Saturday, 22 August 2026
Artificial intelligence is rapidly changing how investors collect and analyse corporate emissions information, but the technology is exposing rather than eliminating one of the biggest weaknesses in climate reporting: the underlying data is still not consistently comparable.
AI systems can scan thousands of sustainability reports, regulatory filings and corporate documents, identify greenhouse gas figures and convert unstructured disclosures into datasets far faster than conventional analyst teams. This creates significant opportunities for asset managers trying to assess climate exposure across portfolios containing hundreds or thousands of companies.
The challenge begins after the number has been found. A company’s reported emissions depend on decisions about which subsidiaries, operations and assets are included, how organisational boundaries are established, which calculation methods are applied and how purchased electricity is treated. Two businesses can therefore publish apparently comparable Scope 1 or Scope 2 figures that have been constructed on materially different bases.
This remains sufficiently important that the Greenhouse Gas Protocol is undertaking a major overhaul of corporate carbon accounting. In July 2026, it announced that it would work with the International Organization for Standardization to combine their corporate carbon accounting frameworks into a single harmonised global standard. The objective is to improve consistency and confidence in greenhouse gas information used across companies, markets and investment decisions.
The development also reinforces an important qualification to claims about the scale of today’s disclosure problem. While there is substantial professional evidence that emissions reporting suffers from gaps and inconsistencies, the assertion that roughly 40% of disclosed Scope 1 and Scope 2 figures lacked adequate scope or operational boundaries as of February 2026 could not be independently confirmed from a sufficiently authoritative source. That percentage should therefore not be presented as an established market statistic.
The wider issue, however, is well documented. In February, the GHG Protocol was still working through fundamental questions surrounding carbon accounting and said that greater harmonisation was necessary to provide clarity, build confidence and support investment. Its Scope 2 consultation, which closed on January 31, attracted almost 1,400 responses across two related consultations.
One of the clearest examples of the problem involves electricity.
Scope 2 covers indirect emissions associated with purchased electricity, steam, heating and cooling. Companies can report electricity emissions using location-based and market-based approaches. The former reflects the emissions characteristics of the electricity system supplying a location, while the latter incorporates qualifying contractual arrangements associated with the energy a company purchases.
An automated system could therefore find two different Scope 2 figures for the same company without either necessarily being incorrect. The investment question is which figure should be used, under which methodology and for what purpose.
The rules themselves are also changing. Proposed revisions to Scope 2 retain the dual reporting structure but seek greater precision in the emissions factors used for location-based calculations. Proposed changes to market-based accounting include requirements concerning the geographic deliverability of electricity and, in certain circumstances, closer matching between the timing of electricity consumption and clean-energy purchases.
The GHG Protocol has explicitly connected this work with rising expectations for credible, investment-quality climate information, arguing that greater consistency is necessary as emissions data becomes more closely integrated with financial reporting and capital allocation.
For investors, this creates another problem for automated analysis: historical figures are not necessarily fixed.
Companies can recalculate earlier emissions following acquisitions, disposals, changes in organisational structures, improved information or revisions to calculation methodologies. An investment system analysing a five-year emissions trend therefore needs to establish whether the historical figures remain comparable rather than simply extracting a number from each annual report.
This is where the distinction between automated extraction and investment-grade data becomes particularly important.
AI is highly suited to the first stage. It can locate emissions figures, identify reporting periods, classify disclosures and analyse large volumes of documents. It can also highlight missing information or inconsistencies that would take human analysts considerably longer to identify.
Reuters reported in January that AI is already being used across sustainability reporting to analyse unstructured information, identify discrepancies and help fill data gaps. But the same analysis identified risks from inaccurate underlying information, assumptions used in estimates, outdated disclosures and automatically generated reports that appear convincing despite weaknesses in their source material.
Greater automation can therefore magnify mistakes as easily as it can improve efficiency. An incorrectly classified emissions number entering an automated investment system can flow into company comparisons, portfolio carbon calculations, risk models and regulatory reporting before the original error is discovered.
The answer is unlikely to be a return to predominantly manual data collection. Instead, climate-data systems are moving towards a layered approach combining artificial intelligence with structured controls.
AI can perform much of the initial document search and extraction. Rules-based systems can then test reporting periods, units, corporate boundaries and methodologies and identify unexpected changes. Human specialists can concentrate on exceptions, conflicting information and situations where accounting judgement is required.
A further requirement is data lineage. For an institutional investor, a portfolio emissions figure should ideally be traceable backwards through the calculation chain: from portfolio to individual company, from company to emissions metric, from metric to methodology and reporting period, and ultimately back to the original corporate disclosure.
This is increasingly relevant as AI moves deeper into investment research. MSCI argued in July 2026 that speed alone is insufficient for institutional use and that AI-generated investment intelligence needs transparent methodologies, identifiable source data and systematic evaluation if decision-makers are expected to rely on the results.
That principle is particularly important for emissions information because many datasets contain a combination of reported figures, calculated values and estimates. Without clear provenance, an investor may not know whether a portfolio metric ultimately rests on a company disclosure, an external estimate or an assumption introduced to fill missing information.
The evolution of international carbon accounting should eventually improve the situation. The decision by the GHG Protocol and ISO to work towards a single harmonised corporate standard could reduce some of the methodological differences that currently complicate comparisons. But the process is not finished. The GHG Protocol says another consultation on Scope 2 is planned during 2026, with the final revised standard currently expected in 2027.
For investment managers, this means the technology used to process climate information must also be capable of adapting as accounting standards change. A system designed around today’s definitions cannot simply assume that future disclosures will be directly comparable with historical data.
The most reliable model is therefore not AI replacing climate analysts. It is a combination of AI extraction, accounting rules, automated validation, source traceability, exception management and specialist review.
Artificial intelligence can dramatically reduce the amount of manual work involved in finding and organising corporate climate information. It can potentially examine thousands of reports in the time previously required to analyse a fraction of them.
But processing more information does not automatically create better information.
For investors, the critical question is no longer whether AI can find a company’s emissions number. Increasingly, it can do that extremely quickly. The more important question is whether the number has been calculated on a comparable basis, whether its methodology is understood, whether historical figures remain consistent and whether the result can be traced back to a reliable source.
Until corporate carbon accounting becomes substantially more standardised, investment-grade emissions analysis will continue to depend on something more sophisticated than artificial intelligence alone.
Source: CIJ.World Research & Analysis Team
