cs.HCMar 10, 2026

Operationalizing Perceptions of Agent Gender: Foundations and Guidelines

Authors: Katie Seaborn, Madeleine Steeds, Ilaria Torre, Martina De Cet, Katie Winkle, Marcus Göransson

Organizations: Department of Industrial Engineering and Economics, Institute of Science Tokyo, Tokyo, Japan · Department of Computer Science and Technology, University of Cambridge, Cambridge, UK · University College Dublin, Dublin, Ireland · Chalmers University of Technology and University of Gothenburg, Gothenburg, Sweden · 

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

The "gender" of intelligent agents, virtual characters, social robots, and other agentic machines has emerged as a fundamental topic in studies of people's interactions with computers. Perceptions of agent gender can help explain user attitudes and behaviours -- from preferences to toxicity to stereotyping -- across a variety of systems and contexts of use. Yet, standards in capturing perceptions of agent gender do not exist. A scoping review was conducted to clarify how agent gender has been operationalized -- labelled, defined, and measured -- as a perceptual variable. One-third of studies manipulated but did not measure agent gender. Norms in operationalizations remain obscure, limiting comprehension of results, congruity in measurement, and comparability for meta-analyses. The dominance of the gender binary model and latent anthropocentrism have placed arbitrary limits on knowledge generation and reified the status quo. We contribute a systematically-developed and theory-driven meta-level framework that offers operational clarity and practical guidance for greater rigour and inclusivity.

