Source-linked AI summary
Ctrl-F-Resist. Practices, Challenges, and Technical Needs of Civil Society Organizations Monitoring the Far-Right Online
Elisabeth Steffen, Helena Mihaljević
TL;DR
This paper examines how Germany-based civil society organizations monitor antidemocratic online dynamics amid limited resources, restricted platform access, legal uncertainty, and limited prior research on their work. Through qualitative interviews with 15 practitioners from 12 organizations, it identifies predominantly manual monitoring, openness to AI-supported media processing and discovery, skepticism toward automated classification, and implications for flexible monitoring infrastructures.
Problem
Existing monitoring research offers limited insight into how civil society organizations conduct long-term, locally embedded monitoring while navigating technical, ethical, and legal constraints.
Method
The study uses qualitative interviews with 15 practitioners from 12 Germany-based civil society organizations engaged in online monitoring and related activities.
Results
Monitoring remains largely manual because organizations lack direct platform access, tailored tools, and ongoing technical support, while participants prioritize discovery, analysis, and documentation over automated classification.
Takeaways & Limitations
Monitoring technologies should be flexible and adaptable across varied organizational practices and thematic foci, emphasizing search, multimodal data preprocessing, evidence collection, and documentation.
Takeaways & Limitations
The study focuses on German-speaking organizations in a limited sample, so transferring its findings across Europe and other political or socioeconomic contexts requires caution.
Abstract
from arXiv · showhide
As far-right actors increasingly exploit online platforms to disseminate ideology and mobilize supporters, civil society organizations (CSOs) play a vital yet underrecognized role in monitoring antidemocratic dynamics online. Unlike fact-checkers or content moderators, CSOs engage in long-term, contextualized analysis, often in resource-constrained settings and under precarious conditions. Despite their critical societal role, CSOs face significant barriers to adopting or co-developing technical solutions, including legal uncertainty, limited platform access, and chronic underfunding. Existing research and tool development efforts have largely overlooked these actors in favor of more institutionally embedded stakeholders. This paper addresses this gap through a qualitative study with 15 practitioners from 12 Germany-based CSOs engaged in online monitoring, positioning them as key yet overlooked stakeholders in the governance of digital spaces. We explore their current practices, challenges, and expectations regarding technological support. Our findings show that monitoring remains largely manual due to the lack of tailored tools, with enhanced search capabilities emerging as the most pressing technical need. While participants express openness to AI-supported features such as media processing and content discovery, many remain skeptical of automated classification, citing concerns around trust, legal usability, and professional credibility. Grounded in these findings, we introduce a conceptual monitoring workflow and describe its implementation in an open-source Telegram monitoring prototype designed to flexibly support diverse monitoring goals. We outline concrete design, policy, and research recommendatios, and introduce the manual labor trap as an empirically grounded concept that explains why monitoring CSOs tend to remain locked into labor-intensive, low-capacity arrangements.
1 Introduction
Civil society organizations are important but underexplored actors in monitoring antidemocratic online dynamics. This study examines their practices, challenges, and technical needs, finding largely manual work, strong demand for multimodal search, and skepticism toward automated classification.
- Research gap: Research on online harms has largely centered on fact-checking and content moderation, leaving CSO-specific monitoring needs and workflows underexplored.The paper identifies tailored technical tools for CSO monitoring as an open challenge.
- Study focus: The study investigates CSO monitoring practices, challenges, and technology expectations through qualitative research with 15 practitioners from 12 Germany-based organizations.Its contribution includes an empirical account of monitoring under complex organizational and political conditions.
- Findings: CSO monitoring remains largely manual because organizations lack tailored technical tools, while ethical and regulatory challenges hinder collaboration and data sharing.These barriers limit resource-efficient, collaborative technical innovation.
- Technical needs: Improved search across audio, images, video, and other modalities is the most frequently articulated technical need.Participants also viewed transcription and OCR as potentially useful AI-supported media-processing tasks.
- Trust and automation: Participants were skeptical of automated content classification because AI may threaten legal validity, professional credibility, and the value of human expertise.Trust concerns include perceptions that AI is unreliable and difficult to use as legally meaningful evidence.
- Design contribution: The paper introduces a conceptual monitoring workflow and implements it in an open-source prototype with configurable components for diverse CSO monitoring goals.The prototype is designed to accommodate varying and often limited technical capacities.
2 Related Work
Related work positions CSO monitoring as distinct, long-term, locally embedded work shaped by scarce resources, fragile data access, and heterogeneous knowledge. Existing tools and research often emphasize intervention or standardized classification, whereas CSOs need flexible support for discovery, interpretation, and human judgment.
- CSO monitoring: CSOs conduct long-term evidence gathering, contextual analysis, reporting, advisory, and educational work that differs from journalism, fact-checking, and content moderation.Their local knowledge can support early warning and context-specific responses.
