A systematized review of empirical research articles on moderation was conducted to characterize the most widely used moderating practices and the most common components of community guidelines in OMHCs. Unlike a full systematic review, a systematized review approach includes some, but not all, of the elements associated with systematic reviews [16]. However, we enhanced the depth of the systematized review findings by purposively sampling gray literature, including community guidelines for OMHCs. This combined approach was selected because moderators adhere to community guidelines that are not always published or reported in academic literature and continually evolve in practice, providing valuable insights into how OMHCs are managed.

Studies were included about moderation practices in OMHCs with the following features: (1) accessed via the internet, (2) the user interface has a text-based component, (3) moderation is asynchronous such as a forum or synchronous such as a chat room, (4) the OMHC focus is on providing support around mental health, including diagnostic-specific communities (eg, for people diagnosed with an anxiety disorder), experiential-based communities (eg, for people who hear voices), and general mental health communities (eg, aimed at people without a specific diagnosis but seeking support), (5) the OMHC has at least one human moderator. Gray literature documents were included if they could be accessed without a login or account creation and contained a list, code of conduct, behavior guidelines, community rules, or details regarding the moderator role. Literature was excluded if the OMHC was primarily for crisis support, audience-based (eg, centered around a content creator), nonmoderated or bot-moderated, or provided one-to-one support from a health care professional and thus lacked a peer support component. Literature was also excluded if it was not written in English or if the document was a review or thesis. The search covered the period between January 2014 and December 2024, reflecting the rapidly evolving landscape of OMHCs since 2014 including the increasing use of mental health chatbots [17].

Three data sources were used:

The range of databases (n=3) and the decision to use the top 10 (Google) and 5 (social media platforms) hits reflect the systematized rather than systematic review methodology, which is intended to identify the most widely used rather than all OMHC moderator approaches.

All OMHCs identified in the search were screened individually for inclusion, and, where available, community guidelines were extracted.

Searches were conducted in April 2024, and the identified literature was collated using Covidence (Veritas Health Innovation Ltd.) software. The search involved initial screening by two reviewers (BRI and MM) of titles and abstracts, followed by full-text screening, with a 10% concordance check, sufficient to achieve high concordance. Any discrepancies were discussed and resolved collaboratively with the researchers. The full-text screening was subsequently reviewed by AC, with further checks and concordance discussions carried out with FN to ensure consistency. The review did not include ongoing database searches over time.

A bespoke data extraction sheet was created and then piloted on 10 documents (5 empirical and 5 gray literature) by FN and AC. Modifications to the data extraction sheet included the addition, refinement, and expansion of characteristic categories. For example, the inclusion of “formality of OMHC” and the differentiation between “intervention,” “formal” and “informal” communities. The finalized data extraction form, shown in Multimedia Appendix 1, was used to record for each document:

Data synthesis was conducted using document analysis on all included documents, ie, academic papers, gray literature, and online web pages, involving skimming, reading, and interpretation of the data. This process followed guidance from Bowen [19] who recommends an iterative process involving content and thematic analysis. A 5-step process was followed. First, 2 analysts (AC and FN) independently categorized data relevant to each objective. For all included documents, verbatim text segments relating to moderator practices and community guidelines were extracted. Second, initial codes were developed through line-by-line analysis of extracted text segments by AC and 10% of codes were checked by FN. Codes for each objective (moderator practice and community guidelines) were differentiated and analyzed separately. Initial codes for a moderator’s practice were descriptive in nature, describing specific actions that moderators undertook. For community member guidelines, initial codes descriptively summarized each guideline and rule within included documents. Third, initial descriptive codes of a similar nature were categorized into subthemes by AC and checked by FN to allow for broader patterns of meaning to emerge. For example, specific mentions of guideline violations or issuing warnings or reminders before taking further action were categorized under “giving behaviour warnings.” Fourth, subthemes of a similar nature were grouped together by AC and checked by FN. The subthemes were reviewed by checking them against the whole data set, informing the merging and integrating of categories. Once confirmed, the label names were reviewed to ensure comprehensibility, eg, “account use” was changed to “account responsibility and security.” This coding framework was then used to create each typology, with all analysts contributing to final choices about the definition and organization of superordinate and subordinate categories and the choices of exemplar text. Finally, a vote counting of the number of documents identifying each category was conducted to determine the strength of the theme of each category, conducted by AC, to indicate how commonly each category was used. This strength of theme analysis did not influence the content of the typology. The 2 primary analysts (AC and FN) came from different academic (psychology and sociology) and mental health professional backgrounds (nursing) and were responsible for all data extraction and analysis. All disagreements were resolved through discussion, comparison with other coded data, and consensus. The other 7 analysts came from different disciplinary backgrounds (nursing, health service research, health psychology, mental health research, technology design research, human computer interaction, public involvement, equality, diversity, and inclusion, and lived experience) and supported the interpretation of themes and resolution of any coding disagreements. The s7 analysts were not involved in the coding of data.