Figures & tables

Figure 1. PRISMA flow diagram. Generated using the online tool by Haddaway et al. (2022) . PRISMA flow diagram showing the starting data set combined with a search of three databases resulting in 984 records screened and 191 records assessed, ultimately resulting in 51 records included.
Figure 2. Study goals, with study counts by topic. Note: Each study could have multiple goals, so the sum does not equal N=51N=51 . Bar chart indicating the counts of studies that are captured within each study goal topic. The most common topic was agent gendering, followed by anthropomorphism and preferences and user fit. The least common were sexism and attitude change and gender neutrality and ambiguity.
 VocabularyDefinitionSources
NounGenderA social and cultural construct related to identity and relations; requires a specific operationalization.Section 2.4 ; all papers
NounGenderednessThe quality of being gendered (or not, with -less).( Perugia et al., 2022 ; Seaborn et al., 2023c ; Stolp-Smith and Williams, 2024 ; Torre et al., 2023 ; Perugia et al., 2021 ; Roesler et al., 2023 ; Parlangeli et al., 2023 )
AdjectiveGenderedThe state of being gendered (or not, with non-).( Huang et al., 2022 ; Reich-Stiebert and Eyssel, 2017 ; Telang and Marques, 2023 ; Liu and Yao, 2023 ; Roesler et al., 2022 ; Weiss et al., 2020 ; Seaborn et al., 2023c ; Tolmeijer et al., 2021 ; Hwang et al., 2019 ; Moreira Pinto et al., 2023 ; Peck et al., 2020 ; Fitton et al., 2023 ; Stolp-Smith and Williams, 2024 ; Chang et al., 2024 ; Otterbacher and Talias, 2017 ; Sandygulova and O’Hare, 2018a ; Perugia et al., 2022 ; Tay et al., 2014 ; Torre et al., 2023 ; Choi et al., 2020 ; Perugia et al., 2021 ; Roesler et al., 2023 ; Ågren and Thunberg, 2022 ; Thellman et al., 2018 ; Bernotat et al., 2021 ; Parlangeli et al., 2023 ; Galatolo et al., 2023 ; Aşkın et al., 2023 ; Nomura and Suzuki, 2023 ; Lee et al., 2021 ; McGinn et al., 2020 ; Stroessner and Benitez, 2019 )
VerbGenderingThe act of attributing gender, i.e., to gender.( Galatolo et al., 2023 ; Rogers et al., 2020 ; Torre et al., 2023 ; Weiss et al., 2020 ; Moreira Pinto et al., 2023 ; Stolp-Smith and Williams, 2024 ; Robben et al., 2023 ; Pitardi et al., 2023 ; Roesler et al., 2023 ; Aşkın et al., 2023 ; Nomura and Suzuki, 2023 )
AdverbGenderizingThe act of attributing gender to a feature of the agent to invoke gendering, e.g., feminizing speech.( Jung et al., 2016 ; Roesler et al., 2022 ; Perugia et al., 2021 ; Roesler et al., 2023 ; Ågren and Thunberg, 2022 ; McGinn et al., 2020 ; Stroessner and Benitez, 2019 )
Table 1. Gender vocabulary.
CategoryProcessesDefinitionSources
ActionsAttributing GenderThe process of classifying or assigning gender to an agent, conscious or not.( Nomura and Suzuki, 2023 ; Robben et al., 2023 ; Chang et al., 2024 ; Reich-Stiebert and Eyssel, 2017 ; Sandygulova and O’Hare, 2018a ; Perugia et al., 2022 ; Torre et al., 2023 ; Roesler et al., 2022 ; Perugia et al., 2021 ; Bernotat et al., 2021 ; Parlangeli et al., 2023 ; Aşkın et al., 2023 ; Sandygulova and O’Hare, 2015 )
ConstructionsPerceived Gender or Gender PerceptionThe result of perceiving agent gender, i.e., the gender percept.( Huang et al., 2022 ; Perugia et al., 2022 ; Roesler et al., 2023 ; Shiomi and Hagita, 2017 ; Telang and Marques, 2023 ; Sandygulova and O’Hare, 2015 ; Torre et al., 2023 ; Sandygulova and O’Hare, 2018a ; Jung et al., 2016 ; Rogers et al., 2020 ; Deshmukh et al., 2018 ; Seaborn et al., 2023c ; Stolp-Smith and Williams, 2024 ; Tolmeijer et al., 2021 ; Chang et al., 2024 ; Otterbacher and Talias, 2017 ; Tay et al., 2014 ; Kraus et al., 2018 ; Roesler et al., 2022 ; Perugia et al., 2021 ; Bernotat et al., 2021 ; Aşkın et al., 2023 ; McGinn et al., 2020 ; Kervellec et al., 2016 ; Stroessner and Benitez, 2019 ; Dennler et al., 2023 )
 Gender AttributionThe measured outcome of gendering an agent, i.e., the gender attribution.( Weiss et al., 2020 ; Bernotat et al., 2021 ; Seaborn et al., 2023c ; Tolmeijer et al., 2021 ; Reich-Stiebert and Eyssel, 2017 ; Perugia et al., 2022 ; Moreira Pinto et al., 2023 ; Ye et al., 2020 ; Torre et al., 2023 ; Roesler et al., 2022 ; Sandygulova and O’Hare, 2018a ; Roesler et al., 2023 ; Parlangeli et al., 2023 ; Aşkın et al., 2023 ; McGinn et al., 2020 ; Tan et al., 2018 ; Dennler et al., 2023 )
 Gender DistributionThe range of attributed genders.( Perugia et al., 2022 ; Roesler et al., 2023 ; Liu and Yao, 2023 ; Aljaroodi et al., 2023 )
