Climate governance is the umbrella term used here for institutions and processes that address mitigation, adaptation, climate-related development, and associated distributional concerns. Decarbonisation governance is a narrower subset concerned with reducing greenhouse-gas emissions and managing the transition of energy, transport, industry, finance, and land use. The reviewed literature includes both domains, so the article does not use climate governance as a synonym for the entire field. Policy instruments, litigation, implementation arrangements, consultation, and informal partnerships are included when they bear on climate-policy decisions (Kern et al. 2024; Tobin et al. 2024a).

The advantage of a multi-level conception is analytical precision about where NGO influence can occur. NGOs may have little control over final legislative adoption but considerable influence over agenda access, evidence standards, consultation procedures, local pilots, or the legitimacy of competing policy frames. These intermediate outcomes are not trivial. They can shape the feasible set of later policy choices. Brusseel, Raeymaeckers, and Verschuere (2025), reviewing advocacy-effectiveness research in nonprofit studies, similarly argue for multidimensional outcome measurement rather than a binary success-failure coding. For climate governance this is particularly important because many policy changes are incremental, nested within broader regulatory regimes, and implemented over long time horizons.

Polycentricity also creates coordination demand. Kern et al. (2024) show that transnational municipal networks evolve over time rather than functioning as fixed structures, while Widerberg (2024) demonstrates that state orchestration can remain central even within ostensibly polycentric arrangements. The implication is that NGOs cannot be assumed to benefit automatically from additional venues. A larger opportunity structure can also produce duplication, competition for attention, and strategic overload. Brokerage becomes valuable when an NGO can reduce these frictions by connecting actors, translating priorities, or moving information between levels that use different institutional languages.

Network mobilisation refers to the deliberate activation and maintenance of relationships through which an NGO can exchange information, coordinate activity, obtain expertise, or approach decision venues. Network position is not measured as a centrality score in this review. It is used only as a conceptual description of relational location where a source documents it. Observed indicators include tie diversity, brokerage roles, repeated interaction, and cross-level or cross-sector links. Direct lobbying is coded by its function within information translation or coalition brokerage; monitoring and implementation partnerships are recorded as activities and outcomes rather than assumed to be stand-alone mechanisms.

Three attributes are distinguished. Structural network properties describe the pattern of ties, such as diversity, brokerage, and embeddedness. Organisational capabilities include expertise, staff, data production, legal knowledge, and the capacity to translate information for a target venue. Relational qualities include credibility, trust, and repeated access, which may develop through prior interaction. These dimensions create opportunities but do not guarantee policy uptake. A dense network may circulate information without changing policy, and an expert organisation may lack access to the venue where a decision is made. Figure 1 therefore separates inputs, mechanisms, policy stages, and outcomes.

Network resources are also stage-dependent. During agenda setting, broad ties can amplify salience and diversify the audiences exposed to a claim. During policy design, smaller trust-based networks may be more consequential because the relevant tasks involve technical detail and iterative negotiation. During implementation, local monitoring, industry links, legal expertise, and administrative relationships become more important. Treating network structure as dynamic rather than fixed therefore avoids a common inferential mistake: assuming that the network that creates attention is necessarily the network that delivers implementation.

Four mechanisms recur across the recent literature and build on established accounts of framing, transnational advocacy, contentious politics, and movement strategy (Benford and Snow 2000; Keck and Sikkink 1998; Tilly 2008; Pulver 2023). Information politics produces evidence, policy analysis, disclosure tools, and interpretive frames. Coalition brokerage connects actors whose capabilities are complementary but whose preferences or organisational routines are not automatically aligned. Legal mobilisation uses procedural rights, litigation, administrative review, and the threat of adjudication to alter obligations and bargaining positions. Symbolic mobilisation uses protest, moral claims, naming strategies, and visible performances to change issue salience and reputational incentives. These mechanisms are analytically distinct, but effective campaigns frequently combine them.

