This work exists within the wider context of the CommonGround project, which aimed to coproduce an online peer support and psychoeducation platform and assess the feasibility and acceptability of the intervention in a randomized controlled trial (fRCT [feasibility randomized controlled trial] [42]; ClinicalTrials.gov: NCT06222346). The CommonGround project is underpinned by a pragmatic mixed methods research paradigm with community participation, which prioritizes what works, for whom, and in what context. The fRCT was delivered remotely and randomized 83 participants with long-term physical health conditions and subthreshold depression to the CommonGround intervention (ie, community members) and 42 participants to the control condition of National Health Service Mental Health webpages for 3 months (findings published in full elsewhere; in press). A group of 187 people with long-term physical health conditions who were interested in managing their mental health better, but did not have subthreshold depression, were also onboarded onto the CommonGround platform to increase the size of the community (community engagement cohort [29]). A nested qualitative process evaluation examined the experiences of community members with long-term physical health conditions and moderators. The qualitative process evaluation adopts an interpretivist-constructivist approach that assumes realities are socially constructed through interactions with others in the cultural context in which they are situated. This paper is reported within the SRQR (Standards for Reporting Qualitative Research) guidelines as per EQUATOR (Enhancing the Quality and Transparency of Health Research) recommendations (Multimedia Appendix 1 [43]).

This study involves data from the nested qualitative study of part of a larger feasibility randomized control trial that was conducted remotely at King’s College London (KCL). The trial and nested qualitative study received ethical approval from the South Central - Oxford C Research Ethics Committee (IRAS ID 328175). The protocol was preregistered (ClinicalTrials.gov: NCT06222346) [29] and outlined the larger feasibility randomized controlled trial and the nested qualitative study (ie, community member and control group interviews, moderator focus group). This project is sponsored and organized by King’s College London. The Research Governance Office can be contacted via email at rgo@kcl.ac.uk.

The data from the nested qualitative study were collected in alignment with the original protocol aims of understanding participant perspectives of the intervention, control condition, and trial procedures generally (published elsewhere [29]) and for understanding perspectives and experiences of moderation specifically (this study).

Informed consent was obtained from all community members at two stages: (1) at T₁ when completing the 2 eligibility screenings, and if found to be eligible for the trial, (2) at T₂ when they consented to participate in the trial. All community members had provided consent to take part in the feasibility randomized controlled trial at enrolment. At enrolment, they were informed that they may be invited to interview after the 3-month intervention period, and that additional consent would be sought at that time if selected. Consent was recapped for those who were invited to interview prior to their agreement to interview.

Community members and moderators were invited to be interviewed or take part in the focus group, respectively, after the intervention period. Five days after the invite, they were contacted again to discuss any questions about what taking part would involve before completing a web-based consent form (via Qualtrics) and scheduling the interview. Participants were allowed more time to consider their participation if needed and asked to contact the research team should they decide they wished to participate.

At the beginning of the interviews and focus group, the information about what the session would involve was repeated verbally, with participants reminded of their right to withdraw and to decline to answer any question, at any time. Community members were reimbursed £20 (US $26.93) for their participation, and moderators were reimbursed in alignment with their hourly wage.

CommonGround was codesigned in collaboration with people with lived experience of long-term conditions or physical-mental health comorbidities, clinicians, academics, and software developers [28]. The anonymous web-based platform was open to all diagnoses and could be accessed via desktop or mobile devices. The key features included a discussion forum with different posting options, reactions, and the ability to comment (peer support component), alongside a resources page of evidence-based self-help information (psychoeducation component). The self-help information consisted of 91 resources that were hand-selected by the research team from reliable, reputable sources and covered various topics ranging from keeping active, managing mental well-being, and managing finances. The resources were selected for their relevance to people living with any long-term physical health condition (ie, were not condition-specific). The CommonGround platform features, functions, and layout remained static throughout the intervention period with one minor exception. During the first week, the software developers made a minor edit to how the latest activity in the community forum appeared. Initially, at launch, the newest post was displayed at the top of the community forum feed. This was changed so that the post with the most recent activity displayed at the top of the feed (ie, existing post with the most recent comment or new post). Further details of the features of CommonGround and screenshots of the platform are available in Multimedia Appendix 2.

