RepRisk screens over 150,000 public sources in 30 languages every day (RepRisk, 2025). MSCI analyses ESG data across 17,000+ issuers using AI to extract information from unstructured documents (MSCI, 2024). S&P Global runs 39 NLP-derived sentiment metrics across the earnings calls of 11,600 companies (S&P Global / Kensho). Bloomberg built a 50-billion-parameter language model trained on 40 years of financial data (Bloomberg, 2023). By the time a human analyst opens your sustainability report, an algorithm has already decided whether it's worth reading.

The Screening Layer You Can't See

Most companies write sustainability reports for human readers. The first audience, in practice, is a machine. Rating agencies, data providers, and institutional investors have invested billions in AI infrastructure that pre-screens, classifies, and scores sustainability communications before any analyst engages. This creates a filtering layer that determines which companies get attention and which get overlooked.

The scale of automated screening makes manual review of every company's sustainability communications physically impossible. RepRisk processes roughly 2.5 million documents daily and distils them into risk signals for 300,000 companies across 108 risk factors (RepRisk, 2025). Morningstar Sustainalytics covers 16,000 companies using machine learning, NLP, and large language models across its entire information value chain (Morningstar Sustainalytics, 2024). MSCI continuously screens media reports, regulatory actions, and civil society sources for ESG controversies (MSCI, 2024). These systems run every day. They don't wait for reporting season.

The investment community has matched this pace. PwC's 2025 Global Investor Survey found that 34% of investors now use generative AI directly in investment decisions, and 36% incorporate alternative data sources (PwC, 2025). BlackRock integrated RepRisk's AI-generated risk signals into its Aladdin platform, putting automated ESG screening directly into the workflow of the world's largest asset manager (RepRisk & BlackRock, 2025). The BIS, Deutsche Bundesbank, and ECB jointly developed Project Gaia, which uses generative AI to extract climate-related KPIs from corporate reports at a scale they described as "previously unimaginable" (BIS, 2024). Central banks are reading your disclosures with AI too.

This is the environment your sustainability report enters. Before a portfolio manager or ratings analyst reads a single paragraph, automated systems have scanned the document, classified its content, compared it against peer disclosures, and flagged whether it warrants human attention. The companies that don't pass that filter don't get a second chance to make their case.

What the Algorithms Detect

AI systems screening sustainability communications are increasingly sophisticated at distinguishing substance from noise. The question for communications teams is whether their reports are optimised for the audience that actually reads them first.

Researchers at the University of Zurich and ETH Zurich developed ClimateBert, a deep learning algorithm trained to identify "cheap talk" in corporate climate disclosures. Analysing annual reports from MSCI World index firms, they found that voluntary climate disclosures are associated with more cheap talk, and that cheap talk correlates with increased negative news coverage and higher emissions growth. Only targeted climate engagement was associated with less (Bingler et al., 2024). The implication is pointed: if your sustainability report contains vague language that sounds good but commits to nothing specific, algorithms can now identify that pattern and quantify it.

A separate study used 13.8 million text samples from corporate reports and news to pre-train NLP models capable of detecting environmental, social, and governance content in corporate disclosures. The communication patterns these models identified explained variations in ESG ratings across providers (Schimanski et al., 2024). What a company says, and how specifically it says it, is now a measurable input to the scores that investors use.

Research on ESG-washing detection has gone further. An NLP-based ESG Sentiment Index compares sentiment analysis scores with substantive sustainability content scores across the reports of 749 globally listed companies to identify companies whose positive language outpaces their actual sustainability commitments (Kobeissi et al., 2024). The distance between what you claim and what you substantiate is being measured algorithmically. The tools for detecting greenwashing are no longer limited to investigative journalists and activist NGOs. They've been industrialised.

EY's 2024 survey of institutional investors found that 57% believe AI could enhance the credibility and accuracy of corporate disclosures by enabling better detection of inconsistencies (EY, 2024). Investors are actively building the infrastructure to stress-test your sustainability narrative.

Why Most Companies Are Writing for the Wrong Audience

The gap between how companies produce sustainability reports and how those reports are actually consumed has become a strategic liability.

The typical sustainability reporting process follows a predictable pattern. The sustainability team compiles data, works with communications to craft a narrative, routes the report through legal review, and publishes a polished PDF. The intended audience is human stakeholders: investors, rating agencies, customers, regulators. The document is designed to be read linearly, with carefully sequenced arguments building toward favourable conclusions.

Automated screening systems don't read this way. They parse. They classify. They compare against sector benchmarks and peer disclosures. They identify whether specific, quantifiable commitments are present or absent. They flag language patterns associated with vagueness, hedging, or unsubstantiated claims. A 120-page sustainability report and a 40-page one that covers the material factors with precision and specificity will produce very different outcomes from these systems. The longer report isn't an advantage. Algorithms reward information density and specificity on material topics, and penalise dilution. The evaluation is sequential. First, can the content be extracted and processed cleanly? Then, is it written at a level that both human readers and AI systems can parse? Then, does it emphasise sustainability themes with sufficient weight relative to generic corporate language? Finally, are those themes aligned to what the market considers material in your sector? A report that fails at any stage immediately impacts the next one.