The systematized search identified 11,347 studies and other documents, from which 80 [11,20-96] met the inclusion criteria. The flow diagram is shown in Figure 1.

In total, 80 documents [11,20-96] were included, as detailed in Multimedia Appendix 2. Documents comprised academic papers (n=47) [11,20-63], webpages (n=29) [64-92], and others (n=4) [93-96], with only slightly more published 2021‐2025 than 2015‐2020. Among these, gray literature contributed 29 sets of community guidelines, compared to 9 sets from academic sources. The 80 sources identified discussed 52 distinct OMHCs, with 4 sources mentioning more than one OMHC. The 52 OMHCs included, their locations or the locations of their lead authors were distributed as follows: United Kingdom (n=12) [35,49,51,57,66,68,71,74,79,80,88], the United States (n=11) [38-40,42-44,65,69,76,90,97], Australia (n=10) [20,24,27,32,33,37,45,48,70,91], Canada (n=2) [23,41], the Netherlands (n=2) [26,60], Taiwan (n=1) [63], and the United Kingdom and North America (n=1) [75], with the remaining 13 OMHCs [61,62,73,77,78,81-87,89] having no geographic access restrictions. The OMHCs were hosted on various platforms, including private non–social media websites (n=30) [20,23,24,26,27,32,33,35,37,42-45,48,49,51,57,60,65,66,68-71,73-76,90,91], social media websites (n=20) [38-40,61,62,77-89,92,97], platforms that incorporate both private and social media components (n=1) [63], and one OMHC [41] that did not report its platform type.

Based on their descriptions, OMHCs were allocated to levels of formality: (1) formal (n=16) [48,49,51,57,65,66,68-71,73-76,90,91], which are professional or training spaces often linked to health care providers or organizations; (2) intervention (n=18) [20,23,24,26,27,32,33,35,37-45,97], targeted platforms delivering specific mental health interventions, usually guided by professionals or trained facilitators; and (3) informal OMHCs (n=18) [60-63,77-89,92], community-based or peer-led spaces where individuals share experiences and support each other, often without professional oversight. As shown in Multimedia Appendix 2, most guidelines were published for formal and informal OMHCs, rather than intervention OMHCs.

Four primary methods for accessing OMHCs were identified: (1) invitation-only groups (n=14) [20,23,24,26,27,32,33,35,37,39,40,45,60,75], such as intervention-related OMHCs for research purposes (n=12) [20,23,24,26,27,32,33,35,37,39,40,45], (2) publicly accessible content with sign-up required to post (n=18) [38,65,68,69,71,76,81-91,97], (3) closed groups (n=18) [42-44,48,49,51,61-63,66,70,73,74,77-80,92], for example, Facebook (Meta) closed groups which require application to join, (4) open community (n=1) [57] allowing free access to content and posts, and one OMHC access type was unspecified. The included OMHCs stated various community purposes related to mental health support, with some more general, focusing on overall mental health support (n=9) [49,51,62,66,70,85,87,88,92]. In contrast, others were more specific, targeting individuals with specific mental health problems (n=33) [20,23,24,26,27,32,33,35,37-45,57,60,61,63,65,76-80,82-84,86,89,97], and many combined both general and specialized support within their OMHC (n=10) [48,68,69,71,73-75,81,90,91].

Different types of moderator characteristics were identified. Three approaches to moderation were identified: premoderation, occurring before the message is visible to other users; postmoderation, occurring after the message is visible to other users; and reactive moderation, occurring when complaints or concerns are raised about a specific post. Most OMHCs used multiple moderation approaches; the most common were postmoderation and reactive moderation (n=22) [20,27,32,35,37,39,41-43,45,48,51,57,60,65,70,71,73,75,85,88,90], followed by all 3 approaches (n=12) [24,26,33,38,44,62,63,74,83,86,89,97], premoderation and reactive moderation (n=2) [82,91], and premoderation and postmoderation (n=1) [68].