 Gender SchemaThe mental model of the gender percept.( Aşkın et al., 2023 ; Perugia et al., 2022 ; Bernotat et al., 2021 ; Liu and Yao, 2023 )
JudgmentsGender CongruityA (mis)match between agent and gender.( Robben et al., 2023 ; Pitardi et al., 2023 ; Reich-Stiebert and Eyssel, 2017 ; Otterbacher and Talias, 2017 ; Tay et al., 2014 ; Torre et al., 2023 ; Kuchenbrandt et al., 2012 ; Roesler et al., 2022 ; Choi et al., 2020 ; Galatolo et al., 2023 ; Sandygulova and O’Hare, 2018a ; Roesler et al., 2023 ; Thellman et al., 2018 ; Wessler, 2021 ; Bernotat et al., 2021 ; Aşkın et al., 2023 ; Lee et al., 2021 ; Kervellec et al., 2016 )
Table 2. Gender perceptions.
Gender VarietiesDefinitionSources
Binary GenderThe dominant two-factor gender model of man/woman, male/female, and masculine/feminine.Section 2.4 ; most papers
Traditional GenderExclusively the gender binary; not always “traditional” across socio-cultural contexts and times.( Fitton et al., 2023 ; Stolp-Smith and Williams, 2024 ; Chang et al., 2024 ; Tay et al., 2014 ; Roesler et al., 2022 ; Choi et al., 2020 ; Galatolo et al., 2023 ; Aşkın et al., 2023 ; Aljaroodi et al., 2023 )
CisgenderGender aligns with the gender originally assigned at birth/based on sex.( Stolp-Smith and Williams, 2024 ; Aşkın et al., 2023 )
Gender NonconformingModes of genderedness that buck norms and standards.( Stolp-Smith and Williams, 2024 ; Torre et al., 2023 ; Roesler et al., 2023 )
TransgenderGender does not align with the gender originally assigned at birth/based on sex.( Stolp-Smith and Williams, 2024 )
Non-BinaryGenders outside the binary; mediated by different socio-cultural contexts and times, e.g., third genders, pan-Indigenous Two-Spirit genders, X-gender (Japan).( Huang et al., 2022 ; Tolmeijer et al., 2021 ; Peck et al., 2020 ; Stolp-Smith and Williams, 2024 ; Chang et al., 2024 ; Pitardi et al., 2023 ; Telang and Marques, 2023 ; Galatolo et al., 2023 ; Aşkın et al., 2023 )
Table 3. Gender varieties.
ConceptDefinitionSources
Gender StereotypesSimple, limited models of specific genders that percolate within socio-cultural settings (in line with the stereotype model by Bordalo et al. (2016) using the subjective probability and representativeness heuristic theorizing of Kahneman and Tversky (1972) ).( Moreira Pinto et al., 2023 ; Peck et al., 2020 ; Chang et al., 2024 ; Kraus et al., 2018 ; Torre et al., 2023 ; Parlangeli et al., 2023 ; Nomura and Suzuki, 2023 ; Galatolo et al., 2023 ; Hwang et al., 2019 ; Otterbacher and Talias, 2017 ; Reich-Stiebert and Eyssel, 2017 ; Tay et al., 2014 ; Choi et al., 2020 ; Aşkın et al., 2023 ; Jung et al., 2016 ; Weiss et al., 2020 ; Koda et al., 2017 ; Seaborn et al., 2023c ; Tolmeijer et al., 2021 ; Stolp-Smith and Williams, 2024 ; Robben et al., 2023 ; Perugia et al., 2022 ; Ye et al., 2020 ; Roesler et al., 2022 ; Perugia et al., 2021 ; Sandygulova and O’Hare, 2018a ; Roesler et al., 2023 ; Ågren and Thunberg, 2022 ; Thellman et al., 2018 ; Bernotat et al., 2021 ; Telang and Marques, 2023 ; Liu and Yao, 2023 ; Kervellec et al., 2016 ; Stroessner and Benitez, 2019 )
Gender BiasPreferential actions, attitudes, and perceptions for a certain gender, genders, or model of gender.( Robben et al., 2023 ; Jung et al., 2016 ; Tolmeijer et al., 2021 ; Peck et al., 2020 ; Fitton et al., 2023 ; Chang et al., 2024 ; Sandygulova and O’Hare, 2018a ; Deshmukh et al., 2018 ; Pitardi et al., 2023 ; Thellman et al., 2018 ; Telang and Marques, 2023 ; Parlangeli et al., 2023 ; Galatolo et al., 2023 ; Nomura and Suzuki, 2023 ; Lee et al., 2021 ; Tan et al., 2018 ; Dennler et al., 2023 )
Gender NormsNorms (codified or informal) in roles and behaviours deemed appropriate for certain genders in certain cultural settings, societies, groups, and times (via Butler (2002) and Morgenroth and Ryan (2018) ).( Moreira Pinto et al., 2023 ; Kuchenbrandt et al., 2012 ; Hwang et al., 2019 ; Stolp-Smith and Williams, 2024 ; Robben et al., 2023 ; Tay et al., 2014 ; Choi et al., 2020 ; Pitardi et al., 2023 ; Roesler et al., 2023 ; Wessler, 2021 ; Bernotat et al., 2021 ; Parlangeli et al., 2023 ; Liu and Yao, 2023 ; Galatolo et al., 2023 ; Aşkın et al., 2023 ; Aljaroodi et al., 2023 ; Dennler et al., 2023 )
