ESG And AI: the reality behind the myth

AI is often positioned as the force that will transform ESG analysis. Advocates present it as the technology that can finally tame the complexity of sustainability data. There is merit in this view, AI can certainly accelerate data collection, sift vast quantities of information, and highlight patterns quickly. Yet the real challenge is not the sophistication of the models, but the material they must process. ESG data is still highly unstructured, inconsistently reported, and frequently unaudited. Until those foundations improve, AI’s role will remain complementary rather than transformative.

Threading a path through complexity

The contrast with financial reporting is stark. Financial data is tagged, standardised, and designed for machine readability. ESG disclosure, by comparison, is fragmented and heterogeneous, scattered across formats and often expressed in qualitative terms.

Three issues stand out:

• Divergent definitions: even basic indicators, such as employee safety, are reported in a multitude of ways, forcing AI models to reconcile apples with oranges.

• Intangible and qualitative values: areas like culture or governance resist neat categorisation; algorithms struggle where judgment, context, and nuance are essential.

• Data integrity: with much ESG reporting unaudited, risks of inaccuracy and greenwashing remain. Outputs are only as reliable as inputs, and poor data compromises any automated analysis.

AI can increase the efficiency of data collection, but efficiency alone does not deliver insight. Without consistency and verification, analysis risks becoming faster, but not necessarily better.

The wings of ambition 

Another dimension lies in the forward-looking nature of much ESG disclosure. Targets such as net-zero pledges or transition pathways are inherently aspirational, designed to signal future intent rather than record past performance. Machine learning models, however, excel in historical analysis, and when asked to interpret ambition, credibility or probability, they falter.

In addition, regulation increasingly requires forward-looking assessments. This creates a fundamental tension: automated tools grounded in the past are being applied to disclosures that live firmly in the future. Human expertise is required to bridge that gap.

A many-headed challenge

Global regulation is moving, but rarely are jurisdictional areas in alignment. The EU’s Corporate Sustainability Reporting Directive (CSRD) represents a leap forward in scope and detail, mandating disclosures across more than 1,000 datapoints and extending coverage to tens of thousands of companies. The UK’s Sustainability Disclosure Requirements (SDR), aligned with ISSB standards, are still being finalised. In the United States, the SEC’s climate disclosure rules face legal challenge, leaving companies uncertain about compliance.

The result is a patchwork of overlapping obligations with regulations leaving a swath of exclusions and exceptions, meaning that a single standard reporting outcome is a long way off. For investors, this landscape makes truly comparable data difficult to achieve. Progress is real, but uneven, and automation struggles in the absence of global consistency.

Forging clarity from complexity

AI’s current limitations in ESG do not represent a failure of technology, rather they reflect the weakness of the underlying data environment. A genuine breakthrough requires a data-first approach:

• clearer, standardised definitions for key metrics

• materiality-driven thresholds to separate what is relevant from what is peripheral

• stronger assurance processes to improve trust in disclosures

• transparent methodologies that explain outcomes, not just deliver scores

Only when this foundation is in place can AI deliver on its promise: not as a replacement for human judgment, but as a tool that amplifies and complements it. The future of ESG analysis will depend less on clever algorithms than on whether the data itself becomes structured, reliable, and genuinely comparable across markets.