Moderators within the OMHCs had differing backgrounds, including health professionals (those with health professional registrations), trained moderators (those who have received some formal moderator training), peer moderators (who have lived experience of mental health problems or a specific diagnosis that is relevant to the OMHC), and volunteer moderators (who volunteer their time to moderate). Most OMHCs had peer moderators (n=13) [39,60,68,77-80,83,85-88,92], and many had a mixture of backgrounds, such as health professionals and peer moderators (n=5) [27,35,43,70,91] or health professionals and volunteer moderators (n=4) [69,71,81,90]. Remuneration practices varied: most were not reported (n=21) [20,23,24,26,27,32,33,35,37,39-43,48,60,61,65,69,76,81], the majority were volunteer roles (n=19) [38,51,62,63,68,77-80,82-89,92,97], some were paid roles (n=7) [44,45,49,57,66,74,75], and others had a mix of paid and volunteer roles (n=5) [70,71,73,90,91].

Moderator practices were synthesized into four categories, summarized in Table 1 and fully described in Supplementary Table S1 in Multimedia Appendix 2.

Some documents used the term “rule” to convey explicit injunctions or prohibitions, and others used “guidance” or “guidelines” to imply flexible recommendations. However, the content’s language was inconsistent, with some “rules” including recommendations, and some “guidelines” including requirements. There was a considerable variation in the depth, specificity, and scope of guidelines across different OMHCs. For example, although hate speech was prohibited across all included OMHCs, the definitions and descriptions varied substantially. Some communities provided highly detailed lists of protected attributes, including those related to disability, gender, sexual orientation, immigration status, religion, race, and political orientation. Others adopted broader, less prescriptive definitions, acknowledging that the dynamic nature of language makes it challenging to codify every form of harmful speech. The community guidelines were synthesized into 3 categories, summarized in Table 2 and fully described in Supplementary Table S2 in Multimedia Appendix 2.

This systematized review developed 2 typologies. The typology of identified moderator practices spanned pre- and postmoderation processes, reactive moderation of content, community-building activities, and member-specific monitoring. The community guidelines typology identified 3 categories, comprising the content of individual posts, membership behavior expectations, and moderation approaches.

Both typologies were intentionally designed to be adaptable and flexible in scope. This adaptability is essential for 2 key reasons. First, the typologies were shaped by the content available within the included OMHCs. They therefore reflect the practices and guidelines that were most accessible, rather than offering an exhaustive catalog of all possible moderation approaches. Second, it is essential to acknowledge that the typologies developed through this review are, by necessity, abstractions of complex, real-world phenomena. Rather than prescribing specific actions, the typologies provide examples and categories that can inform thinking, guide discussions, and support the creation of community-specific guidance. Therefore, translating these typologies into practice will require careful adaptation and nuance to the specific goals, purposes, structures, and cultural contexts of individual OMHCs.

The frequency with which certain practices were cited, particularly those related to flagged content, highlights the centrality of safety within OMHCs. This aligns with prior research showing that maintaining safety is a core responsibility of moderators and a cornerstone of community trust and sustainability. For example, a survey conducted across 10 countries involving 13,500 individuals found that 79% of respondents preferred moderating online content for safety rather than prioritizing freedom of speech [98]. Although there were between-country differences, which make a universal approach to moderation difficult, survey findings highlighted the need for moderation practices that are sensitive to the specific values, purposes, and social or cultural norms of each OMHC.

While safety is paramount, the typologies also demonstrate that moderation is a multifaceted process. Moderation encompasses a broad spectrum of responsibilities, extending beyond screening content, signposting to mental health support, fostering community cohesion, guiding user interactions, and shaping a culture. This reflects previous work showing that moderators are not merely gatekeepers but active “shaping forces,” as cultural agents who influence norms, expectations, and culture within their communities [11]. Our findings also highlight how community guidelines operate as both preventive and educational tools. They provide a visible framework that helps members understand what is expected of them and how to participate safely and respectfully. The types of potentially triggering content varied across communities, including where trigger or content warnings were positioned within posts. These findings provide a foundation that other communities can build upon and extend, while recognizing that additional trigger or content warnings exist beyond those identified in this review [99]. Whilst this accounts for expectations surrounding community behavior, consideration of moderator behavior extended beyond the scope of this review. Understanding how moderator behavior (eg, injection of topics, overstepping the moderator role, or taking on a dual role of community member and moderator within an online community) may influence the culture, safety, and sense of community within a group may be an important point of future research.

