AWS Polly) where we selected the voice “Vicki”. In order to keep the voice characteristics such as speaking rate and intonation as constant as possible, the recording of the female voice was modified by lowering the pitch to create a male voice with similar properties. Using the software package Ableton Live Suite (10.1.6) and the implementation ICRAM Trax Vocal Transformer, we converted the original text-to-speech output of the female voice into a male voice. With the help of the default filter settings “woman to man,” a frequency range of 146.67–270.00 Hz, a mean of 220.0 Hz, and a tuning (A) of 444.0 Hz were used. Additionally, the pitch was lowered by 8.4%. A total of four male versions were created that differed in the percentage by which the pitch was lowered and which were subsequently evaluated by six independent evaluators. The version that sounded the most male and was at the same time closest to the emotional expression and characteristics of the female voice was selected for this study. All recordings were cleaned with a custom noise removal filter using the software package Audacity and adjusted to the same volume by normalizing the amplitude (see other software descriptions above). All speech samples were in German.

The participants’ Need Fulfillment was measured with the BPNSFS. The scale used in this study was built based on a short scale created by Chen et al. (2015). The translation derived from the German adaptation by Heissel et al. (2019). Overall, 12 items were used (four per psychological need). Six items related to Need Satisfaction (two per psychological need) and six items were related to Need Frustration (two per psychological need). The items were also adapted to fit our context. All questions began with “If I used the finance coach for my personal banking operations…” To give an example, a Competence Need Satisfaction item was: “…I would feel competent to perform my banking operations,” and an Autonomy Need Frustration item was: “…I would feel pressured to do things that I would not have chosen myself.” (translated from the original German items). A full list of the adapted items used in this study is found in Appendix B. The participants rated their Need Satisfaction and Frustration for each item on a five-point Likert scale ranging from 1 (not at all) to 5 (very much). Based on previous literature (Chen et al., 2015; Longo et al., 2018), the frustration and satisfaction should be interpreted as separate constructs. In our case, the Frustration indices did not reach Cronbach’s alpha level > 0.66. The Autonomy Frustration index reached a Cronbach’s alpha level of 0.33, the Competence Frustration index reached a Cronbach’s alpha level of 0.66, and the Relatedness Frustration index reached a Cronbach’s alpha level of 0.42. Therefore, we will only continue with analyses of the Satisfaction indices. The Autonomy Satisfaction index reached a Cronbach’s alpha level of 0.74, the Competence Satisfaction index reached a Cronbach’s alpha level of 0.85, and the Relatedness Satisfaction index reached a Cronbach’s alpha level of 0.72. We can also report a good model fit based on Confirmatory Factor Analysis (CFA): RMSEA = 0.078; SRMR = 0.022; CFI = 0.988; TLI = 0.970.

We asked the participants to rate their perception of the finance coach’s dominance in a five-point semantic differential format. The items included were adapted from Mehrabian & Russel’s “dominance” component of their semantic differential scale (Mehrabian and Russell, 1974; see also Bradley and Lang, 1994 and Mara and Appel, 2015). We used the following four pairs of items: influential—influenced; controlling—controlled; dominant—submissive; autonomous—guided. The dominance scale showed good reliability with a Cronbach’s α of 0.72. Based on CFA fit indices, the model is not a good fit: RMSEA = 0.350; SRMR = 0.108; CFI = 0.742; TLI = 0.225. Therefore, we will not include the dominance perception scale in further analyses.

To measure competence perception of the finance coach (e.g., Bergmann et al., 2012), we asked participants to rate the finance coach on a five-point Likert scale—from 1 (not at all) to 5 (very)—using five traits (competent, experienced, intelligent, efficient, and capable). The competence perception scale showed high reliability with a Cronbach’s α of = 0.94. We can also report a good model fit based on CFA: RMSEA = 0.056; SRMR = 0.012; CFI = 0.997; TLI = 0.993.

