The data labs of three German ministries and GIZ developed an AI prototype to help delegations navigate the vast amount of UNFCCC submissions.
The 30th annual Climate Conference (COP30) is bringing together tens of thousands of participants and negotiators from over 190 countries in Belém, Brazil, until 21 November, for one of the most complex international negotiations in the world.
For these negotiators, participating in COP comes with a highly labour-intensive preparation process. Months in advance, they work through a vast amount of position papers, scientific statements, reports and presentations. While this extensive information is needed to capture the diverse situations and targets of countries, the process puts smaller delegations at a considerable disadvantage as they lack the capacity that larger delegations have with dozens of negotiators and extensive back-office support.
Processing UNFCCC submissions
In light of these challenges, the data labs of three German ministries and the Deutsche Gesellschaft für Internationale Zusammenarbeit (GIZ) GmbH developed an artificial intelligence (AI) protoype, which supports delegations with processing UNFCCC submissions to the COP process: NegotiateCOP.
NegotiateCOP is an innovative platform that leverages the UNFCCC submission portal, which contains thousands of country submissions. In its first development iteration NegotiateCOP includes UNFCCC submissions published in 2024 and 2025, including upcoming submissions at COP30. It is powered by large language models (LLMs) to analyse submissions to the portal and currently offers three core functions:
- The Submission Explorer allows content-based navigation of issues and submissions with interactive filters and metadata, identifying key requirements and “red lines” in country positions.
- The Position Comparison Feature offers a systematic comparison of the negotiating positions of negotiations parties, showing overlaps and disagreements.
- The Portal Chat is a tool for navigating and searching documents, issues, and submissions. It uses semantic search and AI to generate answers based on official UNFCCC submissions.
In Practice
Imagine you are preparing to discuss Loss and Damage financing with another delegation: Instead of working through all position papers and submissions individually, you can use NegotiateCOP to gain a comprehensive overview of both side’s “key asks” and “fixed positions”. In this case, you would find that while your delegation is primarily looking to improve the quality of financing instruments for Loss and Damage, the other delegation’s key demand is the flexibility of such instruments – a great starting point for your negotiations.
Following a whole-of-government-approach, the data labs of the Federal Ministry for Economic Cooperation and Development (BMZ), the Federal Foreign Office (AA) and the Federal Ministry for the Environment, Climate Protection and Nuclear Safety (BMUKN) as well as GIZ developed this prototype together with public sector consultancy PD. Right from the start, negotiators and policy experts from Germany and the Global South were part of the agile product development process in order to tailor the product as closely as possible to the needs of real delegates.
The web application is accessible without a login, hosted with IONOS on a server in Germany, and does not track any personal information, queries or interactions on the website, respecting the EU data privacy and AI regulations.
What Comes Next
NegotiateCOP is a first prototype and its journey will be far from over after COP30: Its performance will be evaluated, feedback will be obtained, and the prototype will be developed further to ensure that it is fit for purpose and meets the needs of small delegations and developing country parties.
Try out NegotiateCOP for yourself: https://negotiatecop.org/
If you have feedback on the utility and usability of NegotiateCOP, please feel free to reach out at negotiateCOP@bmz.bund.de. The development team would love to get in touch and develop NegotiateCOP according to your needs.


