The machine writes better greenwash than humans do
Generative AI has a talent that should worry every board member, auditor, and fund manager in New Zealand. It produces sustainability disclosures that read as more polished, more comprehensive, and more convincing than human-drafted originals, even when the underlying data is thin, selective, or wrong.
In 2024, University of Auckland Business School Professor Charl de Villiers and co-authors from the University of Kent and The Open University published research examining how AI would reshape environmental reporting. The core problem is mechanical. AI systems are trained on datasets containing both genuine disclosures and greenwashed ones, and the model cannot tell the difference. It learns the language of credibility, not the substance of truth.
As de Villiers warned in 2024, “AI systems learn from existing data, which may carry social and cultural biases, and these biases can inadvertently be embedded in AI-generated sustainability reports”. The practical result is that a compliance team under deadline pressure can produce a climate statement that looks thorough to investors and auditors while containing material gaps nobody catches.
That is no longer a theoretical concern. It is happening against a regulatory backdrop that is tightening fast.
Four out of 123 companies got it right
New Zealand’s mandatory climate-related disclosure regime has been in effect since January 2023, covering roughly 400 large listed issuers, banks, insurers, and investment managers. Two years in, the numbers are not encouraging.
The McGuinness Institute’s January 2026 analysis of 123 NZ-listed companies found that 72% included full or partial climate disclosures in FY24 annual reports, up from 29% in FY23. Progress, on the surface. But of those 88 disclosing companies, just four made full disclosures. The remaining 84 applied a collective 417 adoption provisions, regulatory carve-outs that let companies defer the hardest parts of disclosure around financial impacts, transition planning, and comparative metrics.
That gap between participation and substance is precisely where AI-assisted greenwashing thrives. A company using adoption provisions to avoid disclosing hard numbers can still produce a beautifully written governance and strategy section, and generative AI makes that easier than ever.
The FMA is watching. Its second annual review cycle assessed 62 climate statements for reporting periods ending between December 2024 and November 2025, with an explicit focus on whether disclosures are fairly presented and free from material misstatement.
The FMA is closing the gap between claims and evidence
On 12 May 2026, the FMA released revised sustainability disclosure guidance for issuers of sustainability-marketed financial products. FMA executive director Louise Unger said the guidance “supports issuers in providing clear and accessible information so investors can be confident their investments align with their objectives.” The four obligations are blunt: claims must be clear, backed by evidence, consistent across channels, and properly disclose external relationships. A correction buried in a footnote cannot rescue a misleading headline.
The Fonterra case shows this enforcement posture is not decorative. In June 2025, Fonterra Brands settled a Greenpeace lawsuit over Anchor butter packaging that claimed “100 percent New Zealand grass-fed.” Fonterra’s own spokesman acknowledged its cows were 96% grass-fed, with the balance coming from imported palm kernel expeller. New Zealand imports around 2 million tonnes of PKE annually, largely for dairy. Greenpeace spokesperson Sinead Deighton-O’Flynn called the settlement “a win against corporate greenwash.”
That case turned on a four-word claim on a butter wrapper. Imagine the enforcement complexity when AI is drafting entire climate statements.
Consumers cannot tell and investors may not either
The verification problem runs deeper than boardrooms. In 2023, EY’s Future Consumer Index found that 55% of NZ consumers said misleading information prevented them from shopping sustainably. A separate 2023 study of 369 packaged seafood products in NZ supermarkets found 41.2% carried at least one self-declared environmental claim, with nearly 80% of those claims vague and contrary to Commerce Commission guidance.
That was before generative AI became a standard business tool. The commercial incentive to make sustainability claims is strong, the consumer ability to verify them is weak, and the technology to produce convincing but hollow disclosures is now freely available.
What boards need to do before the next filing
The practical implications are not complicated, but they require discipline. Boards using AI to assist with climate statements need explicit governance around it: documentation of where AI drafted content, human review of every AI-generated section against underlying data, and sign-off processes that treat machine-assisted disclosures with the same rigour as manually prepared ones.
Auditors need to interrogate data, not prose quality. A fluent, well-structured disclosure that cites real frameworks is not evidence of a genuine transition plan. And procurement and investment teams assessing ESG claims in bids or investor decks need to accept that a polished sustainability narrative is no longer a reliable proxy for accuracy.
The AI did not create the greenwashing problem. But it made the greenwashing indistinguishable from the real thing, and the regulatory system has not caught up.
Sources
- AI’s impact on environmental reporting — University of Auckland (2024-02-21)
- AI’s Impact On Environmental Reporting — Scoop (2024-02-21)
- Analysing Climate Statements in Annual Reports of NZSX-listed Companies — McGuinness Institute (2026-01-14)
- Climate-related Disclosures: Insights from our reviews 2026 — FMA (2026)
- FMA tells issuers how to back their green claims — Insurance Business NZ (2026-05-12)
- Fonterra settles activists’ misleading packaging lawsuit for ‘100 percent NZ grass-fed’ claims — RNZ (2025-06-11)
- Consumers struggling to verify sustainability claims – survey — RNZ (2023-09-19)
- Dolphin friendly? New research shows vague environmental labelling is common on NZ seafood products — RNZ (2023-11-27)