Here, we used two of the ITU items from the Technology Acceptance Model (TAM3; Venkatesh and Bala, 2008) and slightly adapted them to the context of our experiment: “I can imagine using the finance-coach in the future.” and “I would like to be informed about products that are similar to the finance-coach.” Reliability of the items was high, with a Cronbach’s α of 0.87.

Accounting for human autonomy in the development and application of Artificial Intelligence is considered a necessity for public AI acceptance (e.g., HLEG-AI, 2018). The assumed relationship between design features of an AI assistant, Autonomy Need Satisfaction, and acceptance was also evident in our study. Overall, group comparisons supported H1 that a high-agency AI assistant leads to lower satisfaction of the Need for Autonomy on the side of the users. These results are consistent with previous research in which autonomously behaving chatbots were perceived as controlling or threatening (Maedche et al., 2019; Stein et al., 2019; Chaves and Gerosa, 2020). A significant moderated mediation model further demonstrated that men were more likely than women to see their Need for Autonomy undermined by the high-agency female finance coach (in comparison to the other versions) and that this perceived lack of Autonomy Satisfaction led to a lower Intention to Use.

Unlike female participants, men in our sample evaluated the two female finance coaches significantly differently, with their Autonomy Need being most satisfied by the low-agency female assistant. Consistent with previous studies (e.g., Guo et al., 2020), this shows a preference of male users for bots with female connotation. A reason why female features were precisely preferred in combination with a passive interaction style could be due to the fact that here, in conformity with traditional stereotypical gender roles (which still play a greater role for men on average, cf., Gustafsson Sendén et al., 2019; Hentschel et al., 2019), the male part is left to make important decisions himself, similar to the traditional image of the female secretary who assists a male manager (West et al., 2019; Weisman and Janardhanan, 2020).

Results from the female sample did not show any significant differences, suggesting that manipulations of the AI assistant have a smaller impact on female than on male Autonomy Satisfaction. Taking a look at the descriptive trends, however, it is interesting to note that women reported the highest mean score in Autonomy Satisfaction precisely in the condition rated lowest by men, namely the female high-agency coach, while they ascribed the least satisfaction to the male high-agency coach. This could indicate that women may feel their Autonomy Need more satisfied by AI assistants of their same gender and perhaps especially by such that correspond to a more contemporary image of women, with which feminism-oriented users could identify (cf., Sternadori and Abitbol, 2019). In section Gender Differences in the Evaluation of (Non-)Stereotypical AI Assistants, we discuss possible explanations for observed gender differences in the evaluation of the AI assistants in more detail. At the same time, we would like to emphasize that further studies are needed, especially to corroborate the tendencies observed for female participants.

In contrast with what we assumed in H2, there were no significant main effects of finance coach type on how much the Need for Relatedness was satisfied. In line with the Similarity Attraction Theory (Byrne, 1971) and with corresponding findings from the field of human-robot interaction (Eyssel et al., 2012), we had predicted that participants would feel more connected and thus experience greater satisfaction of their Need for Relatedness when the AI assistant matched their own gender. While group differences failed to reach a level of statistical significance, the descriptive mean values indicate the expected trend, as women gave the highest Relatedness Satisfaction scores to a female finance coach and men gave the highest scores to a male finance coach. Explanations for these tendencies may also be found in the developmental and social psychological literature on peer relations, which shows that children and adults are still more likely to form connections and spend time with same-gender peers than with other-gender peers, although it is assumed that more balanced orientations toward different genders could help expectations about gender roles become less rigid (Rubin et al., 2016; Bukowski and DeLay, 2020). Furthermore, some empirical work on parasocial relationships with media figures, for example with characters from science fiction movies (Hall, 2019), also suggests a positive influence of gender identification with screen characters on the perceived social connection with them. Parasocial relationships refer to the emotional attachment that people develop towards a media personality or fictional character (Horton and Richard Wohl, 1956) and have recently become a subject of research also in the context of human interactions with AI characters (Noor et al., 2021). We encourage further research with larger sample sizes to examine under which conditions Relatedness is a function of gender-matched human-machine interactions, including in more social application domains of AI than banking.

