How AI can reliably record greenhouse gas emissions

As the climate crisis intensifies, the demand for transparent and reliable data on corporate greenhouse gas (GHG) emissions has never been greater. Large companies in the EU are legally required to disclose their emissions, but extracting this data from lengthy PDF sustainability reports is a complex and often error-prone task.

Artificial Intelligence (AI), and specifically Large Language Models (LLMs), are increasingly being used to automate the extraction of sustainability data. While these technologies promise efficiency and scale, concerns remain over their accuracy and the risk of misreporting. To address this challenge, researchers at the Social Data Science and AI Lab (SODA Lab) at LMU have developed a gold-standard dataset designed to improve how GHG emissions are recorded, analysed, and compared across industries.

Why accurate GHG emissions reporting matters

Transparent sustainability reporting is critical for:

Regulatory compliance: Companies face increasing pressure from the EU Corporate Sustainability Reporting Directive (CSRD) and global climate disclosure frameworks.
Investor confidence: Reliable emissions data enables financial institutions to assess climate-related risks.
Progress tracking: Consistent data supports corporate net-zero targets and broader climate action strategies.
Digital construction and built environment: Accurate emissions data informs carbon accounting in construction projects, supply chain monitoring, and lifecycle assessment.

Without robust data collection and verification methods, the credibility of sustainability reports is undermined—delaying climate action and reducing trust among regulators, investors, and the public.

The challenge of extracting GHG emissions data

Extracting emissions data from corporate reports is far from straightforward.

Researchers highlight several challenges:

1. Complexity of Sustainability Reports

Reports vary widely in structure and terminology.
Data is often buried in long, unstructured PDF documents.
Different frameworks (e.g., GHG Protocol, ISO standards) complicate comparisons.

2. Limitations of AI and LLMs

Automation risks: While LLMs can scan and extract values, they are prone to hallucinations (generating incorrect or misleading data).
Over-reliance: Teams may mistakenly trust AI outputs without verifying accuracy.
Context sensitivity: AI may misinterpret ambiguous or incomplete disclosures.

3. Inconsistent Company Reporting

According to the LMU team:

Around 50% of sampled sustainability reports contained no usable emissions data.
Most disclosures focused only on Scope 1 (direct emissions) and Scope 2 (indirect emissions from purchased energy).
Scope 3 emissions (supply chain, business travel, transport) were frequently missing or incomplete.

A gold standard for GHG data extraction

To address these challenges, LMU’s Greenhouse Gas Insights and Sustainability Tracking (GIST) group developed a benchmark dataset published in Scientific Data.

Key Features of the Benchmark Dataset

Sampling from global indices: Reports from companies in the MSCI World Small Cap Index and the German DAX.
Rigorous annotation process: Multiple stages of extraction, annotation, and expert verification.
Clear rules and transparency: Strict annotation guidelines ensure consistent treatment of ambiguous cases.
Open access scripts and supplementary materials: Allow researchers and practitioners to replicate results and test AI methods.

This curated dataset acts as a reference point for evaluating AI-based approaches to sustainability reporting. By making assumptions and decisions explicit, it enables fair comparison of methods and improves the reproducibility of research in climate data science.

Implications for digital construction and the built environment

The construction industry is responsible for nearly 40% of global carbon emissions.

Reliable emissions data directly influences:

Building Information Modelling (BIM) integration: Linking verified emissions datasets to BIM workflows for accurate lifecycle carbon analysis.
Supply chain transparency: Identifying carbon-intensive materials and processes.
Green procurement strategies: Supporting low-carbon decision-making in infrastructure projects.
Regulatory compliance: Ensuring construction firms align with EU CSRD and ESG reporting standards.

By improving emissions data reliability, AI-driven methods—grounded in benchmark datasets—can help the construction sector accelerate its transition to net zero carbon buildings and infrastructure.

Future outlook: closing the data gap on the path to net zero

The LMU team’s work demonstrates that AI is not a silver bullet for sustainability reporting, but it can be a powerful tool when supported by rigorous datasets and expert oversight.

Moving forward, key priorities include:

Expanding benchmark datasets to cover more industries and geographies.
Standardising emissions reporting across frameworks to improve comparability.
Integrating verified datasets into digital platforms for real-time emissions monitoring.
Bridging the Scope 3 emissions gap, which remains the largest blind spot in corporate climate disclosures.

Accurate greenhouse gas emissions reporting is vital for regulatory compliance, investor confidence, and achieving net zero goals. While AI and Large Language Models can streamline data extraction from complex sustainability reports, unchecked automation introduces significant risks.

The LMU SODA Lab’s gold-standard dataset provides a critical reference for improving the reliability of emissions data, offering transparency, accuracy, and comparability. For the digital construction sector, this breakthrough paves the way for more trustworthy carbon accounting, enhanced lifecycle assessments, and informed decision-making on the journey to decarbonisation.

By combining AI innovation with robust verification processes, industries can close the emissions data gap—making climate reporting not just faster, but also far more reliable.

Are you a building professional? Sign up for a FREE MEMBERSHIP to upload news stories, post job vacancies, and connect with colleagues on our secure social feed.