Some companies argue artificial intelligence should replace digital tagging of sustainability reports, but research shows AI is not up to the task. Moreover, tagging sustainability data isn’t just a technical exercise; it’s a declaration that companies stand behind, building the trust that businesses, governments, investors, and Civil Society Organisations need to do their work.
As artificial intelligence becomes increasingly commonplace, it is being presented as a solution for a broad swathe of tasks across different sectors. Company sustainability reporting is one example, which will be mandatory for companies operating in the EU with a turnover above €450 million and 1,000+ employees from 2027.
The European Sustainability Reporting Standards (ESRS) require companies to report digitally tagged data about their impact on the environment for the first time, helping data users understand companies’ exposure to risk and their effects on people and the planet.
However, a recent discussion paper from an organisation representing some of Germany’s largest companies argues that the digital tagging¹ of sustainability data is too costly and that AI should take over this task.
But could and should we use AI alone to digitally tag companies’ sustainability reports?
What is digital tagging and why is it important for data users?
In the US and the EU, it is already standard practice for financial reporting data to be digitally tagged. The practice means hundreds of data points can be easily identified and quickly compared. Think of it like adding a footnote or reference to an essay. When a tag is added to a data point in a report, that point is added to a reference list, making it searchable, downloadable and comparable across companies.
Digital tagging will soon be extended to companies’ sustainability data reported under the ESRS. The sustainability standards will inform data users, including investors, governments, and the public, how well they understand and respond to risk. For example, green transition plans, supply chain grievance mechanisms, and workforce data demonstrate a company’s preparedness for environmental or human rights based risks. This information is crucial for enabling investors, policymakers, and consumers to make informed decisions regarding companies.
However, this development has not been universally welcomed, and some companies are arguing that “AI makes iXBRL [digital tagging] reporting increasingly obsolete. […] Given the significant costs and risks for issuers and the lack of added value for investors the only meaningful consequence would be to abolish ESEF/iXBRL reporting requirements entirely.”
Could AI tag sustainability reports instead?
The argument from industry is simple: AI can read a PDF and pull out the data, so why go to the effort of tagging? On the face of it, this seems reasonable.
In practice, however, it’s a different story that presents significant risks.
Current AI models are poor at identifying data types within written reports. One study tested major large language models on their ability to match financial figures to the correct label from a list of over 17,000 standardized US-GAAP concepts found in SEC filings. Even the best-performing AI correctly matched a value to its label only 17% of the time. The research exposes a weakness: AI struggles with semantically similar concepts and cannot reliably tell them apart.
The problem is nuance.
Take an example from an imaginary PDF report: “Net revenues for the year were $4.2 billion.” A human tagger with financial reporting knowledge knows how to choose between: Revenues, Sales Revenue Net, or Revenues Net of Interest Expense. AI does not — at least not yet. It struggles where abstract terms sit close together. Only trained human judgement can tell them apart accurately and reliably.
Should we use AI instead of humans?
Even if AI could do the task accurately, there are good reasons why AI shouldn’t. Experts’ grasp of rules and language, and how the two operate together, is superior to that of AI. However, beyond accuracy there is a critical reason why AI shouldn’t do the task.
Tagging serves a purpose AI cannot replicate: it builds integrity that generates the confidence markets, businesses, and society operates on:
“AI […] operates probabilistically — it learns patterns and makes predictions, but it does not guarantee compliance with strict financial and non-financial reporting rules. If national business authorities, investors and other stakeholders rely solely on AI-extracted data, there is a significant risk of inaccuracies — including misassigned KPIs, incorrect units, and erroneous figures. These errors can for instance distort investor analyses of a given company, with potential consequences for stock prices and market trust.” — We Mean Business Coalition
When a company tags a value like its Scope 1, 2, or 3 CO₂ emissions, it is making a declaration that the figure is correct and that the company stands behind it. As one expert puts it:
“Tagging isn’t just a technical means to identify data for extraction; it is an explicit declaration which reflects a set of judgments about what the data means. There is value in knowing that the preparer has explicitly made those judgments and is willing to put their name to them, and they have to be communicated in a way that is unambiguous.” — CoreFiling
A company declaring a value and standing behind it matters, especially when the goal is to hold companies to account on issues like climate change and human rights.
Where AI does help
This is not to say that AI has no role.
AI-based tagging tools do exist to speed up the tagging process, but they cannot replace, or remove the need for, human oversight.
Likewise, once data is digitally tagged, AI becomes a powerful tool for analysis. As one expert in the field puts it: “XBRL [digital tagging] isn’t a burden; it’s the fuel for AI-powered analysis, ensuring accuracy, comparability, and decision-usefulness.”
Good tagging and good AI analysis are not in competition, they are complementary. One makes the other possible.
Cost of reporting and the price of sustainability risks
The financial cost of digital tagging is often cited by those in favour of AI to ditch it: “iXBRL implementation and maintenance are costly and draw on scarce resources during the critical time of preparing and auditing the annual report.”
It is true, there is a cost to tagging. However, the cost depends on the company’s approach. A digital first mindset and adopting a tagging on the go approach is quicker and more accurate than “bolt-on” tagging after the report is written. For someone with knowledge of reporting standards and the reported content, tagging can be done relatively quickly, as demonstrated in this video.
Putting an exact figure on tagging costs alone is difficult, but the total yearly cost for companies reporting under the European Sustainability Reporting Standards has been put at 0.01% of turnover. By contrast, one projection puts the total cumulative cost of climate hazard exposure for S&P Global 1200 companies at $25 trillion by 2050, equivalent to 74% of their total revenue, or 31% of the index’s total market capitalisation in 2024.
Getting sustainability data into an organised and accessible form is not a bureaucratic exercise. It is how we identify trends and opportunities for investment and sustainable innovation.
Financial markets have long depended on precise, comparable and reliable data. We need that same standard applied to sustainability data. The challenges ahead are shared, so we must rise and meet them together for a sustainable future.
Notes
¹ Digital tagging is also known as Inline eXtensible Business Reporting Language — or iXBRL for short.