The Evolution of Coordination in a Collective Intelligence System: 25 Years of English Wikipedia and the Emergence of Generative AI
Neal Reeves, Maja Świeczkowska, Amy Rechkemmer, Elena Simperl
cs.HC · cs.AI · cs.SI
Quick verdict
25-year Wikipedia analysis finds coordination declining, siloed, with no lasting LLM shift; large-scale and credible, but statistical reporting is thin.
Summary
This empirical study analyses nearly 25 years of English Wikipedia editing (January 2001 to July 2025; 18.24 billion edits) to track how coordination and governance work evolved relative to article production, and whether LLMs changed these trajectories. The authors process full edit histories across five namespaces (main, Talk, User talk, Project, Project talk), filter bots using four sources, and apply Gini coefficients, Markov-chain session metrics (self-transition 'stickiness', spectral gap, mixing time) and an interrupted time-series difference-in-differences design centred on ChatGPT's launch (November 2022). They report that all namespaces declined from the ~2007 peak, governance-related namespaces most sharply (Project talk edits -78.01%), while the top 1% of editors took on a growing share and sessions became increasingly namespace-specialised. DiD models show immediate relative drops in governance engagement after ChatGPT (edits -22% in ns4, -25% in ns5, p<0.001) but no statistically significant long-term trend changes. Main limitations: no confidence intervals or multiple-comparison correction, an untested parallel-trends assumption, an internal inconsistency in the ChatGPT date (Oct 31 vs Nov 30), incomplete bot filtering, and figures not verifiable from the extracted text.
Review complete
Assessment at a glance
5
risk flags identified
● high 0
● medium 3
● low 2
4 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
- Participation in Wikipedia's coordination and governance namespaces has declined relative to article editing since the ~2007 peak, with a shrinking core of editors performing an increasing share of the work (§4.1, §5.1).
- Editing has become more specialised: stationary-weighted self-transition probability peaked at 0.89 (Nov 2024), the spectral gap hit its lowest value in July 2025, and mixing time rose from under 2 to over 5 months (§4.2).
- The release of ChatGPT was associated with immediate relative reductions in governance-namespace engagement (edits -22% in ns4, -25% in ns5, p<0.001) but no statistically significant long-term change in trends (§4.3).
- Talk-page (ns1) coordination has remained comparatively stable relative to article editing, avoiding the sharper governance-namespace declines (§4.1.5, §5.1).
Strengths
Unmatched longitudinal scale
18,241,443,327 edits across the full lifespan of English Wikipedia (21 Jan 2001 - 21 Jul 2025, §3.1), analysed across five namespaces covering content, discussion and governance (Table 1); the authors plausibly claim the first longitudinal evaluation spanning coordination and content spaces (§1).
Multi-method triangulation
Three independent analytic lenses converge: Gini coefficients and top-1% shares (Fig 2), within-session Markov metrics (Fig 5), and DiD-ITS models (§3.5.3), all supporting the same decline-plus-specialisation story rather than a single fragile statistic.
Transparent, multi-source bot filtering
Disciplined causal language on LLMs
Weaknesses
Statistical reporting too thin for the claims
DiD identification asserted, not demonstrated
Internal inconsistency on the intervention date
Operationalisation ambiguity in Markov metrics
Decline percentages lack a stated comparison window
Full Review Report
$1.99 · one-timeUnlock the complete preliminary analysis Evolve produced for this paper.
Report preview
Scope & Claims
The paper (arXiv:2609.16856, cs.HC) presents a longitudinal analysis of English Wikipedia from 21 January 2001 to 21 July 2025, covering 18,241,443,327 edits across five namespaces (Table 1: ns0 Main/articles, ns1 Talk, ns3 User talk, ns4 Project/governance, ns5 Project talk/governance discussion). Three research questions structure the work: (RQ1) how coordination scaled relative to content work over 25 years; (RQ2) what session-level Markov metrics reveal about specialisation and namespace transitions; (RQ3) whether ChatGPT's launch changed engagement in articulation spaces. The central claims are that coordination participation declined relative to content production with a shrinking core absorbing growing workload (ns5 edits -78.01% at peak), that editing became increasingly namespace-siloed (stickiness peaking at 0.89 in November 2024; mixing time rising from under 2 to over 5 months), and that LLMs caused only short-term shocks with no long-term trajectory change. Claims are mostly scoped to relative, quantitative patterns, and the authors explicitly disclaim causal attribution, quality effects, and the equation of all namespace activity with articulation work (Conclusion, §5.3).
The remaining sections cover Methodology Assessment, Statistical Evidence, Results & Reproducibility, Limitations & Risks, and Verdict & Recommendations.
- All 4 strengths & 5 weaknesses, analyzed in detail
- 5 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
Did ChatGPT hurt Wikipedia editing and governance?
According to this study, the launch of ChatGPT (November 2022) coincided with immediate relative drops in Wikipedia's governance and discussion namespaces - edit counts fell about 22% in the Project namespace and 25% in Project talk relative to the main article namespace (p<0.001). However, the authors find no statistically significant long-term change in engagement trends after the shock: the pre-existing trajectories continued largely unchanged.
Is English Wikipedia really declining?
The paper shows that activity declined in all five studied namespaces (articles, Talk, User talk, Project, Project talk) from a peak around 2007 through July 2025 - for example, article-namespace edits fell about 55% and Project talk edits about 78%, based on log-scale comparisons. A small core (the top 1% of editors by activity) now performs a growing share of edits in nearly all namespaces.
What data does the study use and how big is it?
The authors processed the complete English Wikipedia edit history: 18,241,443,327 edits from 21 January 2001 to 21 July 2025. They excluded about 233 million anonymous edits (1.28%) and removed bots identified via four sources (Wikimedia user-group dumps, a former user-groups dump, a community-maintained unflagged-bots list, and the 'All Wikipedia Bots' category).
What are the study's main limitations?
The authors cite four: some bots may remain unidentified; excluding anonymous edits (1.28% of all edits) may shift engagement dynamics; a first-order Markov assumption may miss longer-range editing patterns; and the 1-hour session cutoff may distort low-intensity editors' sessions. A reviewer would add: no confidence intervals or multiple-comparison correction, an untested parallel-trends assumption in the DiD, and an internal inconsistency about ChatGPT's release date (October vs November 2022).
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.
- Figures 1-5 carry much of the descriptive evidence but only captions were extracted; plotted values, peaks and inflections could not be visually verified
- No numerical tables beyond namespace definitions were present in the extracted text, so all coefficient figures rely on prose reporting
- The reference list was garbled by extraction; citation accuracy was not fully audited
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.