In support of H3, we found a positive correlation between the level of perceived Competence Satisfaction and ITU, highlighting the relevance of BPN for user acceptance. H4 assumed differences in Competence Satisfaction as a function of the AI assistant’s agency, while we did not predict whether participants would regard an agentic AI assistant as either a positive or negative influence on their experience of self-competence (Williams and Lillibridge, 1992). The found positive correlation between the perceived competence of the AI assistant and the fulfillment of the Need for Competence (as well as of the other needs) suggests that overall, our participants did not feel threatened by a competent finance coach, but rather perceived it as a supportive resource. This is in line with a recent study on the introduction of AI systems in the workplace, in which positive associations between a system’s usefulness, perceptions of the system as resource, and the fulfillment of psychological needs were found (Stamate et al., 2021). Generally, in the absence of significant main effects and hence in contradiction to H4, our data indicate that neither the degree to which an AI assistant was perceived as competent nor the Competence Need Satisfaction score on the side of the participants were determined by any of our experimental manipulations. However, taking a closer look at the descriptive statistics, we see similar trends as before. Women, in contrast with men, attributed the highest mean value of perceived competence to the female high-agency finance coach and also considered it to best fulfill their own Need for Competence, whereas men reported the lowest Competence Need Satisfaction for this version and the highest for the low-agency male finance coach. According to literature on gender stereotypes (Bakan, 1966; Abele et al., 2008; Hentschel et al., 2019), the agentic and competent traits are closely related to one another and have traditionally been more often attributed to men than to women, both by male and female raters (cf. Nass et al., 1997). The trends in our study results do not support this, but rather imply that women in our sample attributed more competence and more competence satisfaction potential to the female high-agency finance coach than to the male versions. However, these trends would need to be systematically investigated in follow-up studies with greater statistical power.

The finding that both male and female users evaluate an AI assistant with a female voice most favorable in our experiment could be explained by a general inclination of people to prefer female voices (Wiley and Eskilson, 1985; Ko et al., 2006). A number of previous studies suggest that participants of different genders perceive female voices more positively and, for example, attribute more likeability or kindness to them as compared to male voices (Ko et al., 2006; Krahé et al., 2021). However, the outcome that men prefer low agency in connection with the female voice, while women show a tendency to prefer high agency, requires closer examination. We propose to view each of the preferred AI assistants as either conforming or non-conforming to stereotypical gender roles. While the low-agency female finance coach, to whom male participants attributed the greatest need satisfaction on average, aligns with the traditional view of women as non-agentic (but communal), the high-agency female finance coach could be regarded as opposing this stereotype and thus also as a more “feminist” concept of a voice assistant. Following this notion, our data would suggest that different gender groups also differ in how they evaluate technology designs that either correspond or do not correspond to gender stereotypes. From this point of view, female participants in our study exhibited a tendency to be more open to the counter-stereotypical combination of highly agentic and female features, while men expressed a clear preference for the stereotype-conforming combination of low agency and female features.

These interpretations are in line with previous research. As the Role Congruity Theory (Eagly and Karau, 2002) suggests, a perceived incongruity between the stereotypical female gender role and agentic leadership roles can lead to a less favorable view of women. In order to be accepted as leaders, women might thus temper their agency with communal characteristics (Schock et al., 2019). Members of all gender groups have been shown to be susceptible to holding negative biases against role-incongruent individuals. In line with our own findings, several studies indicate that males tend to view agentic women more unfavorably than females do. For instance, if women violate traditional norms by presenting themselves as agentic in job resumés, men were found to perceive this particularly negatively (Tyler and McCullough, 2009). Men judged a fictional female politician less likable when the character used agentic (vs. communal) language, whereas there were no systematic differences in the judgments of female participants (Bray et al., 2020). In a survey with United States college students, only 29.6% of male respondents, in comparison with 40.6% of female respondents, somewhat or fully agreed with the statement “Women who are not feminine are good role models” (Duncan et al., 2019). In a study on social learning, even 5-8-year-old boys were shown to be relatively reluctant to accept facts by girls who were introduced as counter-stereotypical experts (e.g., knowing well about construction or football), whereas gender conformity did not matter that much for girls (Boseovski et al., 2016). This is consistent with findings that already at an early age, boys seem to respond less positively to deviations from stereotypical gender norms than girls (Blakemore, 2003). Linking this research to the current results, the existing empirical evidence is in support of the findings from the current study, suggesting that males, on average, are more prone to hold a role-congruent and gender-stereotypical view of women (and perhaps also of themselves). This may provide further explanation as to why male participants in the current study preferred the low-agency female finance coach and had the lowest rating for the non-conforming female high-agency finance coach.

