NegotiateCOP - AI for climate negotiations

The problem

NegotiateCOP addresses the growing complexity and information overload in international climate negotiations. Conference of the Parties (COP) processes generate thousands of pages of submissions, draft decisions, and negotiation texts that delegations must analyse and compare within very limited timeframes.

This creates a major efficiency and capacity challenge for all negotiators, but especially for smaller and resource-constrained delegations that often lack dedicated analytical teams. As a result, important information can be overlooked, preparation becomes highly time-intensive, and participation in negotiations can become uneven.

The solution

To address the challenge of information overload in climate negotiations, we developed NegotiateCOP - an AI-powered platform that helps negotiators analyse and navigate large volumes of COP documents more efficiently.

The platform uses a RAG (Retrieval-Augmented Generation) architecture built with Haystack, Elasticsearch, and GPT OSS to process official negotiation documents, retrieve relevant information, and generate structured, context-based insights. This allows users to quickly compare country positions, identify negotiation trends, and trace changes across negotiation rounds.

We chose this approach because traditional document review is highly manual, time-consuming, and difficult to scale in fast-moving multilateral negotiations. The goal was not to replace negotiators, but to augment their analytical capacity and make complex negotiations more accessible, efficient, and inclusive - especially for smaller delegations with limited resources.

The impact

As a prototype, NegotiateCOP has already shown strong qualitative impact in supporting negotiators during climate negotiations. Users reported that the platform significantly reduced the time needed to search, review, and compare negotiation documents, allowing them to focus more on strategy and coordination rather than manual document analysis.

The platform enabled negotiators to identify relevant country positions and text changes across large document sets within minutes rather than hours. This was particularly valuable for smaller delegations with limited analytical capacity.

NegotiateCOP also generated strong visibility and positive feedback beyond the direct user community. The project received international attention, including coverage in Germany’s Heute Journal and an Associated Press article, highlighting its potential as an innovative AI application for multilateral negotiations.

While the current version remains a prototype, the next iteration for COP31 will significantly expand the document database, improve AI capabilities, and introduce more advanced analytical features such as longitudinal position tracking and country profiles.

Lessons learned

What worked particularly well was the close collaboration between policy experts, negotiators, and technical teams from multiple ministries and organisations. Early user interviews and testing sessions with German and international negotiators helped ensure that the platform addressed real negotiation workflows and practical pain points.

The RAG-based approach also proved effective for navigating large volumes of negotiation documents while keeping outputs grounded in source material. Users especially valued the ability to quickly compare positions and identify relevant information across multiple texts.

At the same time, the prototype highlighted important limitations. Negotiation documents are highly complex, politically nuanced, and often inconsistently structured, which makes reliable retrieval and contextual interpretation challenging in some cases. In addition, users wanted more advanced features such as historical position tracking, better summarisation of negotiation dynamics, and personalised workflows.

If we were starting again, we would invest earlier in standardised data structures and long-term document pipelines to simplify scaling and improve consistency across negotiation cycles. We would also involve a broader range of international users even earlier in the development process to further strengthen usability across different negotiation contexts and capacities.

NegotiateCOP was developed within the regulatory and governance framework of the German public sector and international negotiation processes. Several legal and policy considerations shaped both the technical design and operational approach of the project.

A key consideration was compliance with data protection and IT security requirements, particularly regarding the handling of sensitive negotiation-related documents and user interactions. Although the platform only processes public COP documents, the system architecture was designed with secure data handling and transparency principles in mind.

The project was also shaped by broader public sector requirements for transparency, interoperability, and vendor independence. This influenced the decision to build the platform on modular, largely open-source components such as Haystack and Elasticsearch, as well as on GPT OSS.

In addition, the project operated within standard public procurement and governance structures for interministerial digital innovation projects involving multiple German ministries and implementing organisations.