Venue choice is central because mechanisms have different conversion rates across arenas. Kagan (2024) shows that nonprofit advocacy can be better understood when strategies are linked to the venues in which decisions are made. Expertise has high value in bureaucratic and standards-setting venues. Participation law becomes powerful when NGOs can mobilise it outside the courtroom as well as within it (Abbot and Lee 2024). Moral and justice frames can activate broader publics, but their effect depends on audience fit and political context (Gulliver et al. 2023; Rawlyk and Willness 2025). Symbolic mobilisation can shift media and electoral attention, yet policy uptake still depends on institutions capable of transforming that attention into proposals and implementation routines (Fabel et al. 2025; Thomas-Walters et al. 2025).

This venue-mechanism fit clarifies why the same NGO may appear conciliatory in one arena and confrontational in another without necessarily changing its substantive goals. Strategy should be interpreted as adaptation to the institutional task. An organisation can litigate to create procedural leverage, publish expert reports to define a feasible policy option, and cooperate with regulators to translate the resulting obligation into administrative practice. The analytical object is therefore the portfolio and sequence of actions, not the ideological label attached to a single tactic.

Insider–outsider accounts explain differences in access and public contestation but can make repertoires appear mutually exclusive. Advocacy-coalition theory emphasizes shared beliefs and policy learning, while multiple-streams scholarship explains how policy entrepreneurs couple problems and solutions during windows of opportunity (Sabatier and Jenkins-Smith 1993; Kingdon 2011). Transnational advocacy research and structural-hole theory explain cross-border ties, information exchange, and brokerage (Burt 2004; Keck and Sikkink 1998). Venue-based approaches clarify where nonprofit advocacy occurs (Kagan 2024). Embedded network advocacy builds on these accounts by specifying how a portfolio can change across agenda setting, policy design, and implementation, and how implementation feedback can alter subsequent strategy. Table 1 summarizes the distinction.

Embedded, disruptive, and hybrid pathways are analytical descriptions of advocacy portfolios. Embedded activity relies on repeated access, expertise, consultation, and coalition maintenance. Disruptive activity raises public, procedural, or reputational costs when agendas are closed or inertia is strong. Hybrid activity links these repertoires, for example when public attention is followed by technical proposals or when legal action opens a procedural route for administrative response. Neither pathway is uniformly superior. Embedded participation can narrow agendas or create dependence; disruption can raise attention without producing policy uptake. Evidence on protest and radical-flank effects similarly indicates context-dependent outcomes (Jarke-Neuert et al. 2023; Badullovich et al. 2024; Ostarek et al. 2024).

Table 1 compares the proposed framework with adjacent approaches. The distinction is intended to make the theoretical contribution precise: the contribution is not a new label for insider and outsider tactics, but an explicit, stage-linked account of conversion, feedback, and observable outcomes.

The evidence is organised in three layers. The core review contains 24 peer-reviewed publications dated 2023–2025. This window covers the three most recent complete publication years at the search cut-off of 28 September 2026; partial-year 2026 publications, including Yu et al. (2026) and Li and Farid (2026), are used only as contextual updates. The corpus was assembled through targeted cross-publisher scholarly searches, DOI and publisher-page checks, and backward and forward citation searches. Representative terms combined climate or climate governance with environmental NGO, civil society, network, advocacy, framing, coalition, litigation, protest, policy access, and China. The full search strings and date are recorded in the supplement. Searches were run on 28 September 2026. Representative combinations were ‘climate governance OR climate governance’ with ‘environmental NGO OR civil society’ and ‘advocacy OR network’; ‘climate litigation NGO OR environmental organisation policy legal mobilisation’; and ‘environmental NGO China climate policy advocacy policy networks’. Additional searches paired climate justice, framing, coalition brokerage, protest, and policy access with the governance terms. DOI records, publisher pages, and cited-by and reference lists were checked to confirm publication year, peer-review status, and relevance. The 2023–2025 boundary is a reproducible calendar rule: it captures three complete years without mixing a partially observed 2026 into the sample. It is not a claim that this period marks a theoretical break or represents all recent work.