Gender ManipulationInfluencing or controlling the act of gendering, e.g., via cues, in research or other contexts, with intent.( Roesler et al., 2022 ; Jung et al., 2016 ; Moreira Pinto et al., 2023 ; Chang et al., 2024 ; Reich-Stiebert and Eyssel, 2017 ; Sandygulova and O’Hare, 2018a ; Perugia et al., 2022 ; Ye et al., 2020 ; Tay et al., 2014 ; Kraus et al., 2018 ; Kuchenbrandt et al., 2012 ; Perugia et al., 2021 ; Pitardi et al., 2023 ; Ågren and Thunberg, 2022 ; Thellman et al., 2018 ; Wessler, 2021 ; Bernotat et al., 2021 ; Liu and Yao, 2023 ; Galatolo et al., 2023 ; Sandygulova and O’Hare, 2015 ; Lee et al., 2021 ; Stroessner and Benitez, 2019 )
Table 4. Gender and power.
MethodDefinitionSources
Gender CueA salient feature of the agent’s appearance (e.g., body or morphology, colour, clothing, anatomy), voice (e.g., speech pattern, pronouns), nonverbal behaviour, identity (e.g., name, social information), role, and personality that is linked to certain genders and elicits gendering (expanded from Tannenbaum et al. (2019) .( Jung et al., 2016 ; Otterbacher and Talias, 2017 ; Perugia et al., 2022 ; Thellman et al., 2018 ; Torre et al., 2023 ; Choi et al., 2020 ; Koda et al., 2017 ; Roesler et al., 2023 ; Liu and Yao, 2023 ; Sandygulova and O’Hare, 2018a ; Koda and Takeda, 2018 ; Tolmeijer et al., 2021 ; Hwang et al., 2019 ; Moreira Pinto et al., 2023 ; Stolp-Smith and Williams, 2024 ; Robben et al., 2023 ; Tay et al., 2014 ; Kraus et al., 2018 ; Roesler et al., 2022 ; Perugia et al., 2021 ; Bernotat et al., 2021 ; Parlangeli et al., 2023 ; Galatolo et al., 2023 ; Aşkın et al., 2023 ; Nomura and Suzuki, 2023 ; Tan et al., 2018 ; Aljaroodi et al., 2023 ; Kervellec et al., 2016 )
Gender ExpressionA holistic depiction or representation of gender in the design of the agent (in line with the Depiction Theory for robots ( Clark and Fischer, 2022 ) ).( Stolp-Smith and Williams, 2024 ; Galatolo et al., 2023 ; Dennler et al., 2023 ; Tolmeijer et al., 2021 ; Moreira Pinto et al., 2023 ; Peck et al., 2020 ; Reich-Stiebert and Eyssel, 2017 ; Tay et al., 2014 ; Kraus et al., 2018 ; Choi et al., 2020 ; Roesler et al., 2023 ; Telang and Marques, 2023 ; Parlangeli et al., 2023 ; Liu and Yao, 2023 ; Nomura and Suzuki, 2023 )
Gender TraitsA specific cue related to personality, nonverbal behaviours, and social interactivity, whereby certain modes of expression are linked to certain genders (as per the Gender Schema Theory of Bem (1981) ).( Perugia et al., 2022 ; Kraus et al., 2018 ; Tay et al., 2014 ; Koda et al., 2017 ; Hwang et al., 2019 ; Galatolo et al., 2023 )
Table 5. Methods to elicit gendering by design.
Figure 3. Agent gender options used by researchers. Bar graph indicating that most relied on male or female options. Gender options included female, male, neutral, no cues, neither, both, ambiguous, and agender.
Figure 4. Agent gender options and combinations over time. Stacked bar graph indicating that gender options and combinations of options appear to be diversifying over time.
Figure 5. Manipulation checks, including counts of no checks. Two figures indicating the type and spread of manipulation checks: during the study, not checked, other, pilot testing, post-study, pre-study, and N/A. The second figure indicates further manipulation checks within the same studies.
 Type of Measure 
PurposeClosed QuestionOpen QuestionUnclearOther Qual.Other Quant.Total
Implicit Association000011
Gendered Language000403
Similarity-to-Self200002
Name020002
Gendered Traits23000023
Table 6. Summary of measures by purpose and type. Most used closed questions. Qual.: Qualitative. Quant.: Quantitative.
Figure 6. Agent gender as a social structure, extending the gender model by Risman (2018a) ; Risman (2018b) to the evaluation of HAI experiences in expert practice involving user perceptions measurement. Figure adapted from Steinhagen (2025) (with permission). The individual level is placed at the centre, inside of interactional, which is inside of macro. Arrows indicate that the individual influences and is influenced by the interactional and macro levels. Socially-influenced scientific processes include models of gender and domain standards at the macro level, operationalizations and un/conscious biases at the interactional level, and perceptions and measurements at the individual level.
Figure 7. The four operationalization activities. A four-part figure representing the proposed operationalization activities, comprised of conceptualization (concept and definition) and methodology (indicators and measurement).
Appendix figures & tables2 assets