Differences also emerged regarding language deemed explicit or offensive. Most OMHCs banned such language outright, while others permitted it under specific conditions, such as when it expressed frustration without targeting others. A minority required posts containing explicit language to include trigger warnings. None of the communities published exhaustive lists of prohibited terms, reflecting the inherent fluidity of language and the need for moderator discretion in applying these rules. The availability of community guidelines varied significantly across different platform types. Publicly accessible informal communities, such as those on Reddit (Steve Huffman, Alexis Ohanian, and Aaron Swartz) or Facebook (Meta, Inc), were more likely to publish their guidelines in full. Most intervention-based OMHCs either provided minimal information or did not share their guidelines. While specific guidelines may exist, they may not have been publicly available, possibly to maintain confidentiality or because the details have been published only within academic literature. As a result, the available evidence is skewed towards informal and formal communities, leaving gaps in our understanding of how interventions or closed OMHCs operate. For example, private platforms such as Discord (Discord Inc.) could not be included due to restricted access. This highlights a significant limitation for both research and practice, as some of the most vulnerable populations may be engaging in spaces that are effectively hidden.

Our findings align closely with Skousen et al (2020) [100], who identified 5 categories of practices that members used to moderate successfully in online patient communities: instructive, semantic, connective, administrative, and policing. These categories correspond to our typologies, particularly in areas such as providing support (instructive), facilitating and guiding conversations (semantic and connective), and intervening when rules are breached (policing). Notably, administrative practices focused on community growth were not evident within our analysis, likely because public-facing documents such as codes of conduct do not capture internal strategies or behind-the-scenes activities undertaken by moderators. Although not explicitly stated, the typologies imply that many asynchronous forum moderators take on an instructive role by explaining how to navigate the platform, clarifying its technical features, and signposting users to relevant subforums or resources. For instance, the community guidelines identified issues relating to members holding multiple accounts, though they do not specify how moderators address this. This particular issue around multiple accounts also reflects a broader, increasingly recognized problem in online research [101].

This study has several limitations. First, our search was limited to English-speaking communities and English-language sources, most of which were based in Western countries. This limits the cultural and linguistic reach of the typologies, which may not fully capture the practices of communities operating in other contexts. Adaptation of the findings to other cultures will require adopting, adapting, and releasing inherited norms [102]. Second, because our analysis focused on publicly accessible guidelines, it may not fully reflect the private or intervention-based OMHCs or the behind-the-scenes practices of moderators. For example, intervention-based communities often provided minimal information about moderators’ roles or responsibilities, and closed informal communities were inaccessible to researchers. Third, it did not include substance use–focused communities, which may operate under different moderation policies, user needs, and risk dynamics, thereby limiting the generalizability of the findings across the broader spectrum of online mental health support spaces. Fourth, our search period was 2014 to 2024. An updated MEDLINE search for OMHC publications conducted up to July 16, 2026, identified only 13 further papers, none of which related to moderator practices. Fifth, the absence of formal quality appraisal, combined with purposive gray-literature sampling of community guidelines (using Google and social media platforms only), limits the ability to weight sources and introduces a possible selection effect. Sixth, we used a systematized rather than systematic review methodology. Specifically, protocol registration, exhaustive search strategy, and database searching (eg, using an unprofiled Google search), formal quality appraisal, and analyst reflection were not undertaken. As the goal of this review was to establish the primary features of OMHC approaches to safety and establishing a sense of community, issues of selection and interpretation bias are less relevant, so the extra resources needed for a systematic review were not justified. Finally, while community guidelines provide valuable insight into the rules and expectations governing OMHCs, they may not always map neatly onto moderators’ day-to-day practices. Moderators may engage in nuanced decision-making and relational work that is not explicitly documented in written policies.

The typologies developed in this study provide a foundation for future work to support moderators and strengthen OMHCs. While this review focused on OMHCs, the findings may have broader intersectoral implications.