Eligibility required an English-language peer-reviewed journal article published in 2023–2025 that addressed climate policy, climate-related governance, or an advocacy mechanism relevant to the framework. Empirical NGO studies were eligible, as were adjacent studies of movements, public opinion, legal processes, intergovernmental coalitions, business, or governance when they provided contextual evidence on a mechanism or outcome. These actor types are separately coded and are not counted as NGO evidence. Duplicate versions were consolidated at the article level. The full 24-item included inventory is supplied. Historical retrieval totals and record-level exclusion counts were not retained, so this is reported as a structured review rather than a PRISMA-compliant systematic search; no screening flow counts are inferred. Figure 2 shows the auditable evidence architecture. Sources were excluded from the core when they were editorials, reports, dissertations, non-English publications, or journal articles without a substantive link to climate governance or one of the specified advocacy mechanisms and outcomes. An adjacent study was retained only when it clarified a mechanism or an outcome boundary; its actor class remains visible in the coding. Selection prioritized relevance to the framework and coverage across policy stages, actor types, mechanisms, and institutional settings. It did not seek a statistically representative sample, and findings are not weighted by journal prestige, sample size, or study design. The search log preserves representative query language and known screening stages, while the absence of archived result exports prevents retrospective reconstruction of a record-by-record exclusion flow.

The 16 observations in the second layer are illustrative mechanism observations, not a conventional comparative case sample. They were selected to cover the four mechanisms, agenda, design, and implementation stages, and variation in institutional setting and actor type. The unit is stated separately for every observation because it ranges from an organisation or campaign episode to a legal process, network, or public-opinion study. Eleven observations draw directly on publications in the 24-item corpus; five provide contextual benchmarks or adjacent mechanism evidence. The third layer consists of external contextual benchmarks from UNEP climate-litigation reports, Climate Action Network’s public profile, and Yu et al.’s (2026) compilation of Chinese climate-policy documents. These sources describe institutional scale; they do not independently validate NGO influence or the proposed causal sequence. The selection aim was to examine how mechanisms appear in different contexts, not to select cases because they confirmed the proposed framework. Each observation has an explicit unit and source, and only attributes documented by that source are coded. The observations are not independent in a statistical sense: some derive from the core publications, and others describe the same broad institutional setting from a different evidentiary angle. They therefore illustrate mechanism boundaries and evidence limits rather than provide a second denominator for the publication-level results. Mixed-actor or non-NGO observations remain contextual in all substantive claims.

Each included publication was coded on separate dimensions: publication year; stated scope; locations of empirical observations; actor class; study design; analytical method; data source; causal-identification strength; advocacy mechanisms; and outcomes. Study design was not treated as interchangeable with analytical method or data source. A source could receive multiple mechanism and outcome codes. Information and framing covers knowledge production, disclosure, and interpretive work; coalition brokerage covers coordination across actors or venues; legal mobilisation covers litigation and procedural rights; symbolic mobilisation covers protest and public or reputational signals. Direct lobbying, monitoring, service provision, and implementation partnerships are coded by the function they serve and by the policy stage, not forced into a fifth mechanism. A separate primary code is retained only for sensitivity checks. A mechanism is a documented activity or process linking an actor’s action to an intermediate policy-relevant response; it is not simply an organisational attribute. An outcome is separately coded as a condition the source examines, whether or not the mechanism appears to cause it. This rule avoids treating an NGO’s expertise as an advocacy tactic or treating a court judgment as proof of implementation. Direct lobbying, for example, is coded as information translation when the source focuses on supplying evidence to a decision venue, or as brokerage when it organises coordination among actors. Monitoring is coded as an activity, while verified administrative response is an outcome. Financial leverage and capacity building are likewise resources or supporting activities unless the source documents a distinct mechanism and outcome.