- Research gap: Academic knowledge about CSO monitoring remains limited and often derives from grey literature rather than systematic research.The paper identifies direct empirical research on how CSOs conduct monitoring as scarce.
- Structural conditions: Chronic reliance on short-term funding limits CSOs’ investment in infrastructure, monitoring tools, and dedicated technical staff.In-house tools developed by larger organizations are typically tailored to their own resources and not openly shared.
- Data access: Monitoring depends on fragile, restricted, and opaque platform data access, particularly as affordable analytics access and messaging-platform access decline.These conditions create structural vulnerability for platform-dependent data collection.
- Technical limitations: Current learning-based systems may not align with CSO concepts or contextual sensitivities, especially for locally embedded phenomena and underrepresented language settings.General-purpose classifiers may miss phenomena that do not fit platform policies or legal thresholds.
- Technological support: Monitoring prioritizes observation and sensemaking, suggesting stronger demand for assistive human-AI tools than for systems making intervention or classification decisions.Related work describes openness to AI for narrow tasks while emphasizing human control, transparency, adjustability, and user control.
3 Methodology
The study combines qualitative interviews with practitioners from Germany-based CSOs, thematic analysis, and an iterative research-to-practice collaboration developing an open-source prototype. Ethical safeguards addressed the sensitivity of participants’ work, while the sample and researcher relationship introduce important scope and positionality considerations.
- Research design: The research-to-practice project paired qualitative inquiry with iterative development of an open-source prototype informed by CSO requirements.The study provides groundwork for defining and prioritizing software features.
- Scope: The study focused technically on Telegram while seeking transferable insights into CSO workflows, challenges, and requirements across platforms and communication modalities.Interview questions therefore included other platforms and services.
- Participants and interviews: Participants represented organizations engaged in monitoring, education, archiving, analysis, publication, consulting, communication, and security-related work.Three participants had educational backgrounds in technology that sometimes shaped their discussion of tool functionality.
- Participants and interviews: The researchers conducted 13 remote semi-structured interviews with 15 participants from 12 CSOs between October 2024 and January 2025.Interviews lasted 30 minutes to one hour and generally involved one participant.
- Ethics: The study protected participants through informed procedures, local transcription, anonymization, encrypted storage, and deletion of audio and video files.Participant and organization details were withheld because disclosure could increase exposure to monitored actors.
- Positionality: Mediated participant access fostered trust but may have shaped how interviewees framed issues because researchers were known to develop AI-based detection tools.The authors note that participants nevertheless frequently expressed critical views of automated detection.
- Data analysis: Thematic Analysis used inductive, data-driven analysis alongside research-question-structured interview guidelines.The coding process developed 21 themes and 497 codes through collaborative review and discussion.
4 Findings
CSO monitoring is largely manual and fragmented across platforms, modalities, and ad hoc documentation practices. Participants identified stronger cross-platform and multimodal search, reliable evidence handling, and legally workable collaboration as central needs, while facing access, legal, security, and tool limitations.
- Current practices: Monitoring remains largely manual, relying on platform clients, screenshots, notes, local files, and occasional in-house scripts.Telegram API use and systematic databases are rare, leaving documentation and data management time-consuming and unsystematic.
- Relevant platforms and services: CSOs monitor diverse and shifting platforms, with cross-platform analysis needed to build a complex picture of actors and phenomena.Platform relevance varies over time, and newer services can disrupt established archiving workflows.
- Multimodality of content: Effective monitoring must cover multimodal content, but audio and video are costly to store, process, and review manually.Text remains comparatively efficient, while podcasts, livestreams, demonstrations, sharepics, and coded emojis can provide important evidence.
- Technical infrastructure and data access: Existing tools are constrained by limited platform access, subscription costs, API limits, poor transparency, and cumbersome manual processing workflows.Instagram, Facebook, Meta, and X access is especially restricted or costly, while large data volumes exceed available analytical capacity.
- Ethical, legal, and security challenges: Legal uncertainty and security risks constrain evidence retention, monitoring scope, collaboration, and practitioner safety.Data protection and copyright can prevent sharing, while harassment risks lead practitioners to use disposable or mock accounts.
- Needs and expectations: The most prominent technical need is advanced search across unknown channels and multimodal content, supplemented by dashboards and configurable monitoring.Participants requested global Telegram search, alerts, saved queries, transcript-based discovery, and interfaces for filtering, statistics, and channel-level patterns.
5 Discussion and Recommendations
The discussion links CSO monitoring needs to precarious infrastructures, recommending flexible, human-centered tools alongside policy and research changes. It also frames persistent manual work as a structural problem requiring collaborative sociotechnical scaling.