For developers and moderators, transparent decision-making is essential to ensuring fairness, consistency, and community trust. A key challenge is balancing user safety with the need for open, authentic expression. These sometimes competing goals create ethical and legal dilemmas that require careful judgment. Overly strict or opaque moderation can undermine authenticity, whereas insufficient moderation may expose members to harm. The typologies offer a practical tool and starting point, providing a framework for translating and articulating specific guidelines for moderation decisions, clarifying the moderator role, and supporting practices across different online community platforms. It is noteworthy that only 6 (11.5%) OMHCs had an appeals procedures, suggesting more attention needs to be given to governance arrangements enhancing OMHC trustworthiness. These considerations are especially relevant in the context of evolving social media regulations. Platforms are continually adjusting their policies on freedom of speech, while some countries, such as Australia and the United Kingdom, are introducing restrictions on social media access for children under 16 years. The typologies can help developers anticipate challenges, maintain member safety, and preserve meaningful community engagement in these shifting regulatory contexts. Similarly, the 4 moderator practices identified in Table 1 can provide the basis for a bespoke training package for community members wishing to take on a moderator role, with community-specific guidance for each component, eg, new member screening processes.

In clinical practice, the findings can assist clinicians to recognize safer versus less safe OMHCs. Safer OMHCs are likely to have guidelines for members covering post content and member conduct including many elements from Table 2, and transparent role expectations for moderators spanning premoderation to postmoderation, with member-monitoring and community-building features aligned with Table 1 approaches. This knowledge can inform recommendations for service users seeking online peer support to help them find safe, well-moderated communities. Future research could quantitatively examine whether specific moderation processes are associated with higher engagement, greater satisfaction, or better symptom or psychosocial outcomes. For example, comparing the effects of community-building processes with those centered on member and content monitoring to determine how these different approaches influence user engagement, satisfaction, or outcomes in distinct ways. Research could also examine whether specific moderation processes mediate changes in an individual’s symptoms and psychosocial outcomes. Quantitative studies should be complemented by qualitative approaches such as interviews or ethnographic observation to capture the day-to-day realities, challenges, and decision-making processes involved in moderation.

Finally, our focus on human moderators can inform likely evolution towards automated moderation. Our findings are entirely consistent with a recent independent study which concluded “Moderation is critical in making online peer communities feel safe and engaging” [15] (p.1). Automated moderation of OMHCs is not yet common, but experience from 2 related areas raises concerns. Automated moderation of social media platforms can both reduce some forms of harmful content and create new risks, such as curbing dissemination of harm reduction and health promotion materials through erroneously interpreting it as promotion of drug use and sales [103], or misunderstanding evolving language and failing to recognize “algospeak” used to evade algorithmic moderation [104]. Chatbot therapists have many potential benefits, such as immediate availability and reduced stigma, but also carry risks including addiction, social isolation, misuse of data, hallucinations, bias, and lack of safety [105]. It is known that community moderators develop “knowledge-in-action” which then allows moderator teams to develop “groupwise reflective practice” [106], which becomes self-moderation in a mature community [15]. The contribution, if any, of automated OMHC moderation at each stage will require careful and disinterested evaluation, with equal attention to risks and benefits.

The typologies presented here suggest that moderation in OMHCs involves a wide range of interrelated responsibilities that extend well beyond the monitoring of content or enforcement of rules. Moderators are required to navigate multiple functions simultaneously, including assessing potential risks to member safety, managing boundary issues, responding to problematic content, facilitating supportive interactions, and fostering community engagement. In some cases, moderators also undertake actions related to safeguarding, such as monitoring individual risk or contacting external services. Taken together, these activities point to moderation as a complex form of sociotechnical and relational work that sits at the intersection of community facilitation, risk management, and mental health support. As many OMHCs rely on peer moderators or individuals with lived experience, this complexity raises important questions regarding the training, supervision, and institutional support required to sustain these roles safely and effectively. Future research should therefore examine how moderators are prepared for these responsibilities and what forms of training, guidance, and organizational structures best support them in balancing community safety with the facilitation of supportive peer interactions. Using ethnographic methods, such as in-depth observation and interviews with moderators in their day-to-day activities, could provide valuable insights into the nuances of moderation processes and the real-world challenges moderators face in carrying out these responsibilities.

As individuals increasingly experience online communities as a central part of their social network, there is a growing need to manage OMHCs well. Our identification of moderator practices and OMHC guidelines contributes to this goal.