Supporting Allyship in Virtual Collaboration with Artificial Intelligence: A Scenario-Based Study

1 Introduction

• Virtual collaboration creates barriers to participation, coordination, and shared information for people with disabilities, with potential effects on work and belonging.Examples include screen-sharing barriers for blind and low-vision professionals and sensory overload or inflexible communication norms for neurodivergent professionals.Source: §1 · p. 1
• Accessibility technologies such as screen readers and captions address technical access, but mixed-ability teamwork also depends on interpersonal and relational dynamics.Team practices include sharing access needs, coordinating support, and taking on accessibility-related labor [33].Source: §1 · p. 1
• AI's potential to shape coordination and group dynamics motivates asking whether it can support allyship, a context little examined in AI research.Prior work describes AI as influencing collaboration through co-creation and proactive participation in group discussion [32].Source: §1 · p. 2
• The study used four allyship scenarios with 18 mixed-ability team members to elicit reactions to novel AI designs.Participants included ten disabled and eight non-disabled people.Source: §1 · p. 2
• Participants valued AI that helped teammates express, interpret, and coordinate around access needs, but were wary of actions without consent, misrepresentation, or weakened interpersonal meaning.They anticipated benefits from reducing repeated disclosure and making support more actionable while preserving human judgment and relational control.Source: §1 · p. 2
• The authors propose allyship-support tools that scaffold how teammates recognize, communicate, coordinate, and respond to access needs.They frame this as a shift from technology providing individual access to technology scaffolding interpersonal and organizational processes.Source: §1 · p. 2

2.1 Accessible Virtual Collaboration

• Accessibility barriers can persist despite technical workarounds because access also depends on teammates sharing materials, describing visuals, and coordinating support.These practices are often ad hoc and uneven, depending on whether teammates know about needs, understand useful support, and are willing to act.Source: §2.1 · pp. 2–3

2.2 Allyship in Disability Contexts

• In mixed-ability teams, allyship requires disabled teammates to express needs and allies to have actionable ways to respond, yet stigma and uncertainty can impede both.Prior work identifies allyship as important to participation, coordination, and collaboration [33].Source: §2.2 · p. 3

2.3 Technology that Supports Interpersonal Dynamics in Collaboration

• Technology research has studied stable supporter roles and general collaboration dynamics, but not reciprocal allyship embedded in ongoing mixed-ability teamwork.AI research has found potential for social connection alongside concerns about trust, inclusivity, authenticity, and autonomy [32].Source: §2.3 · p. 3

3 Method

• Scenario-based design used hypothetical future-use narratives and storyboards to elicit expectations and concerns about technologies before implementation.A formative study informed the main scenarios.Source: §3 · p. 3

3.1 Formative Study: Informing the Creation of Scenarios

• The formative study refined the allyship problem space and narrowed AI design possibilities to support focused reactions to specific scenarios.Design dimensions included embodiment, visibility, and initiative; mismatches in initiative can undermine trust and acceptance.Source: §3.1 · p. 4

3.1.1 Participants and Procedure.

• Six participants—four disabled and two non-disabled—discussed past allyship challenges and reacted to storyboards comparing AI embodiment, visibility, and initiative.Researchers used semi-structured interviews and descriptive coding to capture preferences, concerns, and relevant experiences.Source: §3.1.1 · p. 4

3.1.2 Findings.

• Formative participants described difficulty articulating needs, discomfort requesting support, uncertainty about offering help, and concerns about overstepping or misinterpretation.They preferred text-based, private, passive AI unless users had specified preferences in advance.Source: §3.1.2 · p. 4

3.2 Scenarios

• The four main scenarios covered accessibility profiles, anonymous support requests, support-opportunity prompts, and phrasing alternatives for offering help.The scenarios reflected allyship situations and AI design preferences identified in the formative study.Source: §3.2 · pp. 4–5

3.3 Participant Recruitment and Demographics

• Eligibility required mixed-ability teamwork and regular virtual meetings; disabled participants also had to report disability-related challenges in virtual meetings.The main study included 18 participants: ten disabled and eight non-disabled.Source: §3.3 · p. 6

3.4 Procedure

• All participants followed the same protocol, discussing four accessible storyboards and their reactions, team contexts, role dynamics, and coordination implications.Sessions lasted 60 minutes; each scenario walkthrough and interview took around ten to fifteen minutes.Source: §3.4 · pp. 6–7

3.5 Data Analysis

• Researchers first descriptively coded concrete interaction accounts, then interpretively coded concepts and compared patterns across participants using a matrix.The analysis produced two high-level themes, six second-level themes, and 26 interpretive codes.Source: §3.5 · pp. 7–8

4.1.1 Addressing Breakdowns in Interpreting Needs and Responses.

• Teammates struggled to interpret needs and whether help would be welcome, while some disabled participants were unsure what support to request.Misread needs could lead to unwanted support, and uncertainty could persist even when people already used strategies to manage challenges.Source: §4.1.1 · p. 8
• Fear of professional judgment led some disabled participants to keep needs private and rely on AI individually rather than seek interpersonal support.That choice reduced disclosure but also limited opportunities for teammates to recognize and respond to needs.Source: §4.1.1 · p. 8

4.1.2 Reducing the Burden of Expressing and Coordinating Access

• Participants anticipated that accessibility profiles could reduce repeated explanations of preferences and needs across collaborators.Both disabled and non-disabled participants saw value in making interaction preferences visible up front.Source: §4.1.2 · p. 9
• Participants expected visible preferences to help teammates provide clearer, more appropriate support without sidestepping the issue.They viewed profiles as a way to translate stated needs into actionable support.Source: §4.1.2 · p. 9
• Preset support options could signal that accessibility was normalized in the team or organization, affecting disabled participants’ comfort selecting them.Comfort depended in part on what offering these options conveyed about the surrounding team context.Source: §4.1.2 · p. 9