The CommonGround platform was governed by the Community Principles, Moderation Policy, Safeguarding Policy, and Privacy Policy, which were all published on the platform (and are available on the CommonGround projects Open Science Framework webpage [44]). These policies were coproduced in collaboration with our research advisory group (10 members living with a range of long-term health conditions, including 6 females and 4 males (self-reported sex), aged between 25 and 73 years), software developers, and the Samaritans Online Harms Team [45]. Key principles of the moderation approach included that the platform was anonymous, was not moderated 24/7, and did not provide one-to-one or crisis support. The moderation team also had access to a moderator handbook, developed in consultation with a leading disability charity [46], which included guidance on how to moderate different scenarios. The CommonGround platform had an admin backend that allowed moderators to edit and delete posts, delete comments, review flagged posts, and escalate posts for further action. Community members were able to view all policies and “flag” posts for review by moderators. Prior to the launch of the community, moderators were trained by the research team on how to use CommonGround and how to implement the governing policies. The moderator training was developed based on consultations with the community manager at SCOPE [47]. Each moderation session was expected to take approximately 60 minutes per day, and to occur at some time between the hours of 8 AM and 8 PM. Moderators were able to choose whether to moderate across a single or multiple sessions, and exactly when to moderate, dependent on the needs of the community (eg, when there was a moderation concern to monitor) and their personal schedules. Moderators were required to complete a handover after each moderation session of outstanding tasks, posts, or comments to watch, and any other comments. Moderators were instructed to review the previous moderation handover at the beginning of their session and to complete any outstanding tasks as a priority. Then moderators were instructed to review new content and complete any moderation tasks as needed (ie, in response to what had been posted). These moderation tasks may have involved any (or all) of the following: edit posts, delete posts, delete comments, review posts with attachments, review flagged posts, comment on posts to guide discussions toward a productive versus harmful narrative, and contact users to issue warnings. A Microsoft Teams chat was also created to support shared decision-making and provide updates to the moderation team. All moderators were of equal status and consulted with the research team (HGJ, GL, and MH) as required for additional support or to resolve any queries they had on the best course of action. For instance, if a moderator was unsure about whether to moderate or not (eg, whether to delete a comment, an action that could not be reversed), they were encouraged to discuss this with the team to come to a collective decision. If required, any decisions to issue warnings to community members (part of the “three strike” policy) or to remove community members from CommonGround would be made by joint decision of the moderators and research team. At any time, moderators could request an informal debrief with the research team, which included a liaison psychiatrist. A feedback session for moderators to share experiences was conducted halfway through the 3-month intervention period.

There was a separate engagement team whose exclusive focus was on cultivating engagement on CommonGround by sharing discussion topics to stimulate conversation and weekly circular emails with community updates and reminders to log in. The engagement team had no moderating or safeguarding responsibilities, but could use the “flag this post” button if they came across something they felt required attention (as a regular community member could also do). The engagement team consisted of our co-researcher with lived experience and 2 other research team members. The profiles of the Engagement Team on the platform were clearly identifiable as being distinct from both regular community members and the Moderation Team via their avatar and account username. As the engagement team involved different team members to the moderation team, had unique roles and responsibilities, and were not involved in any moderation activity, the engagement team will not be discussed further in this manuscript.

Data from the community members (interviews) and moderators (focus group) were analyzed together using NVivo software (version 14, released 2023; Lumivero). An inductive thematic analysis was performed to sort the data into overarching themes that were developed inductively and from an interpretivist-constructivist position [49,50]. The first author (HGJ) familiarized herself with the entire dataset by reading and rereading the transcripts. Once immersed in the data, an initial list of codes was generated, which were constantly compared with one another to identify patterns of shared meaning that formed provisional themes. An experienced qualitative researcher (VL) reviewed the initial coding framework to help refine the codes and provide an additional perspective on interpreting the data. VL and HGJ also met regularly to discuss the analysis and explore different interpretations. The transcripts were then reviewed, and relevant extracts of the data were coded into NVivo nodes (the refined codes). Through an iterative process, codes were refined, added, and condensed as needed before being organized into themes. Following further discussions, a consensus was reached on the analysis, and the themes were named and defined. Throughout, our lived experience co-researcher (EA-F) ensured the coding framework was reflective of lived experience terminologies and definitions.

As moderators reflected on their experiences of moderating the CommonGround platform, they highlighted areas where moderation could be improved in the future. Areas for potential improvements and evidencing quotes are presented in Table 3 below.

Online peer support promises a low-cost, easily accessible, and scalable way to support mental well-being and self-management among people living with long-term physical health conditions [1,24]. The risks and potential adverse events that can occur when engaging with peers via web-based communities are well recognized [9,27,31-34]; yet, research into how best to moderate communities to keep them safe is limited, particularly in the context of new web-based health communities [37]. In this work, we explored the perspectives and experiences of both community members and moderators together to evaluate the acceptability and feasibility of our co-designed moderation procedures and identify areas for future improvements. Overall, community members and moderators both had positive views of our moderation procedures. Moderators found it feasible to implement the polices but identified opportunities to improve moderation as the community culture evolves. Based on these findings, we provide recommendations to guide the development and implementation of moderation practices for new communities (Table 4).

