The World Bank Group is testing how artificial intelligence can make regulatory research faster and more accessible through a project examining building energy laws across 119 cities in 110 economies. The initiative uses AI to answer detailed legal questions, cite the exact provisions supporting its conclusions and acknowledge when available evidence is insufficient.

The research supports the next edition of the World Bank Group’s Building Green dataset, which examines building energy regulations and their enforcement. The updated dataset is scheduled for publication in spring 2027.

The approach demonstrates how AI can assist large-scale legal research while maintaining a strong emphasis on accuracy, transparency and independent verification.

How the AI Research System Works

Researchers designed a structured process in which AI answers 60 legal questions for each city. The questions cover matters such as whether particular building energy requirements are mandatory, which legal instruments establish them and what penalties may apply.

The system prioritises official legal sources, including government gazettes, ministry websites, municipal websites and standards bodies. Each answer must be supported by a relevant legal excerpt and a link to the underlying source.

Rather than forcing an answer when the evidence is unclear, the system can respond that insufficient evidence has been found. This helps distinguish verified legal requirements from conclusions that cannot be adequately supported.

The workflow also checks for contradictions and unsupported claims before the answers proceed to further review.

Independent AI and Human Checks

One of the project’s central lessons is that a plausible answer is not necessarily a legally accurate answer. Early testing revealed that the AI could sometimes identify the substance of a requirement while citing the wrong legal instrument.

To reduce such errors, the World Bank team introduced three verification stages:

• Automated consistency checks: Identify contradictions and unsupported responses.
• Independent AI review: Anthropic’s Claude reviews the legal basis produced by the drafting model under analyst supervision.
• Human verification: Researchers check every answer against the underlying legal sources.

The drafting model is therefore not allowed to serve as the sole judge of its own work.

The researchers also refined their methodology to account for outdated regulations, amendments and situations in which detailed requirements appear in subordinate regulations rather than parent legislation.

Early Results Show Significant Time Savings

In the first 21 economies reviewed, approximately four out of five AI-generated answers were correct. Accuracy reached 75% or higher in 18 economies, with the strongest results reaching approximately 95%.

The system performed particularly well on questions requiring straightforward yes-or-no or fixed-choice answers, with nearly nine out of ten responses in these categories correct.

The team estimates that the workflow saves approximately 1,500 researcher hours. Previously, a researcher needed roughly two to three days per jurisdiction to identify relevant laws and prepare 60 answers. Reviewing AI-generated responses now takes around four to six hours per city.

The reported AI usage cost for testing and production between March and August 2026 was approximately $3,200 across all 119 cities, averaging about $27 per city.

These figures demonstrate the potential efficiency gains, while also showing why independent verification remains necessary.

Separating Written Laws From Real-World Enforcement

The project distinguishes between what regulations require and how those requirements operate in practice.

AI is used for the legal research because published laws can be examined and cited. A separate questionnaire collects information from local architects, engineers and officials about enforcement, inspection procedures, compliance costs and the practical availability of qualified professionals.

This distinction matters because a city may have comprehensive energy efficiency regulations on paper without consistently enforcing them. Combining legal findings with local expert assessments can help policymakers identify gaps between formal requirements and actual implementation.

Why Building Energy Laws Matter

Building energy codes establish minimum efficiency requirements for aspects of construction and building operation. Effective standards can help reduce energy consumption, lower emissions and support more sustainable urban development.

As cities expand and new buildings are constructed, the quality of today’s regulations can influence energy use and environmental performance for decades.

The World Bank’s research illustrates how AI can help governments and development institutions compare complex regulatory systems across many jurisdictions. Its broader lesson is that reliable AI-assisted research depends on clear questions, traceable evidence, independent checks and a willingness to acknowledge uncertainty.

The goal is not to eliminate human expertise, but to use technology to accelerate research while preserving the accountability needed for trustworthy legal data.