4.2.1 Distributing the Labor of Allyship.

• Participants saw AI as a way to track access needs and distribute coordination beyond a single project manager or attentive ally.They noted that one coordinator could struggle as team demands grew, and that allies may lack time or resources to respond consistently.Source: §4.2.1 · p. 9
• Responsibility for initiating accessibility support varied with team role: an organizer proactively checked needs, while a collaborator deferred when another person was responsible.One participant described reaching out to all seminar attendees as an organizer, but not taking the lead as a fellow lab member.Source: §4.2.1 · p. 9

4.2.2 Accounting for Organizational and Social Constraints.

• Disclosure-support tools depended on a safe environment, while visible needs in larger teams did not ensure follow-through because responsibility could feel diffuse.Participants described fear that requests could affect workplace evaluation, and a contrast between larger teams where “someone will get to it” and smaller teams where someone was more likely to step in.Source: §4.2.2 · pp. 9–10

4.2.3 Preserving Control, Authenticity, and Human Judgment.

• Participants worried AI prompts could make help feel inauthentic if support seemed to reflect AI compliance rather than an ally’s own care and judgment.They valued technology easing interaction without removing the human-to-human act of offering and receiving support.Source: §4.2.3 · p. 10
• Participants feared AI-generated rephrasing could distort tone, intentions, or how a person came across.They emphasized control over self-expression and the importance of sensitivity to social nuance.Source: §4.2.3 · p. 10
• Some participants worried reliance on AI could substitute for learning to navigate difficult allyship conversations directly.One participant felt greater comfort with a manager after working through an initially uncertain conversation.Source: §4.2.3 · p. 10

4.2.4 Supporting Coordination Without Overstepping Social Bound-

• Participants were more comfortable with AI coordinating low-effort support, such as turning on captions, than requests requiring time commitments.For substantial demands on another person’s schedule, they preferred direct conversation so availability and expectations could be negotiated.Source: §4.2.4 · pp. 10–11
• Participants wanted AI to preserve situational consent, keeping decisions about initiating, offering, or accepting support with the people involved.They were more comfortable with a reminder than AI contacting teammates or scheduling support based on a saved preference.Source: §4.2.4 · p. 11

5 Discussion

• Participants valued AI designs that could reduce repeated disclosure, surface support opportunities, and distribute allyship labor, but were less receptive when AI risked misrepresenting intentions or weakening support’s authenticity.The discussion argues that AI should scaffold recognition, coordination, and reflection while preserving agency and relational control.Source: §5 · p. 11

5.1 Toward Allyship-Support Tools as a Class of Accessibility Technologies

• Allyship-support tools scaffold relational coordination around access needs, keeping collaborators central rather than focusing only on individual access or routing work to helpers.The tools address how teammates recognize needs, determine appropriate support, and coordinate without overstepping.Source: §5.1 · p. 11
• Broadening these tools beyond disability labels may accommodate varied support preferences, but risks leaving disability-specific barriers unnamed and reducing explicit discussion and shared responsibility.The discussion also cautions against automating away productive interpersonal friction through which teammates learn to support one another.Source: §5.1 · pp. 11–12

5.2 Design Guidelines for Allyship-Support Tools

• Design guidance calls for making needs actionable, reducing coordination burdens with user control, scaffolding human judgment, and preserving authentic interpersonal learning.Suggested mechanisms include structured preferences, opt-in requests, editable profiles, and prompts that support rather than replace socially meaningful decisions.Source: §5.2 · pp. 12–13

5.3 Considerations for AI in Allyship-Support Tools

• AI may help with context-sensitive coordination, but its role requires careful limits, human oversight, and attention to responsibility, intervention, and relationship effects.The discussion raises risks around accountability, authoritative or incomplete recommendations, and interventions that could short-circuit interpersonal learning.Source: §5.3 · pp. 13–14

5.4 Limitations and Future Work

• The findings reflect hypothetical scenarios emphasizing private, text-based, passive or consent-driven AI and recruits experienced with mixed-ability teams, not deployed use or less-experienced teams.The authors call for studies with prototypes or deployed systems in real teams and broader participant perspectives.Source: §5.4 · p. 14

6 Conclusion

• The scenario-based study involved 18 participants, 10 disabled and 8 non-disabled, considering AI designs for allyship in virtual collaboration.Source: §6 · p. 14
• Participants perceived AI as useful for expressing, interpreting, and coordinating around access needs, but less acceptable when it displaced user control, human judgment, or support’s relational meaning.Source: §6 · p. 14
• Allyship-support tools scaffold how teammates recognize, communicate, coordinate, and respond to access needs within ongoing collaborative relationships.Source: §6 · pp. 14–15

• Figure 2 presents the Accessibility Profile Support scenario.Source: Figure 2 · p. 18
• Figure 3 presents the Anonymous Support Request scenario.Source: Figure 3 · p. 19
• Figure 4 presents the Support Opportunity Prompts scenario.Source: Figure 4 · p. 20
• Figure 5 presents the Phrasing Alternatives for Offering Help scenario.Source: Figure 5 · p. 21

• Table 4 summarizes the cross-case matrix on reducing labor around expressing, interpreting, and coordinating access needs.Source: Table 4 · p. 22
• Table 5 summarizes the cross-case matrix on preserving human judgment, control, and team dynamics.Source: Table 5 · p. 22