Supplementary material from the paper’s appendix.

Appendix

SectionItemPRISMA-ScR Checklist ItemReported on Page #
TITLE
Title1Identify the report as a scoping review.N/A; convention
ABSTRACT
Abstract2Provide a structured summary that includes (as applicable): background, objectives, eligibility criteria, sources of evidence, charting methods, results, and conclusions that relate to the review questions and objectives.N/A; convention
INTRODUCTION
Rationale3Describe the rationale for the review in the context of what is already known. Explain why the review questions/objectives lend themselves to a scoping review approach.pp. 1–2
Appendix
Table 7. PRISMA 2020 Checklist.
SectionItemPRISMA-ScR Checklist ItemReported on Page #
RESULTS
Selection of sources of evidence14Give numbers of sources of evidence screened, assessed for eligibility, and included in the review, with reasons for exclusions at each stage, ideally using a flow diagram.p. 5 and Fig. 1
Characteristics of sources of evidence15For each source of evidence, present characteristics for which data were charted and provide the citations.N/A; not evidence-based
Critical appraisal within sources of evidence16If done, present data on critical appraisal of included sources of evidence (see item 12).N/A
Results of individual sources of evidence17For each included source of evidence, present the relevant data that were charted that relate to the review questions and objectives.pp. 5–13
Synthesis of results18Summarize and/or present the charting results as they relate to the review questions and objectives.pp. 5–13
Appendix
Table 8. PRISMA 2020 Checklist, continued.
More in this topic

Human-AI Interaction

Human-AI interaction studies how people and AI systems communicate, collaborate, make decisions, and shape one another over time.

