AI Sustainability Reports Need Verifiable Data and Claims

A sustainability report that once took months to assemble can now be drafted with generative AI in hours.

The faster draft creates a new problem for the team approving it. A sentence can sound authoritative even when its number came from the wrong year, its comparison uses a different reporting boundary, or its source cannot be found, and fluent prose gives no outward sign of which of those three is true.

Reporting teams combine facility records, utility bills, supplier information, estimates and prior-year disclosures, and the issue with AI in this process is bigger than whether the tool makes up facts outright. If those inputs are inconsistent, an AI tool can turn the inconsistency into fluent prose without resolving it, leaving executives to approve a polished account that is harder to check than the spreadsheets behind it. The useful measure of AI in reporting is whether it shortens production time while preserving a clear path from each material claim to its evidence.

Fluent Language Can Conceal a Data Problem

Consider a company that reports lower energy use after closing one facility and acquiring another. An AI-generated draft might accurately calculate a percentage change from two supplied totals, and still describe that change as an efficiency improvement even though the company's operating footprint changed underneath the number.

A similar problem arises when an emissions figure uses estimated supplier data in one year and reported supplier data in the next. The arithmetic may be correct, but a confident sentence about progress would need to explain the change in method, not just the change in the number.

IFRS S1 Already Requires Companies to Disclose Estimates and Assumptions

These are reporting judgments, not copyediting issues. Under IFRS S1 guidance, companies applying the standard identify estimates and disclose the significant inputs, assumptions and calculation methods behind them. Whether a particular company must apply IFRS S1 depends on its reporting commitments and jurisdiction, but the underlying discipline is useful more broadly. Readers need to know what was measured, what was estimated and what can reasonably be compared, whether or not a formal standard requires them to say so.

AI can help reporting teams organize records, flag apparent inconsistencies and prepare text for review, and those uses become more valuable when the system preserves links to the underlying data rather than smoothing over the seams between them. A practical review starts with the claims most likely to influence a decision, among them emissions totals, reductions against a baseline, renewable electricity purchases, water withdrawals, waste diversion and assertions about suppliers. For each, reviewers should be able to locate the source record, identify the reporting period and boundary, reproduce the calculation, and see who approved any estimate or adjustment.

That process also helps distinguish an unsupported sentence from a disputed one. If two facilities classify the same waste stream differently, the answer is to reconcile the definitions, not to ask AI to rewrite the paragraph more carefully. Better structured data points to a real benefit here. The Global Reporting Initiative's machine-readable Sustainability Taxonomy, built on XBRL, is designed for faster collection and more comparable disclosures, though structure alone does not establish that a company's source data or interpretation is correct, a limit most ESG assurance programs are still working through.

ISSA 5000 Takes Effect for Reporting Periods Starting December 15, 2026

Assurance raises the stakes for documentation. The International Auditing and Assurance Standards Board's ISSA 5000 provides an approach to assurance across sustainability topics and reporting frameworks, and it becomes effective for engagements on sustainability information reported for periods beginning on or after December 15, 2026, in jurisdictions that adopt it. Separately, the board has been examining how AI and other emerging tools affect quality management in audit and assurance work, and its 2026 account of global roundtables highlights governance, risk management and trust as issues practitioners are still working through. Neither development creates a blanket requirement to disclose every use of AI in a sustainability report.

For a company seeking assurance once ISSA 5000 applies, the immediate question is whether a reviewer can examine the evidence behind a disclosure. An AI-written paragraph without a preserved source trail adds work at exactly the point when a company needs to demonstrate how it arrived at the claim, a gap already showing up in how AI tools trained for speed can miss the operational context a human reviewer would catch.

Technology leaders and sustainability teams can close that gap before publication starts to look like production. Defining which data an AI tool may use, and preserving the source references and calculation versions behind each figure, keeps the evidence attached to a claim instead of buried behind it. Routing material claims to the people who own the underlying records, rather than to whoever drafted the paragraph, keeps ownership where the knowledge already lives, an approach some teams are already applying to cut disclosure production time without giving up that control. Words like improved, reduced or on track deserve a review of their own, separate from the number they describe, since they interpret the data rather than merely repeat it.

AI may make the next report faster to produce. Its value to the business will depend on whether the people signing off can still explain every consequential number and the claim built around it.