- Empirical findings: CSOs rely on labor-intensive manual monitoring because they lack direct platform access, tailored tools, and sustained technical support.These constraints reflect limited resources, restricted platform access, and legal uncertainty.
- Design recommendations: Participants favor multimodal discovery, analysis, and documentation over automated classification, which raises concerns about workflow fit, transparency, and legal credibility.Search should combine multimodal and multilingual capabilities with lexical and semantic methods, while preserving human oversight.
- Design recommendations: Privacy and evidentiary requirements create retention, export, and external-service trade-offs for monitoring infrastructure.Tools must balance data minimization and secure deletion with forensic-quality, legally admissible evidence.
- Policy and research recommendations: The paper recommends recognizing CSOs institutionally and aligning funding, legal frameworks, and research priorities with their long-term monitoring work.Research priorities should emphasize foundational capabilities such as robust preprocessing for noisy, real-world, and non-English data.
- Policy and research recommendations: Interactive systems with lightweight human feedback loops are proposed to improve context-specific AI support and trust within everyday workflows.The recommendation responds to skepticism toward generic, English-centric AI models and automated classification.
- The Manual Labor Trap and Inter-organizational Collaboration: The manual labor trap describes how brittle, labor-intensive arrangements consume resources needed to build durable infrastructure, making collaborative scaling a proposed alternative.The paper cautions that collaboration itself is shaped by governance, ownership, power, and unequal capacity to contribute.
6 Limitations
The study’s findings are bounded by its German-speaking sample, variation across political and organizational contexts, and limitations in participant expertise and requirements analysis.
- Scope: The German-speaking sample limits transferability, especially across lower-income and authoritarian contexts with different infrastructure, threat models, and evidentiary conditions.The authors call for broader European and international research.
- Scope: Practices and needs varied within the sample, making design implications difficult to generalize while still requiring organization-specific adaptation.Core needs such as multimodal processing, robust search, and legal evidence handling appear broadly relevant but are not one-size-fits-all.
- Methodological limitations: Interviews with monitoring professionals, sometimes prompted by suggested features, may have narrowed the range of articulated technical needs.The authors also state that deeper requirements analysis would benefit from additional participatory and contextual methods.
7 Conclusion
The paper positions CSOs as crucial but understudied actors whose heterogeneous, precarious monitoring work requires adaptable technologies. It contributes empirical findings, actionable recommendations, and the manual labor trap as a basis for theorizing collaborative scaling.
- Conclusion: CSOs are crucial yet understudied actors whose diverse activities and obstacles require flexible technologies adaptable to varied organizational practices and thematic foci.The conclusion emphasizes the heterogeneity of civil society monitoring.
- Conclusion: The paper provides an empirically grounded account of monitoring as collaborative, infrastructural, and epistemic work under precarious, adversarial, and legally uncertain conditions.This extends understanding of sociotechnical systems beyond well-resourced institutional settings.
- Conclusion: It translates these findings into design and research implications focused on flexible infrastructure, search, multimodal preprocessing, evidence collection, and documentation.The recommendations are intended to remain adaptable across thematic domains.
- Conclusion: The manual labor trap explains why CSOs remain in labor-intensive, low-capacity arrangements and motivates theorizing collaborative sociotechnical scaling.The proposed theory concerns alternative collaborative modes of organization and infrastructure maintenance.
A.1 Participants
Table 5 provides an overview of the interview participants and their organizational and technical backgrounds.
- Participants: The participant overview identifies each employing organization’s thematic focus.
- Participants: The table records each participant’s main activity within the study.
- Participants: The overview indicates whether participants had formal technical education.
A.2.1 Information Phase. •
The information phase introduced the study and established the interview context, including confidentiality and the organisation’s work.
- The interview began with information about the research, its objectives, and confidentiality.
- Participants were asked to describe their organisation’s work in general terms.
A.2.2 Warm-up Phase.
The warm-up phase asked participants to identify their organisation’s focal topics and main work focus.
- Participants were asked to describe their organisation’s focal topics and main focus of work.Examples included counselling, monitoring, and educational work.
A.2.3 Main Phase.
The main phase examined organisations’ online monitoring practices, research processes, challenges, resources, software needs, AI expectations, infrastructure, and collaboration.
- Participants were asked about analysing antidemocratic actors online, including relevant phenomena, platforms, Telegram, media types, methods, and languages.
- The interview explored typical research processes, encountered technical or legal challenges, and available human and financial resources.
- Participants discussed existing research software, its helpful features, and desired improvements or missing functionality.
- Questions addressed where AI could support participants’ work, ideal functionality, generative AI, and the role of generated content in monitoring.
- The phase also covered technical infrastructure, information exchange, and data sharing with others.
- The interview concluded by inviting additional comments and informing participants about publication and workshops to test the software.