To our knowledge, this is the first study to integrate the experiences of community members and moderators to understand moderation in a newly launched peer support community. It was universally agreed that moderation is a critical pillar of community safety, reflecting the perspectives of members [31] and moderators [9] of other established communities that moderation is essential for creating a safe space and mitigating risks. As in other web-based health communities, we also found that the community members and moderators had particular concerns about the need to moderate medical misinformation [9,31,37]. This highlights the importance of tailoring moderation practice to the unique aims and sensitivities of the specific community, and the need to identify these through collaboration with the target audience (Recommendation 1, Table 3). While moderators reported a few instances where they were required to enforce the moderation policies, they felt that the moderation blueprint with the clear escalation procedures was appropriate and applicable to the community. The minimal need for moderator intervention may reflect the relatively low activity levels on CommonGround, where fewer user-generated posts naturally reduced the need for moderation. Equally, it may be because members understood and were happy to abide by the moderation policies [30]. Community members found “no red flags” in the moderation policies, suggesting that these were aligned with their expectations and moderation needs.

The apparent success of moderation also suggests that co-designing moderation practices is critical in ensuring their acceptability among the community, and that having clear boundaries and explicit guidance for moderators can lay the foundations for successful moderation from “Day One” of the community (Recommendations 1-3, Table 4). Community members valued being able to easily access the moderation policies when needed, underlining the need to share moderation practices transparently, which can enhance legitimacy, perceived consistency, and accountability of moderation ([51]; Recommendation 4, Table 4). Improving the communication of reactive moderation decisions may help to increase understanding, promote adherence to norms, and improve future behavior [51,52]. Together, these findings underscore the importance of having clear, comprehensive moderation practices tailored to the community, including the specific content of policies and the transparency, tone, and language of communications. Where the capacity to codesign moderation policies and practices is limited, examples from existing communities can be drawn upon, particularly as key moderation practices and experiences are likely common to any online peer community, such as the use of peer moderators [30].

By integrating the perspectives of community members and moderators, we gained unique insight into areas of moderation that may require review and adaptation as the newly formed CommonGround community evolves. The anonymous nature of web-based communities is often seen as a double-edged sword: anonymity can facilitate stigma-free and honest posting for some [53], whilst others feel that anonymity hinders the ability to form the meaningful and enduring social connections that are the cornerstone of peer support [54,55]. We found similar mixed opinions among our community members. The desire for reduced anonymity may grow as community members become more willing to share personal information if it makes accessing social support from peers easier, even if this could introduce other risks [56,57]. The reputable KCL ownership and branding positively influenced perceptions of safety. However, as some members appreciated the anonymity or acknowledged its value for others, offering members the ability to choose their preferred level of anonymity may be appropriate. Future work should explore with lived experience experts how to implement features that allow personal control over anonymity to ensure that anonymity can serve as a facilitator to engagement for those who desire it, while also removing this potential barrier to social connection for others [58,59]. These divergent views on a core aspect of CommonGround’s moderation practice highlight the importance of reflecting on moderation practices by listening to the community and adapting as necessary (Recommendation 6, Table 4).

The moderation practice and style needed for CommonGround as a new community were recognized to be different from those of larger, well-established communities. In well-established communities, community members can be seen to be actively self-moderating by detecting medical misinformation, questioning information shared, reminding peers not to give medical advice, and reporting trolls to moderators [31,37,60]. While CommonGround moderators and community members agreed that peer-to-peer moderation could be more powerful and feel less punitive, they recognized that CommonGround could not self-moderate yet. The limited volume of user-generated content and the weak sense of community and ownership among members is likely to have hindered the emergence of self-moderation. The shift toward self-moderation must occur organically, as the premature withdrawal of moderators could harm community cohesion and reduce perceptions of safety if members are not comfortable acquiring moderation responsibilities [31]. Indeed, as prior work often cites “super users” (ie, those who engage with the community often, and are often observed to be generating larger volumes of content than the average user) in contributing to how this culture evolves, who often only emerge as a community evolves and may also facilitate the shift toward self-moderation [61]. As voiced in our research and prior work [31], there may always be a need for professional moderators to deal with high-risk or serious issues, even if the community takes ownership over other moderation activities. Although used infrequently on CommonGround, the moderators valued features like “flag this post” as critical tools for encouraging a community towards self-moderation. Indeed, moderators of other web-based health communities have urged members to use such self-moderation tools more to help shape moderation norms by defining what content they deem unacceptable [30]. Therefore, while the community culture needed for self-moderation cannot be artificially created, the features and functions that support self-moderation can be implemented early (Recommendation 5, Table 4). Future work should evaluate member and moderator perspectives as a community evolves through different stages of self-moderation to determine what additional practices may help support a successful transition.