One author performed the initial coding and a second-pass consistency audit against the included publications and source metadata. No independent second coder was available, so intercoder reliability and a kappa statistic are not claimed. Ambiguous cases were resolved using the dominant documented function in the source, while all other documented mechanisms were retained as secondary codes. For example, a litigation article receives a legal code and may also receive information or brokerage codes when the publication documents disclosure or coalition work. Among 23 publications with at least one mechanism code, the initial primary assignments are information/framing 9, brokerage 8, legal mobilisation 3, and symbolic mobilisation 3. Under the recorded alternative assignments, the counts are 9, 10, 1, and 3; the leading primary category changes. The main results therefore rely on multi-label coding, and no single-mechanism ranking is treated as stable. One-at-a-time recoding is reported in the replication files. The primary-code rule is applied only when one mechanism is the publication’s central analytical focus or carries the most sustained process evidence. Where two mechanisms are equally central, the primary code follows the mechanism most directly tied to the reported outcome; the other remains in the multi-label set. A second plausible assignment is recorded where reasonable readers could disagree. This procedure makes interpretive choices inspectable but does not eliminate subjectivity. The initial and alternative counts show that even a one-paper reassignment can change the leading category. For that reason, the paper does not rely on primary-code percentages to rank mechanisms, and conclusions are checked against the multi-label pattern and the outcome matrix.

Outcomes use a stable five-part taxonomy: agenda and salience (A); discourse, representation, or legitimacy (D); policy design or adoption (P); implementation or administrative practice (I); and relational or procedural access (R). An outcome is recorded when it is analysed or synthesized in the source, not merely named as a future research priority. Evidence strength is classified separately as causal design, association, process evidence, or synthesis/context. This allows the review to distinguish an outcome category from the strength of evidence about that outcome. Table 2 reports the multi-coded mechanism-by-outcome matrix; the accompanying workbook also lists the evidence tier and source for every code. The matrix is a measure of co-occurrence in the same publication, not evidence that a given mechanism produced a given outcome. It also permits a study to discuss several outcomes without asserting that each was achieved. The resulting profile is intentionally asymmetric: procedural and relational access is frequently discussed, policy design appears in a smaller group, and implementation codes are uncommon. For implementation, the distinction between a review that identifies the outcome domain and an empirical article that follows administrative practice is retained in the workbook. This additional evidence tier prevents a high cell count based on broad review coverage from being mistaken for multiple direct observations of implementation change.

Because the corpus combines experiments, observational studies, qualitative process accounts, legal analyses, and reviews, no common numerical risk-of-bias score was imposed. Instead, the coding records causal identification, temporal ordering, outcome operationalization, data source, and process-tracing or triangulation evidence as separate descriptors. The review gives each publication equal weight only in descriptive counts and does not pool effect sizes. Figure 2 and the supplementary workbook report the 24 included records, complete study-level codes, 16 illustrative observations, search protocol, codebook, source map, and figure data. External benchmarks are kept analytically separate: litigation totals indicate reported case counts, CAN reports membership scale, and Yu et al. describe government policy documents. None measures NGO access or policy influence. The evidence appraisal is question-specific rather than a single quality score. A causal-design label is reserved for studies with a defined outcome and a comparison, experimental manipulation, or credible quasi-experimental strategy. Association denotes a measured relationship without a design that resolves the main alternative explanations. Process evidence requires temporal ordering and source material that documents relevant steps, but it does not by itself rule out rival causes. Synthesis/context describes a review, conceptual analysis, official benchmark, or source used to define the wider setting. For all empirical studies, the coding also records whether outcomes are explicitly operationalized, whether evidence comes from more than one source, and whether a temporal sequence is observable. These fields make the limits comparable without implying that a survey experiment is a better tool than process tracing for every research question.

The included corpus contains four publications from 2023, ten from 2024, and ten from 2025. These counts describe the composition of the purposively assembled sample; they do not demonstrate growth in the research field. The window ends in 2025 because it is the latest complete three-year period at the cut-off date; 2026 publications are cited as contextual updates but not mixed into the core denominator.

