Skip to main content
Aug 21, 2023

AI tool allows anyone to generate score for sustainability reports

Team of researchers wants to use tech to expose greenwashing

A new artificial intelligence (AI)-powered tool has launched that allows users to assess the quality of climate-related disclosures in sustainability reports.

Anyone can upload a report and get a score based on how well it performs against the 11 TCFD recommendations.

The team of researchers behind the project, called ChatReport, wants to open up access to ESG analysis and lower the reliance on ESG ratings agencies.

Markus Leippold
Markus Leippold, University of Zurich

The AI model highlights the possibilities of generative AI to source and interpret information, says Markus Leippold, a professor at the University of Zurich and one of several academics who developed the tool.

He notes, however, that the AI model is reviewing only corporate disclosure, not a company’s actual actions to tackle climate change. ‘It’s just about analyzing the quality of the talk,’ he says.

Climate and AI

It’s already possible to use a generative AI tool like ChatGPT to ask questions about a company’s disclosure, but the results are unreliable and may include invented information, referred to as ‘hallucinations’.

While built using ChatGPT, ChatReport is designed to stay grounded and use only information it has retrieved from the report in question.

Once registered and logged in, users can upload any sustainability report. After a few minutes, the platform provides a headline ‘conformity score’ out of 100, along with additional marks and text commentary against the TCFD recommendations.

‘Overall, the disclosure provides some insight into the board’s oversight, but lacks specific details,’ says ChatReport about one report inputted by IR Magazine. There is also a ‘ClimateQ&A’ box where you can ask your own questions about the report’s contents.

Transparency matters

ChatReport is the latest in a series of initiatives from the same team focused on sustainability information. Two years ago, it published a paper about a language model called ClimateBert that could analyze TCFD disclosures, although it wasn’t released as an online tool.

Next came an AI model trained on the reports produced by the Intergovernmental Panel on Climate Change (IPCC), allowing users to ask questions about the latest trends in global warming.

IPCC reports are ‘basically the most authoritative sources of climate science,’ says Leippold.  ‘But people do not read IPCC reports – so why not give them the kind of tool where they can ask any questions they have?’

The idea for ChatReport, meanwhile, came from a long-standing interest in the ESG ratings industry.

‘I always thought that what rating agencies do is so non-transparent and they disagree on the ratings,’ says Leippold. ‘Why not provide a tool so that everybody… can assess the quality of information companies are providing?’

Feedback call

For their part, IR teams may find such tools useful to review their disclosure against TCFD guidelines, as well as check whether AI models can correctly pick up information in their reports.

Leippold says he welcomes any feedback on the project. ‘The good thing is we are academics, so we don’t have a business model behind it,’ he says. ‘We can make errors, right? We just throw out something, ask for feedback and hope to improve it, but don’t claim it is the ultimate truth.’

Looking to the future, Leippold says it would be interesting to expand the scope of the analysis beyond the TCFD recommendations.

‘These kind of things can be easily automated,’ he says. ‘If we had more resources available, I would tell one of my research assistants to encode… the GRI guidelines or some other reporting guidelines.’

But these are all ‘little pieces’ of the bigger project, he adds, which is to put everything together and create a system for AI-based greenwashing detection: ‘In the next four years, that is what we are looking for.’