PwC's 2023 Global Investor Survey found that 94% of investors believe corporate sustainability reporting contains unsupported claims (PwC, 2023). EY's 2024 survey confirmed that 96% of finance leaders are concerned about the integrity and reliability of nonfinancial data (EY, 2024). Investors are filtering for credibility, and the filtering is increasingly automated. When these concerns are paired with AI tools purpose-built to detect exactly the inconsistencies they're worried about, the result is a screening environment that's becoming harder to pass with generic, qualitative narratives.

Companies providing so much information that investors struggle to discern which factors are truly material to the business aren't demonstrating thoroughness. They're creating noise that automated systems filter against. Every immaterial data point buried in a lengthy report pushes the material ones further from the surface where algorithms can find them.

What Machine-Readable Credibility Looks Like

The companies whose sustainability communications perform well in this environment share specific characteristics.

They lead with material factors. MSCI's methodology focuses on financial materiality, with key issues weighted by industry rather than by universal checklist (MSCI, 2024). Companies that structure their reports around the same materiality logic that rating agencies use to score them are more likely to surface the right information in automated screening. This means fewer topics covered with more depth and specificity, anchored to the factors that investors and rating agencies in your sector consistently identify as financially significant.

They use precise, quantitative language. When ClimateBert identifies "cheap talk," it's detecting vague commitments, aspirational language without timelines, and claims without metrics. The opposite of cheap talk is a specific commitment with a baseline, a target, a timeline, and a verification mechanism. Companies that populate their reports with measurable, time-bound claims on material factors give algorithms the structured data they're trained to find.

They maintain consistency across channels. Automated systems read everything. RepRisk screens news media, social media, NGO publications, and regulatory filings (RepRisk, 2025). S&P Global runs NLP sentiment analysis across earnings call transcripts (S&P Global / Kensho). Bloomberg's language model processes financial documents at scale (Bloomberg, 2023). If your sustainability report tells one story and your CEO's earnings call tells another, algorithms will flag the discrepancy. And beyond your own channels, what's being said about your company across thousands of news sources shapes how algorithms evaluate your credibility too. The gap between your internal narrative and your external perception is now measurable. Consistency across all public communications on sustainability topics has become a prerequisite for credibility in an environment where cross-referencing is automated.

They treat data quality as a communication asset. Deloitte found that 57% of senior leaders cite data quality as their top sustainability challenge (Deloitte, 2024). Only 15% disclose Scope 3 emissions despite every major framework demanding it. Companies that invest in producing verifiable, audit-ready sustainability data on their material factors create a foundation that survives automated scrutiny. The data spine underneath the narrative is now the first thing screening systems evaluate.

The Question Your Communications Team Should Be Asking

Your next sustainability report will be read by algorithms before it reaches a human. Those algorithms are trained to detect vagueness, flag inconsistencies between your public channels, measure the gap between your claims and your evidence, and compare your disclosure quality against every company in your sector.

The communications teams still writing sustainability reports as narrative documents for human consumption are producing content optimised for an audience that encounters it second. The first audience is a classification system that decides in seconds whether your company merits deeper analysis or gets filtered into the middle of the pack.

The question is whether your sustainability communications are built for the audience that actually reads them first.

References

• RepRisk (2025). "RepRisk Approach & Methodology."
• RepRisk & BlackRock (2025). "RepRisk and BlackRock Expand Collaboration to Drive Transparency on Business Conduct."
• MSCI (2024). "ESG Ratings Methodology."
• S&P Global / Kensho. "Textual Data Analytics: Sentiment Scores and Behavioral Metrics."
• Morningstar Sustainalytics (2024). "Leveraging AI for ESG Assessments." White Paper.
• Wu, S. et al. (2023). "BloombergGPT: A Large Language Model for Finance."
• Bingler, J., Kraus, M., Leippold, M. & Webersinke, N. (2024). "How Cheap Talk in Climate Disclosures Relates to Climate Initiatives, Corporate Emissions, and Reputation Risk." Journal of Banking & Finance, 164.
• Schimanski, T., Reding, A., Reding, N., Bingler, J., Kraus, M. & Leippold, M. (2024). "Bridging the Gap in ESG Measurement: Using NLP to Quantify Environmental, Social, and Governance Communication." Finance Research Letters, 61, 104979.
• Kobeissi, A., Rahaman, M. & Gokhan, N. (2024). "ESG-Washing Detection in Corporate Sustainability Reports." International Review of Financial Analysis.
• BIS / Deutsche Bundesbank / ECB (2024). "Project Gaia: Enabling Climate Risk Analysis Using Generative AI."
• PwC (2023). "Global Investor Survey 2023."
• PwC (2025). "Global Investor Survey 2025."
• EY (2024). "Global Corporate Reporting Survey."
• EY (2024). "Global Institutional Investor Survey."
• Deloitte (2024). "Sustainability Action Report."