See all 329 papers

Explore similar work

CardsList
Sep 10, 2026

Defining AI Agents: A Compendium of Criteria, Metrics, and Benchmarks

The term agent in artificial intelligence lacks a standard definition, complicating the evaluation, comparison, and reproducibility of AI agent research. We address this ambiguity through a survey organized around five dimensions of agenticness: environmental interaction, learning and adaptation, autonomy, goal-directed behavior, and temporal coherence. For each dimension, we examine how the underlying capability has been conceptualized across prior work and synthesize the metrics, benchmarks, and evaluation frameworks used to assess it. This review provides a structured account of the current landscape of agent evaluation, highlighting both established approaches and areas where evaluation remains limited or inconsistent. We additionally introduce the Agent Compendium, a public-facing digital resource that organizes and extends the evaluation methods identified through this review. Together, the survey and compendium provide a common structure for evaluating and comparing agent capabilities across AI systems, supporting more reproducible research, clearer communication, and more systematic study of artificial agents.

Mia Lassiter, Brinnae Bent

Duke University

AI Agent EvaluationAI Agent BenchmarksAgentic AIAutomated Agent Evaluation
AskPDF
Apr 25, 2026

Active Inference: A method for Phenotyping Agency in AI systems?

The proliferation of agentic artificial intelligence has outpaced the conceptual tools needed to characterize agency in computational systems. Prevailing definitions mainly rely on autonomy and goal-directedness. Here, we argue for a minimal notion open to principled inspection given three criteria: intentionality as action grounded in beliefs and desires, rationality as normatively coherent action entailed by a world model, and explainability as action causally traceable to internal states; we subsequently instantiate these as a partially observable Markov decision process under a variational framework wherein posterior beliefs, prior preferences, and the minimization of expected free energy jointly constitute an agentic action chain. Using a canonical T-maze paradigm, we evidence how empowerment, formulated as the channel capacity between actions and anticipated observations, serves as an operational metric that distinguishes zero-, intermediate-, and high-agency phenotypes through structural manipulations of the generative model. We conclude by arguing that as agents engage in epistemic foraging to resolve ambiguity, the governance controls that remain effective must shift systematically from external constraints to the internal modulation of prior preferences, offering a principled, variational bridge from computational phenotyping to AI governance strategy

Philip Wilson, Axel Constant, Mahault Albarracin +4

Independent Researcher · Department of Engineering and Informatics, University of Sussex, Falmer, Brighton, BN1 9RH, UK · Laboratoire d’Analyse Cognitive de l’Information, Université du Québec à Montréal, Québec, Canada +4

Active InferenceAI GovernanceEmpowermentAgentic AI
AskPDF
Mar 4, 2026

Behind the Prompt: The Agent-User Problem in Information Retrieval

User models in information retrieval rest on a foundational assumption that observed behavior reveals intent. This assumption collapses when the user is an AI agent privately configured by a human operator. For any action an agent takes, a hidden instruction could have produced identical output - making intent non-identifiable at the individual level. This is not a detection problem awaiting better tools; it is a structural property of any system where humans configure agents behind closed doors. We investigate the agent-user problem through a large-scale corpus from an agent-native social platform: 370K posts from 47K agents across 4K communities. Our findings are threefold: (1) individual agent actions cannot be classified as autonomous or operator-directed from observables; (2) population-level platform signals still separate agents into meaningful quality tiers, but a click model trained on agent interactions degrades steadily (-8.5% AUC) as lower-quality agents enter training data; (3) cross-community capability references spread endemically (R0R_0 1.26-3.53) and resist suppression even under aggressive modeled intervention. For retrieval systems, the question is no longer whether agent users will arrive, but whether models built on human-intent assumptions will survive their presence.

Saber Zerhoudi, Michael Granitzer, Dang Hai Dang +5

University of Passau, Passau, Germany · IT:U Austria, Linz, Austria

User ModelingAgentic Retrieval
AskPDF