Few studies have explored the experiences of moderators, but some key challenges consistently emerge [37]. First, a common challenge lies in efficiently identifying new content to reduce the risk of unmoderated posts escalating into worst-case scenarios [9,37]. Moderators were also mindful of how this challenge could intensify as the community grows, underscoring the need to develop and test novel tools that enhance moderation efficiency as CommonGround is scaled. Another challenge that our moderators faced was navigating the moderation threshold where they must evaluate whether to take action [9,30,51]. The views of the moderators echo those of moderators in other communities, who want to prioritize safety but are concerned that premature or overbearing moderation can stifle conversation and potentially damage community culture [62]. Moderators observed that once they commented, the community ceased to comment further. This pattern has been observed in other communities [62], with moderators interpreting this to reflect disengagement from the community. However, members typically felt that the moderators had struck the right balance, suggesting that the quiet comment section may instead reflect the generally low engagement we had on CommonGround, where most members were “readers” preferring to observe posts rather than actively create content. This disconnect between how moderators believe their actions are perceived and how the community actually experiences them highlights the importance of considering both perspectives, particularly if informing changes to moderation practices (Recommendation 6, Table 4).

The benefits that the CommonGround moderators experienced as a result of operating as a team have been documented among other moderation teams. For instance, the CommonGround moderators shared how it became easier to navigate the moderation threshold over time and as a result of seeking support from their team, as they were able to exchange experiences to align their moderation practice, seek emotional support, and access the expertise of other members for second opinions on complex situations [63]. This adds to the literature highlighting how the moderation team becomes a form of peer support in itself that can help improve moderation efficiency, consistency, and reduce the moderation burden [9,64]. This highlights how it is important to consider how best to support moderators whilst they are supporting the community and give them a dedicated space to connect and support one another (Recommendation 7, Table 4).

There are some limitations that should be considered for the interpretation and transferability of our findings and recommendations. First, this study explored the perspectives of community members and moderators who participated in the new community that was live for only 3 months. Many community members shared that they did not engage with CommonGround as they would have liked to due to personal circumstances or having limited time available during the 3-month trial. While lower levels of engagement and forum activity could be expected during early stages of community development, the experiences shared in this work are contextualized within a small community with relatively low engagement and sense of community in a research setting. As such, our findings and recommendations may not apply to communities launched with larger populations or with high volumes of user-generated content from the outset. Second, our findings should be interpreted relative to the nature of our moderators: professional, paid moderators with clinical expertise who were trained specifically about the moderation practices for CommonGround. Prior research suggests that professional moderators are often perceived as more trustworthy and credible than informal moderators, particularly around their moderation of medical topics [31]. Indeed, our moderators were aware of how their clinical expertise influenced their approach to moderation, and therefore, our findings may not generalize to moderation teams of volunteers, peers, or those without clinical or moderation backgrounds. Third, there is a potential for selection bias in the community members’ interviews, as those who consented to participate may have more positive views of web-based peer forums or the moderation and governance of CommonGround. However, we did purposefully sample community members based on their engagement levels, probing those who had never logged in or only logged in a few times for their underlying reasons, motivations for low (or no) engagement, to unpack whether moderation or governance played a role in their disengagement. Fourth, although community members were anonymous to one another, they were aware that the moderators could link their pseudonyms to their personal details if required for moderation purposes. This awareness that they could be identified in exceptional circumstances may have influenced what members posted, meaning that the CommonGround environment may not directly reflect non-research anonymous communities.

All moderators and community members needed internet access to participate in the web-based community and interviews or focus groups. While 96% of households in the United Kingdom have internet access [65], for those without or who have limited access, their ability to engage with the community and interviews may have been limited. Additionally, some participants may have experienced barriers in accessing CommonGround privately, which may have had a downstream impact on the content of the community forum, and therefore the moderation needed [66]. Future work should consider how digital exclusion and privacy influence the nature of web-based communities, in particular their moderation.

Our work adds to our understanding of the critical role moderation has in shaping safety in online peer support communities designed to support mental well-being, and the importance of creating appropriate and trustworthy governance that is specifically tailored to the needs of the community members. By exploring perspectives on moderation in a new community and by integrating the experiences of both community members and moderators, we have gained unique insights into how to enhance our codesigned moderation practices going forward. We have provided a transparent account of the creation, implementation, and reflection on our moderation policies and governance, and recommendations on how to approach the moderation of new communities in the future.