ClimateCheck 2026: Shared Task on Scientific Fact-Checking and Disinformation Narrative Classification of Climate-related Claims Hosted as part of the NSLP 2026 Workshop at LREC 2026 12 May 2026 Palma de Mallorca, Spain
https://nfdi4ds.github.io/nslp2026/docs/climatecheck_shared_task.html Competition on Codabench: to be announced
Task Overview The rise of climate discourse on social media offers new channels for public engagement but also amplifies mis- and disinformation. As online platforms increasingly shape public understanding of science, tools that ground claims in trustworthy, peer-reviewed evidence are necessary. The new 2026 iteration of ClimateCheck builds on the results and insights from the 2025 iteration (run at SDP 2025/ACL 2025), extending it by adding training data, a new task on classifying disinformation narratives in climate discourse, and a focus on sustainable solutions. We offer two tasks:
* Task 1: Abstract retrieval and claim verification Given a claim and a corpus of publications, retrieve the top 5 most relevant abstracts and classify each claim-abstract pair as supports, refutes, or not enough information. Evaluation: Recall@K (K=2, 5) and B-Pref (for retrieval) + Weighted F1 (for verification) based on gold data; additional unannotated documents will be evaluated automatically. In addition, we will ask participants to use CodeCarbonhttps://codecarbon.io/ to assess emissions and energy consumption at test inference. * Task 2: Disinformation narrative classification Given a claim, predict which climate disinformation narrative exists according to a predefined taxonomy. Evaluation: Macro-, micro-, and weighted-F1 scores based on annotated documents.
Important dates
* Release of datasets: December 15, 2025 (task 1); December 19, 2025 (task 2) * Testing phase begins: January 15, 2026 (Codabench link TBA) * Deadline for system submissions: February 16, 2026 * Deadline for paper submissions: February 20, 2026 * Notification of acceptance: March 13, 2026 * Camera-ready papers due: March 30, 2026 * Workshop: May 12, 2026
We encourage and invite participation from junior researchers and students from diverse backgrounds. Participants are also highly encouraged to submit a paper describing their systems to the NSLP 2026 workshop. Datasets, Evaluation, and Rankings
Task 1: Abstract retrieval and claim verification The dataset for task 1 will follow the same structure as the 2025 iteration, but with triple the amount of available training data. We will evaluate systems on both abstract retrieval and claim verification tasks in an end-to-end manner. Abstracts retrieval will be evaluated using Recall@K and B-Pref, while claim verification will be evaluated using weighted F1 scores. Gold annotations will be used for both, with an automatic evaluation approach to evaluate incomplete judgments iteratively (more details will be announced soon). In addition, this year’s iteration will focus on coming up with sustainable solutions, encouraging the development of systems that can potentially be used in real-world scenarios. Thus, we will ask participants to use the CodeCarbon library when running the test inference to measure emission rates and energy consumption. This will not, however, be counted towards the final rankings. Task 2: Disinformation narrative classification The dataset for task 2 will consist of the same claims used for task 1, each annotated with labels denoting whether the claim is an example of a known climate disinformation narrative, and if so, which one(s). We follow the CARDS taxonomy (levels 1 and 2) to annotate our claims in a multi-label manner. Results will be evaluated using macro-, micro- and weighted-F1 scores. Participants can take part in task 1, task 2, or both tasks (better yet – think of ways to incorporate task 2 into the task 1 pipeline!). Organisers
* Raia Abu Ahmad (DFKI, Germany) * Aida Usmanova (Leuphana University, Germany) * Max Upravitelev (XplaiNLP Group, Technische Universität Berlin, Germany) * Georg Rehm (DFKI, Germany)