Geographic scope and empirical location are coded separately. Figure 3 counts only explicitly identified regions in which observations were collected or analysed, and permits multiple regions per publication. Global-scope reviews or studies without a specified observation geography are not assigned to a region. The figure therefore describes known locations in this sample and cannot establish the field-wide representation of any region or the Global South.

Note. Multi-region coding is permitted. Global scope and unspecified locations are not treated as observed regions.

Five core publications focus specifically on China: one legal-process study, one study of NGO policy networks, one China-focused climate and social-inclusion synthesis, one NGO-litigation comparison, and one comparison of local advocacy coalitions. Their actor types and designs differ. The China subset is consequently treated as a small, scope-conditioned body of evidence, not as a representative national sample or a test of a general relationship between hierarchy and advocacy effectiveness.

Figure 4 groups study designs into qualitative or process studies, reviews or conceptual analyses, quantitative observational or network studies, legal or socio-legal studies, and experimental or quasi-experimental studies. It reports evidence class and analytical method separately. Exact study-design labels, analytical methods, and data sources are recorded as distinct variables in the workbook. These descriptive dimensions do not imply that one design is inherently stronger across all research questions.

The strongest causal identification in the corpus comes from studies of message or protest effects on public opinion, voting, political attention, and media behavior. For example, Ballew et al. (2025) test climate messages across countries, while Fabel et al. (2025) analyse protest and electoral outcomes in Germany. These designs inform mechanisms related to communication and symbolic action, but they do not estimate the effect of NGO advocacy on policy implementation or emissions.

Qualitative and legal studies are more suited to identifying sequences and institutional mechanisms. Abbot and Lee (2024) trace NGO use of participation law under the Aarhus Convention; Zhang (2024) examines how expertise and policy networks shape access in China; and Li (2025b) compares two NGO-initiated climate cases. These sources support process descriptions, not a common ranking of policy effectiveness. Review articles and governance studies provide context and are tagged accordingly.

Implementation remains the weakest part of the evidence chain. Three publications discuss implementation as an outcome domain, but only one core empirical study provides direct process evidence about administrative practice; none estimates a comparable causal effect on compliance, investment, or emissions. Agenda, discourse, and procedural outcomes are more frequently observable in the available material. Figure 4 and the implementation evidence field make this difference explicit. This asymmetry matters for interpretation. An observed media or agenda response can be measured over a short interval, whereas implementation may depend on budgets, administrative capacity, enforcement, firm decisions, local coordination, and changes in political leadership. Evidence that a proposed policy entered a document does not show that agencies implemented it, that regulated actors complied, or that emissions changed. The review therefore treats implementation as a separate outcome and reports both its frequency and its evidentiary basis. This is a description of what the included publications establish, not proof that NGO advocacy has no implementation effects.

Multi-label coding identifies information and framing in 16 publications, coalition brokerage in 14, legal mobilisation in four, and symbolic mobilisation in seven. Counts overlap because one publication may document several mechanisms. They describe what the included publications analyse, not the frequency or effectiveness of NGO tactics in the world. Figure 5 reports the same multi-label measure for the full corpus and the China-focused subset. The co-occurrence matrix shows how often each mechanism appears alongside agenda, discourse, policy-design, implementation, and relational-access outcomes. Information and framing are associated in the coding with all five outcome domains, while brokerage is commonly discussed alongside policy design and access. Symbolic mobilisation appears more often with agenda and discourse than with implementation. These are descriptive patterns in publication content. Because a cell can reflect a review’s synthesis or a qualitative process account as well as an empirical estimate, the matrix cannot be read as a comparative test of predictive power. It is useful for identifying which mechanism–outcome links deserve better-designed follow-up research.

