1. What is EnvironmentGPT?
EnvironmentGPT is UNEP’s first public beta chatbot designed to make trusted environmental science easier to access. Built on a Retrieval-Augmented Generation (RAG) architecture, it provides accurate, cited answers drawn exclusively from UNEP-approved publications. Unlike general-purpose AI models trained on open internet data, EnvironmentGPT ensures transparency, scientific integrity, and reduced hallucinations. Future versions will be integrated into UNEP’s World Environment Situation Room (WESR).
2. Why It Was Built
UNEP’s mandate includes strengthening the science–policy interface. Policymakers, practitioners, and the public need rapid access to reliable environmental knowledge, yet information remains fragmented and complex to navigate. EnvironmentGPT helps bridge this gap by enabling users to:
- Summarize large volumes of information quickly
- Receive explanations tailored to different expertise levels
- Integrate knowledge across climate, biodiversity, pollution, water, land, health, and economic domains
-Access referenced, authoritative content in accessible formats
General-purpose LLMs cannot be assumed reliable for technical environmental topics; EnvironmentGPT offers a domain-specific, retrieval-based alternative aligned with UNEP’s priorities on climate stability, healthy ecosystems, and pollution reduction.
3. What Documents It Uses
The knowledge base is built using a three-tier selection framework prioritizing high-consensus global assessments, authoritative UN-led publications, and high-quality regional or peer-reviewed literature. Over 220 validated reports form the initial corpus, including:
- GEO-7
- Emissions & Adaptation Gap Reports
- Frontiers Reports
- IPCC & IPBES summaries
- Relevant FAO, WHO, and other UN publications
- Selected open-access scientific meta-analyses and systematic reviews
All documents were reviewed by UNEP’s Office of Science to ensure quality and thematic coherence.
4. Key Features
- Latest UNEP flagship assessments integrated from launch
- Filtering by publication or publication series
- Audience modes: Public, Policymaker, Scientist
- Multiple LLM options with sustainability profiles
- Confidence score based on breadth of supporting evidence
- Environmental impact metrics (energy, water, emissions, minerals per query), using EcoLogits
- Conversation memory for follow-up questions
- Full reference transparency
Improved AI environmental impact assessments
The Global Partnership on Marine Litter (GPML) Digital Platform is a multi-stakeholder platform that compiles different resources, connects stakeholders, and integrates data. This project is developing a matchmaking system to automatically match stakeholders based on their interests, skills, and other relevant information using a content-based recommender engine, natural language processing, and collaborative filtering.
Activity Type AI Tools/SolutionsResearch/Reports/AssessmentsThis project uses machine learning to identify areas of surface water from satellite imagery at a global level. It produces an annual dataset of surface water extent as a formal reporting indicator for SDG 6.6.1, which is the only globally derived satellite-based indicator in the SDG framework that member states can use in their national reporting.
Activity Type AI Tools/SolutionsResearch/Reports/AssessmentsThis exploratory research venture between UNEP and UNCTAD shows how machine learning models can help policymakers and researchers design data-driven policies that efficiently and effectively allocate government resources to maximize inclusive and sustainable prosperity and development.
Activity Type Research/Reports/AssessmentsAI Tools/Solutions