Ctrl-F-Resist. Practices, Challenges, and Technical Needs of Civil Society Organizations Monitoring the Far-Right Online
Elisabeth Steffen, Helena Mihaljević
cs.HC · cs.AI · cs.CL · cs.IR
Quick verdict
Interview study of 15 practitioners at 12 German CSOs: monitoring stays manual, search is top need; solid qualitative work, new concepts partly programmatic.
Summary
The paper presents a qualitative interview study (13 interviews, 15 practitioners, 12 Germany-based civil society organizations, October 2024–January 2025) of how CSOs monitor far-right activity online, analyzed with Thematic Analysis into 4 meta-themes, 21 themes, and 329 codes. Monitoring remains largely manual — screenshots, manual platform search, ad-hoc storage — because tailored tools are lacking. Key challenges are restricted platform data access, multimodal content, legal uncertainty (GDPR versus evidentiary needs), and security risks; enhanced search, including multimodal full-text search, alerts, and channel discovery, is the most pressing technical need. Participants welcome AI for media processing (ASR, OCR) but distrust automated classification, citing reliability, legal usability, and professional credibility. The authors derive a conceptual monitoring workflow, implement it as an open-source Telegram prototype (lexical and semantic search, transcription, a BERT-based conspiracy classifier, dashboards), and evaluate it formatively in two co-design workshops, with a longitudinal test ongoing. They contribute ten design and three policy recommendations and the 'manual labor trap' concept. Limitations include a Germany-only sample, purposive and snowball sampling, participants who are monitoring rather than AI experts, and formative-only evaluation of the prototype.
Review complete
Assessment at a glance
4
risk flags identified
● high 0
● medium 2
● low 2
5 strengths · 5 weaknesses · 5 author questions
AI text signal = Binoculars zero-shot detector score over 20 sampled passages (Fast-DetectGPT co-reported) — absence does not prove human authorship, and a flag does not imply misconduct.
Levels reflect verification complexity — not a quality score. Evolve never rates your paper.
Key claims under review
- Online monitoring by the studied CSOs remains largely manual (screenshots, manual search, ad-hoc storage) because tailored tools are lacking, with enhanced multimodal search, alerts, and channel discovery as the most pressing technical need.
- Practitioners accept AI for narrow media-processing tasks (ASR, OCR) but substantially distrust automated classification, citing misclassification risk, legal usability in court, and professional credibility.
- Legal uncertainty (GDPR versus evidentiary needs) and security threats structurally impede inter-organizational data sharing and collaboration.
- A conceptual monitoring workflow can be operationalized as an open-source, locally deployable Telegram prototype suited to low-capacity CSOs, validated formatively in co-design workshops.
- The 'manual labor trap' concept explains why resource-scarce CSOs stay locked into labor-intensive arrangements, motivating a proposed CSCW theory of collaborative sociotechnical scaling.
Strengths
Genuine gap filled with first-class evidence
Sections 1-2 establish that CSO monitoring is underexplored relative to fact-checking and content moderation (building on Bäumler et al. [9]); Sections 4.1-4.4 deliver the most granular empirical account to date of how CSOs actually work — manual Telegram search ('trigger the full-text search manually every time', P4), scrollshot documentation (P6), and rare evidentiary rigor via cryptographic hashing and Internet Archive archiving for a 'chain of evidence' (P13).
Unusually transparent qualitative methodology
Section 3.5 documents the full coding trajectory: 714 codes/25 themes → 497 codes/21 themes → final 4 meta-themes, 21 themes, 80 subthemes, 329 codes, with dual coding of two interviews and RQ mapping applied only after theme stabilization. The full code list is published on Zenodo (record 20447764) and the interview guide appears in Appendix A.2.