In the five China-focused publications, information and framing appear in four, coalition brokerage in three, and legal mobilisation in two; symbolic mobilisation is not coded in this small subset. This absence is a feature of the selected literature, not evidence that symbolic action is absent from Chinese environmental politics. The data cannot establish that hierarchical systems systematically favor one portfolio. At most, the subset generates a proposition for comparative research across multiple political systems.

The China publications document different parts of the sequence. Zhang (2024) traces expertise and policy-network access; Li (2025b) examines institutional conditions in two legal cases; Wang (2025) compares coalition processes in four cities; Zhu (2023) provides legal context; and Yang and Walker (2024) frame just transition and social inclusion. Their units and evidence classes differ, so these findings are presented as bounded observations rather than pooled evidence of a hierarchy effect.

Legal mobilisation can combine procedural pressure with information and brokerage, but the coding records these functions separately when a source documents them. Zhu (2023) provides institutional context, while Li (2025b) traces two NGO-initiated cases. Neither source supports a general causal estimate. The broader legal literature likewise points to access, disclosure, and administrative response as distinct outcomes rather than a single win–loss measure.

Table 2 reports the number of publications that connect each mechanism to each outcome category. Cells are descriptive co-coding counts; they do not show effect sizes, direction, or causal strength.

Visibility and policy conversion are distinct outcomes. Studies of protest and communication identify changes in public concern, message response, media attention, or political behavior, but these intermediate outcomes are not equivalent to policy-text change. Rawlyk and Willness (2025) examine adaptive climate communication; Liu and Kuang (2024) analyse framing and emotional responses; and Fabel et al. (2025) estimate protest-related political outcomes. The reviewed evidence is consistent with a translation hypothesis, but does not compare the relative effectiveness of advocacy portfolios. The reviewed studies do not share an outcome metric, observation period, or counterfactual. A change in public concern, a new consultation opportunity, a revised policy text, and verified compliance are qualitatively different endpoints. They are ordered in the framework because they may occur at different stages, not because every advocacy episode moves through them in sequence. Some efforts end at visibility; others enter a venue without producing a textual change; still others may affect administration without generating broad attention. Any claim about conversion must therefore identify the endpoint and the temporal link that is observed.

Justice framing can connect climate policy to distributional concerns. Gulliver et al. (2023) document how Australian environmental NGOs frame climate justice, while Ogunbode et al. (2024) find an association between climate-justice beliefs and climate action or policy support across 11 countries. Fabre et al. (2025) estimate public support for global redistributive and climate policies; their unit is public preferences rather than NGO campaigns. These studies address different units and outcomes. They show why framing may matter, but do not establish a common NGO effect or a direct route from public support to durable policy change.

Communication with decision makers adds another potential link. Ettinger et al. (2025) review climate and health communication to policy makers, and Pavenstädt and Rödder (2024) examine how climate movements negotiate evidence-based and political narratives. These studies inform a theoretical account of translation across audiences. They do not show that every translated message is adopted or implemented.

The implementation gap is both empirical and causal. Administrative routines, market conditions, political turnover, and policy feedback can mediate or confound the relationship between advocacy and implementation. Longitudinal network measures, comparison units, administrative data, and credible quasi-experimental designs would be needed to identify such effects. The present review records implementation evidence separately rather than inferring it from agenda or design outcomes.

Four gaps recur in the coded publications. First, implementation evidence is sparse. Second, network evolution is often invoked but rarely measured across policy stages; Kern et al. (2024) offer an exception by tracing municipal-network change. Third, comparative evidence remains uneven across regions. Latin American partnership research and a Tunisia-focused study widen the evidence base, but this purposive sample cannot support a field-wide claim about Global South representation (Tosun et al. 2024; Ben Youssef 2024). Kalatzi Pantera (2025) examines global integration and environmental mobilisation, adding a relevant macro-level context while leaving the organisation-to-policy conversion sequence for direct study.

