A new generation of AI tools promises to automate data collection, gap analysis, and narrative generation for ESG reports. We talked to the sustainability teams who have been testing them for 18 months, and the results are more nuanced than the vendors admit.
Key Takeaways
When Holcim's global sustainability team began piloting an AI-assisted reporting platform in late 2024, the initial results felt transformative. Data that had previously required six weeks of manual aggregation across 70 operating companies was compiled in under 72 hours. Gap analysis against GRI Standards and the CSRD's European Sustainability Reporting Standards was flagged automatically. The team estimated it had recovered more than 800 person-hours in the first reporting cycle alone. Then came the first narrative draft, and the mood shifted considerably.
The AI had produced polished, confident prose. It had also invented a comparison to an industry benchmark that did not exist, cited a CDP methodology update from a year that predated the CDP's founding, and subtly misstated the company's Scope 2 boundary approach. None of these errors were obvious at a glance. Two were only caught because a team member happened to be personally familiar with the source material. The experience is not unique. Across the 14 sustainability teams ESG Impact Weekly spoke to for this investigation, representing organisations from mid-cap manufacturers to FTSE 100 conglomerates, the pattern repeats: impressive operational gains in structured data tasks, serious reliability problems in open-ended narrative generation.
To be clear, the efficiency case for AI in ESG reporting is real. The tools performing best are those focused narrowly on what AI does well: pattern recognition across large structured datasets, automated gap mapping against frameworks, and alert generation when reported figures fall outside expected ranges. Workiva's AI layer, which the company rolled out progressively through 2025, has drawn consistent praise for its ability to flag inconsistencies between narrative claims and underlying data tables, a task that previously consumed days of manual cross-referencing. Persefoni's carbon accounting integration now surfaces anomalous emission factors automatically, reducing the risk that outdated conversion values persist into final disclosures. These are genuine advances, and the teams using them for these bounded purposes report high satisfaction.
The picture changes when AI is deployed for what vendors sometimes call "narrative intelligence," the automated drafting of qualitative disclosures, target-setting rationales, and stakeholder-facing commentary. Here, the underlying limitation of large language models becomes a material liability: these systems are optimized to produce fluent, confident text, not accurate text. In a domain where precision is legally and reputationally consequential, fluency without accuracy is worse than no automation at all.
"The risk is not that the AI produces obviously wrong output. The risk is that it produces wrong output that reads exactly like right output. In a reporting context, that distinction is everything."
Dr. Naomi Brandt, Director of Sustainability Reporting, Ørsted
ESG reporting is a particularly dangerous domain for AI hallucination for several reasons. The frameworks are complex and frequently updated, meaning an AI trained even six months ago may apply outdated requirements. The source material is heterogeneous, mixing quantitative operational data, legal commitments, board-level policy statements, and third-party assurance opinions. And the downstream consequences of inaccuracy range from regulatory penalties under CSRD and SEC climate rules to reputational damage if errors are identified by journalists or activist investors after publication.
Among the teams we surveyed, the most common mitigation strategy is a two-stage review protocol in which all AI-generated narrative is treated as a zero-draft and subject to full editorial review by a subject-matter expert before it enters the formal drafting pipeline. Several teams have introduced what one sustainability director described as a "citation audit": every factual claim in an AI-generated section must be traced to a named primary source before the section advances. This adds time back to the process, partially eroding the efficiency gains the tool was meant to deliver. But teams using this approach consistently report fewer late-stage revisions and greater confidence in final output.
The emerging consensus among practitioners is that AI earns its place in the ESG reporting stack when it is deployed as a co-pilot for structured tasks, not an autonomous author for qualitative disclosure. The teams that have had the best 18-month experiences are those that scoped their AI deployment carefully, resisted vendor pressure to automate narrative generation prematurely, and invested in internal upskilling so that human reviewers could meaningfully evaluate AI output rather than simply accepting it.
For sustainability teams evaluating platforms now, the most important due diligence question is not "what can this AI generate?" but "what is the provenance of every claim it makes, and how does the platform surface that provenance for review?" Vendors who cannot answer that question clearly are selling a product that is not yet ready for high-stakes regulatory disclosure. The market will mature, and the tools will improve. But the organisations that treat AI as a complete solution today, rather than a carefully bounded accelerant, are taking on risk that no efficiency gain can justify.
From small businesses to global enterprises, companies are racing to adopt dedicated carbon accounting tools. We put five leading platforms through a rigorous head-to-head evaluation covering accuracy, auditability, and integration depth.
Remote sensing technology is giving investors and supply chain teams unprecedented visibility into deforestation, methane leaks, and land use change in real time. Here is how it works in practice and who is building the leading capabilities.
The SEC's finalized climate disclosure rules are reshaping how public companies report on climate risk. We break down the requirements, timelines, and practical steps sustainability teams need to take immediately.