Exemplary ethics for adversarial-context research
Closed practice-to-artifact loop
Candid positionality and limitations
Weaknesses
Thin intercoder reliability
Demand characteristics in needs elicitation
Formative-only prototype evaluation
Proposed-not-demonstrated theory
Narrow sample base
Full Review Report
$1.99 · one-timeUnlock the complete preliminary analysis Evolve produced for this paper.
Report preview
Scope & Claims
This PACMHCI article (Vol. 10, No. 6, CSCW068; arXiv 2609.00808) studies how Germany-based civil society organizations (CSOs) monitor far-right activity online, positioning CSOs as stakeholders distinct from fact-checkers and content moderators. Three RQs structure the work: current practices and workflows (RQ1), technical/ethical/legal challenges (RQ2), and needs and expectations toward technology (RQ3). Central claims: (1) monitoring remains largely manual for lack of tailored tools; (2) enhanced search — multimodal full-text search, alerts, channel discovery — is the most pressing need; (3) practitioners accept AI for media processing (ASR/OCR) but distrust automated classification on grounds of trust, legal usability, and credibility; (4) a conceptual workflow can be implemented in an open-source Telegram prototype; (5) the 'manual labor trap' explains lock-in of low-capacity arrangements, with inter-organizational collaboration as the proposed escape.
The remaining sections cover Methodology Assessment, Statistical Evidence, Results & Reproducibility, Limitations & Risks, and Verdict & Recommendations.
- All 5 strengths & 5 weaknesses, analyzed in detail
- 4 risk flags with severity, section & reviewer questions
- Methods & statistical evidence assessment
- 5 questions a peer reviewer would ask
- Full six-dimension evaluation report & recommendations
One-time purchase · this paper only
One-time purchase, this paper only. AI-generated preliminary analysis — not peer review.
Frequently asked questions
What is the 'manual labor trap' in this paper?
An empirically grounded concept the authors introduce to describe how resource-poor civil society organizations monitoring far-right online activity stay locked into labor-intensive, brittle manual workflows (screenshots, manual platform search, ad-hoc storage). These arrangements consume the very time and money needed to build durable infrastructure, so the low-capacity state reproduces itself; the authors propose inter-organizational collaboration as a possible escape route and call for a CSCW theory of 'collaborative sociotechnical scaling'.
Who was studied and how?
The authors interviewed 15 practitioners from 12 Germany-based civil society organizations that monitor far-right activity online, in 13 semi-structured remote interviews conducted October 2024 to January 2025. Data were analyzed with Braun & Clarke Thematic Analysis (final: 4 meta-themes, 21 themes, 329 codes), complemented by two one-day co-design workshops testing an open-source Telegram monitoring prototype.
What do civil society monitors actually want from AI?
Mostly assistive media processing — automated speech transcription and OCR to make audio, video, and images searchable — plus better search across channels, saved queries with 24/7 alerts, and tools to discover new relevant channels. They remain skeptical of automated content classification, citing misclassification risk, the need to explain evidence in court, and the professional credibility of human expertise.
Is the monitoring tool or data from this study available?
Yes, partially. The Telegram monitoring prototype is open source (project site: arai-telegram.github.io), and the full qualitative code list is published on Zenodo (record 20447764). The interview transcripts themselves are not public, to protect anonymized participants.
About this review
This is an AI-generated preliminary review by Evolve.Pub, produced from the openly available text of the paper. It is not peer review and does not represent the authors' views. The canonical record of the paper is at the publisher link above.
- The provided text is a sampled excerpt: three '[...sampling gap...]' regions were skipped (parts of Sections 2.4-4.1.5, part of Section 4.6 including the head of Table 2, and part of the Conclusion). Judgments touching those regions, especially full Table 2 contents, are made from the visible remainder.
- Figures 1-5 are referenced and described in prose but their graphical content is not present in the extracted text; thematic-map structure was assessed from the surrounding text only.
Go deeper on this paper
Wrote a paper? Stress-test it before reviewers do.
Upload your manuscript for the same structured scrutiny — reasoning, statistics, evidence, figures, and reproducibility.