Fourth, climate governance overlaps with biodiversity, development, trade, public health, energy security, and social inclusion. Oelschläger and Kim (2024) examine climate-biodiversity framing, Ferré et al. (2025) study coalitions around development and renewable energy, and Hrabanski and Le Coq (2025) analyse agricultural framing at COP27. These studies illustrate contested issue boundaries; their actors are not all NGOs.

A final gap concerns representation in global institutions. Marquardt et al. (2024) document unequal speaking and listening relations in youth representation at climate conferences, while Zhao (2024) examines legitimacy claims associated with non-state deliberation. Ruiz-Campillo and Bueno Rubial (2025) trace agenda development by the G77 and China negotiating bloc, an intergovernmental comparator rather than an NGO case. Presence, access, and influence remain distinct outcomes and are coded separately.

Climate Action Network (CAN) illustrates the scale of an umbrella network. Its public profile reports more than 2,500 member organisations across more than 130 countries (Climate Action Network, n.d.). This is an organisation-reported membership benchmark, not an influence measure. The size of a network alone does not show whether its members coordinate across policy stages or whether their proposals affect policy decisions.

The partnership study by Tosun et al. (2024) analyses 195 Latin American NGOs across 21 countries and finds that partnership choices vary with institutional and organisational conditions. This observation illustrates brokerage under heterogeneity, but does not estimate a universal effect of network participation. The full set of 16 observations and their units, actor classes, mechanisms, outcomes, and source limits are listed in Table 3 and the accompanying matrix.

Polycentric governance does not remove the state or other powerful actors. Widerberg (2024) examines state orchestration, while Tobin et al. (2024a) document business participation in climate governance. These are contextual studies, not NGO evidence. They show why advocacy across venues takes place within unequal distributions of authority, expertise, and implementation capacity.

Fridays for Future is a decentralized social movement, not an NGO, and is included only as contextual evidence on symbolic mobilisation. Fabel et al. (2025) analyse nearly 4,000 German protests and estimate effects on voting, politician attention, and media behavior. Nisbett and Spaiser (2023) discuss the moral power and institutional limits of youth activism. These findings concern movement outcomes; they should not be read as direct estimates of NGO influence.

Symbolic action can alter the interpretive environment, but public response is heterogeneous. Brehm and Gruhl (2024) report increased climate concern following strikes and civil disobedience in Germany; Badullovich et al. (2024) examine how public perceptions of protest relate to climate-action support; Gan et al. (2024) assess whether the age and gender of climate protesters shape public support; Jarke-Neuert et al. (2023) analyse collective-action dynamics. Together, these studies motivate conditional expectations rather than a presumption that visibility produces policy uptake.

Evidence on radical-flank effects points to possible complementarity between disruptive and moderate actors. Ostarek et al. (2024) report increased public support for moderate organisations following radical protest. Thomas-Walters et al. (2025) review activism research and find that public-opinion and media outcomes are more commonly established than direct policy or emissions effects. The evidence therefore supports a distinction between attention and policy conversion, while leaving the conversion sequence open to stronger tests.

UNEP reports provide discrete benchmarks for the scale of climate litigation. The 2017 global review identified 884 cases; the 2020 status review reported at least 1,550 cases as of 1 July 2020; the 2023 review counted 2,180 cases as of 31 December 2022; and the 2025 review reported 3,099 cases as of 30 June 2025 (United Nations Environment Programme and Columbia University 2017; United Nations Environment Programme and Sabin Center for Climate Change Law 2021; United Nations Environment Programme 2023; United Nations Environment Programme 2025). Figure 6 displays these as edition-specific bars, not a harmonized growth series. Coverage, databases, and case-identification practices differ across editions, and the totals do not identify NGO-initiated cases or advocacy effects.

Note. Each bar is a source-specific cumulative count. Jurisdictional coverage and case-identification practices vary by edition; the bars should not be interpreted as a harmonized longitudinal growth rate. Sources: UNEP and Columbia University (2017); UNEP and Sabin Center for Climate Change Law (2021); UNEP (2023, 2025).

