New Text-Mining Method Reveals Which Climate Topics Governments Fight Over at the IPCC

Every few years, the world’s governments gather in long, often exhausting sessions to approve the summary reports of the Intergovernmental Panel on Climate Change (IPCC), the United Nations body that distills the state of climate science into language policymakers can use. These approval meetings are famously contentious: delegates line-by-line negotiate the exact wording of statements about global warming, and the final text carries enormous weight because it feeds directly into climate negotiations under the UN Framework Convention on Climate Change (UNFCCC). A new study published in PLOS Climate now offers the first systematic, quantitative way to measure how intensely different governments engage with different topics in those meetings, and its findings complicate a long-standing assumption about who really controls the IPCC’s output.

For decades, scholars and observers have described IPCC approval sessions as dominated by a handful of large, well-resourced delegations, chiefly from wealthy countries, which can afford to send big teams of scientists, lawyers, and negotiators to scrutinize every sentence. The new research, conducted by Patrick Bayer, Lorenzo Crippa, Hannah Hughes, and Erlend A. T. Hermansen, does not overturn that picture entirely, but it adds a crucial layer of nuance. By applying natural language processing techniques to the meeting records, the team shows that engagement levels vary dramatically by topic, meaning that different groups of countries step forward or step back depending on what is being discussed.

The methodological innovation at the heart of the paper is what the authors call a text-as-data approach. Rather than relying on labor-intensive qualitative coding, in which human researchers read transcripts and manually classify each intervention, the team used computational tools from natural language processing to map the content of government statements and quantify how much attention each delegation devoted to particular subjects. The approach links the textual record of who spoke, how often, and about what, to the topical structure of the underlying scientific report, producing a topic-level map of engagement for every participating government.

To demonstrate the technique, the researchers focused on the IPCC’s Sixth Assessment Report (AR6), the most recent comprehensive evaluation of climate science, and specifically on two topics that proved pivotal in the first Global Stocktake of the Paris Agreement, the process by which countries collectively assess progress toward climate goals. Those two topics were carbon management, which covers technologies and practices for capturing and handling carbon dioxide, and nature-based solutions, which harness ecosystems such as forests and wetlands to mitigate and adapt to climate change. Both became identified policy approaches for national mitigation and adaptation plans in the Global Stocktake outcome, making them ideal test cases for studying governmental attention.

The choice of these two topics is analytically significant. Carbon management, including approaches like carbon capture and storage, tends to be championed by countries with large fossil fuel industries that see continued hydrocarbon production as compatible with ambitious climate targets, provided emissions can be captured. Nature-based solutions, by contrast, attract strong interest from countries with extensive forests or vulnerable ecosystems, as well as from those that emphasize the co-benefits of biodiversity protection and community resilience. Because the two topics carry different political and economic implications for different states, they offer a natural experiment for testing whether engagement in IPCC meetings is uniform or topic-dependent.

The analysis delivered what the authors describe as robust evidence that governments’ engagement levels do indeed vary by topic. In practical terms, this means a delegation that is nearly silent during discussions of one chapter may become one of the most active participants when another subject comes to the floor. The pattern suggests that countries allocate their scarce negotiating resources strategically, intervening where the scientific framing has the greatest potential to affect their national interests in subsequent UNFCCC negotiations. The IPCC approval process, in this view, functions as a key site for legitimating the knowledge that later flows into treaty negotiations, and governments treat it accordingly.

This finding nuances, rather than contradicts, the established literature. It remains true that a few large country delegations dominate overall speaking time across the agenda, and the new study does not dispute that asymmetry in resources and influence. What it adds is the recognition that dominance is not uniform: the topical distribution of attention reveals coalitions of interest that a purely aggregate view would miss. Small delegations may concentrate their limited capacity on the one or two issues that matter most to them, achieving a form of targeted influence that raw counts of interventions conceal.

The technical machinery behind the study is worth unpacking for what it promises beyond the IPCC. Text-as-data approaches in political science typically combine automated topic models, which discover latent themes in large document collections, with measures of actor-level attention, such as the frequency and length of statements mapped onto those themes. By aligning the structure of the AR6 report with the record of government interventions in approval meetings, the researchers could compute engagement scores per country per topic and then assess the robustness of the resulting patterns. The authors emphasize that the method complements, rather than replaces, qualitative research: close reading remains essential for interpreting why a delegation intervenes and what its wording choices mean, but the computational layer makes systematic comparison across dozens of countries and many agenda items feasible for the first time.

The implications reach well beyond climate science. Intergovernmental bodies from biodiversity treaty conferences to plastic-pollution negotiations all rely on similar approval and drafting processes in which scientific assessments are translated into policy-relevant language. A tool that can map which governments are most active on which issues, at scale and without years of manual coding, gives scholars a new instrument for studying the politics of global knowledge production. It also offers transparency benefits: civil society observers and journalists could use such techniques to document where contested scientific claims attract the heaviest governmental pushback, and how those battles shape the summaries that ministers later cite as the scientific basis for policy.

For the IPCC itself, the study arrives at a moment of introspection about the pace and inclusiveness of its procedures. Approval sessions have grown longer and more contentious as the stakes of climate policy have risen, and debates over topics such as carbon management have exposed deep fault lines between countries betting on technological fixes and those prioritizing ecological approaches. By quantifying engagement at the topic level, the new method provides an empirical foundation for conversations about representation and fairness in the panel’s most consequential meetings. The authors’ demonstration that engagement is strategically selective, not uniformly distributed, suggests that future assessments of whose voices shape climate science summaries should look not just at how much countries speak, but at what they choose to speak about.

Subject of Research: Quantifying topic-level government engagement in IPCC approval meetings using text-as-data methods

Article Title: Mapping and quantifying engagement levels of governments in IPCC approval meetings by topic from text

Article References: Bayer, P., Crippa, L., Hughes, H., & Hermansen, E. A. T. (2026). Mapping and quantifying engagement levels of governments in IPCC approval meetings by topic from text. PLOS Climate, 5(9), e0001057. https://doi.org/10.1371/journal.pclm.0001057

Image Credits: AI Generated

DOI: 10.1371/journal.pclm.0001057

Keywords: IPCC, climate policy, natural language processing, text-as-data, UNFCCC, Global Stocktake, carbon management, nature-based solutions, Sixth Assessment Report, intergovernmental negotiations, PLOS Climate, government engagement