In light of these results, and given the widespread human tendency to perceive machines as social agents (Reeves and Nass, 1996; Nass and Moon, 2000), it seems plausible that reluctance to counter-stereotypical characters in interpersonal relationships may be transferred to human-machine relationships as well. Thus, males, relative to females, might encounter non-conforming female AI assistants with more skepticism and feel more threatened than supported in their BPN. Taking responsibility for independent decisions is related to stereotypically male gender roles. Such gender stereotypes do not have to be static, but are influenced by actual and perceived changes in what roles different genders occupy in society (Eagly et al., 2000). With increasing efforts toward gender equality, a.o. supported by feminist initiatives, it was found that the stereotypical view of women increased in agentic traits in recent years, whereas the male stereotype showed less change in either agentic or communal traits (Gustafsson Sendén et al., 2019; Hentschel et al., 2019). Thus, the finding that male participants evaluated the low-agency female assistant most positively and the high-agency female assistant most negatively on average could also be due to the fact that men are still influenced—even more strongly than women—by stereotypical expectations related to their own gender category and thus cannot easily hand over the role of the active, decisive part to a (female-featured) interaction partner.

Consistent with previous research, women in our experiment did not appear to have as clear preferences as men (Tyler and McCullough, 2009; Boseovski et al., 2016; Bray et al., 2020). In the female subsample, the manipulations of the AI assistant did not lead to any statistically significant differences in judgments. However, there was a tendency for women to rate the counter-stereotypical high-agency assistant with female voice as most favorably across all outcome variables. Although this trend could still be a coincidental finding and thus must be carefully re-assessed by follow-up studies, it should be interesting to reflect on potential underlying motives of female participants. First, their seemingly greater openness to the use of high-agency finance coaches could be associated with individual differences in the perception of the AI assistant as either a competing authority that limits one’s own experience of autonomy, or as a supportive tool for the achievement of shared objectives (cf. Stamate et al., 2021). Against the backdrop of women’s still lower self-efficacy in financial matters (Amatucci and Crawley, 2011; Farrell et al., 2016), women in particular might assume an AI-based finance coach will support them in becoming more autonomous and independent of advice from third parties, rather than narrowing their space for decision-making. Further research could look at potential interaction effects between gender, application context, and perceptions of AI as a helpful resource in relation to BPN satisfaction. Second, women may be more attentive to stereotypical portrayals of their own gender. They might perceive such portrayals as outdated and rate them as less favorably accordingly. Such behavior would likely be associated with positive attitudes toward feminism and a stronger rejection of traditional gender norms. Including such constructs was beyond the scope of this paper, but should be considered in future work. Third, females might prefer the high-agency female assistant, because it is the one, they would most readily identify with. In past studies, women were found to have a stronger tendency to anthropomorphize machines (e.g., Abel et al., 2020) and to identify with fictional screen characters than men (Aytulun and Sunai, 2020). Women who view themselves as independent and assertive might therefore prefer to identify with an artificially intelligent counterpart that corresponds to a more contemporary image of women. This approach is supported by the observation that women in the high-agency female condition of our study also reported the highest Relatedness Need Satisfaction, although again these results merely represent descriptive trends.