At the NaturschutzDigital 2026 conference on the island of Vilm, Hannes Schritt and Aaron Best showed how generative AI can help analyse Natura 2000 management plans to identify conflicting objectives in peatland protection.

Peatlands store carbon and provide a home for rare species and habitats. Rewetting them can benefit both, but it can also lead to conflicts: Protected habitats and species that have developed under the existing conditions may be affected by higher water levels. How do Natura 2000 management plans deal with this? Ecologic Institute is investigating this on behalf of the German Federal Agency for Nature Conservation (BfN), together with biota and the Succow Foundation.

At the NaturschutzDigital 2026 conference (“Generative Artificial Intelligence in Nature Conservation: Perspectives from Research and Practice”), held from 2 to 5 November 2026 at the International Academy for Nature Conservation on the island of Vilm, Hannes Schritt and Aaron Best presented on 4 November how they use AI for this purpose. The plans often run to 50–300 pages. The team therefore developed a hybrid approach that combines three steps. First, a search tool scans the documents for keywords (syntactic analysis) and allows a preselection of suitable plans. Then an AI model analyses the relevant passages for their content (semantic analysis): Which rewetting measures are planned, which conflicting objectives are named, and how are they resolved? Finally, experts consolidate, verify and assess the results. Multi-step prompts with fixed identifiers for measures, conflicts, obstacles and synergies ensure that every finding remains traceable and is not counted twice.

The approach was developed in several pilot phases starting in February 2025. The results were compared with manual analyses and the prompts were repeatedly adjusted. It became clear that the models have become significantly more reliable in a short time.

The key lessons from the project:

• Analysis goals and assessment criteria must be clear from the start.
• Expert verification by humans remains indispensable.
• Prompts, document preparation and quality-control steps need to be retested after every model change.
• Spending less time with the original documents changes how subject knowledge is built. At the same time, skills in working with AI grow.

The analysis feeds into recommendations for improving management planning, which Ecologic is compiling in a guidance document for the BfN.