Legal mobilisation may matter beyond the judgment. Abbot and Lee (2024) show how NGOs use participation rights under the Aarhus Convention; Barritt (2024), Whittaker (2023), and Pieraccini (2024) examine environmental rights, information, and public-interest procedures. These studies support a sequence of possible outcomes, including admissibility, disclosure, procedural access, administrative response, and policy change. Each step requires separate evidence.

Katz et al. (2025) describe transnational legal-clinic collaboration, while Fisher (2025) examines public-law narratives in climate cases. In China, Zhu (2023) provides legal context and Li (2025b) compares two NGO-initiated cases. Clinics and legal processes are not assumed to be NGOs. Across these sources, litigation can serve more than one function, so multiple codes are retained where the publication documents information or brokerage alongside legal action.

A useful legal-outcome sequence distinguishes filing and admissibility, information disclosure, procedural access, administrative response, norm clarification, reputational effects, negotiated settlement, and downstream policy change. Comparative work should not reduce these outcomes to a single win–loss indicator because standing rules, remedies, and court–agency relations differ across jurisdictions.

The China-focused publications offer a useful setting in which to examine advocacy under administrative hierarchy, but the evidence base is small and heterogeneous. Zhang (2024) traces expertise and policy-network access; Li (2025b) compares two NGO-initiated cases; Wang (2025) examines local coalition processes; Zhu (2023) provides legal context; Yang and Walker (2024) discuss just transition and social inclusion. These sources do not form a common comparative test of political hierarchy or advocacy success. Their value here is theory generation and identification of mechanisms for future research.

Yu et al. (2026) identify 26,057 climate-related policy documents issued in China from 2000 to 2022: 1,938 national, 14,193 provincial, and 9,926 municipal documents. Figure 7 reproduces this government-document distribution as context for the scale of the policy system. The dataset does not measure NGO brokerage, access, or influence. It indicates that policy authority and outputs are distributed across levels, which may create translation demands; whether NGOs meet those demands is a separate empirical question.

Note. The bars report counts in Yu et al. (2026), n = 26,057. The same study identifies 19,609 mitigation and 6,448 adaptation documents. These counts describe government policy documents, not NGO activity or influence.

Legal action offers a further bounded observation. Zhu (2023) describes the institutional context for climate litigation in China, and Li (2025b) traces two cases initiated by Friends of Nature. These sources suggest that institutional adaptation may shape the effects of litigation, but they cannot establish how all Chinese NGOs use law or how legal strategies compare with public mobilisation across political systems.

Li and Farid (2026) provide a post-window example of transnational policy entrepreneurship around the Belt and Road Initiative. It is treated as contextual evidence, not included in the 24-publication corpus. Such work may extend advocacy across domestic policy networks, international standards, corporate governance, and finance, but the strength and consequences of these links require direct study.

Embedded access can support translation and repeated interaction, but it can also produce dependence, selective access, professionalization, agenda narrowing, or co-optation. These are potential trade-offs, not secondary qualifications. The framework therefore treats expertise and brokerage as possible resources whose effects depend on accountability, access conditions, and opportunities for public scrutiny. The China observations generate a hypothesis for comparison with other hierarchical or semi-authoritarian settings; they do not confirm a general hierarchy effect. These risks can also alter the evidence that researchers observe. Organisations with stable access may be more visible in policy documents and interviews, while groups excluded from consultation may leave fewer records. Professionalization can improve the technical quality of submissions while shifting accountability away from affected communities. Legal action can clarify procedure and also expose organisations to resource costs or political constraints. A comparative design should therefore record who gained access, who remained absent, the conditions attached to participation, and whether organisations retained the capacity to contest implementation. These are potential moderators and countervailing outcomes, not reasons to presume either success or failure.

Table 3 presents selected mechanism observations to illustrate how the framework distinguishes NGO evidence from movement, public-authority, business, and other contextual evidence. The full 16-observation coding matrix is supplied in the data workbook.