Article Summary
- Different ESG rating providers can and do assign the same company very different scores, because there's no single standardized formula for what counts as "good" governance or "good" environmental performance.
- AI models used for ESG screening lean heavily on self-reported corporate disclosures, which means a company skilled at sustainability reporting can sometimes score better than one with genuinely stronger practices but weaker disclosure habits.
- An ESG label on a fund describes its screening methodology, not a guarantee of investment performance, environmental impact, or that every underlying holding meets your personal ethical standards.
"The four most dangerous words in investing are: 'this time it's different.'"
Sir John Templeton
A fund with a leaf icon and the word "sustainable" in its name looks straightforward until you open the fact sheet and find an oil major sitting in the top ten holdings. That contradiction isn't necessarily a mistake — it's usually the output of an AI-driven scoring model that weighed the company's governance practices, emissions-reduction targets, and safety disclosures, and decided its ESG score cleared the bar relative to its industry peers. Understanding how these scores get built, and how much they can disagree with each other, matters more than trusting the label on the fund itself.
How the Scoring Models Actually Work
Modern ESG scoring relies heavily on natural-language processing to digest an enormous volume of unstructured text: annual sustainability reports, regulatory filings, earnings call transcripts, and ongoing news coverage. The AI model scans this text for signals related to dozens of environmental, social, and governance factors — carbon emissions targets, board diversity, labor practices, data privacy incidents, executive pay structure — and converts those signals into a numeric or letter score, often benchmarked against a company's industry peers rather than against companies as a whole.
Some providers supplement this text-based analysis with alternative data sources, like satellite imagery used to estimate a facility's actual emissions or supply-chain mapping tools that flag labor-practice risk several tiers down from the company being scored. This is genuinely more sophisticated than the manual, analyst-driven ESG research of a decade ago, but it inherits a familiar AI limitation: the model is only as good as the data it can access and interpret, and much of the most decision-relevant information about a company's real-world practices simply isn't captured in public disclosures at all.
Why the Same Company Gets Different Scores From Different Providers
One of the most well-documented issues in ESG investing is that major rating providers frequently disagree, sometimes substantially, about how sustainable the same company is. This happens because each provider's AI model uses a different methodology: different weightings between the environmental, social, and governance pillars, different peer groups for benchmarking, and different judgment calls about which controversies matter most. A company might score strongly on governance from one provider and weakly on environmental impact from another, and the two overall scores can land in very different places.
This divergence matters practically because it means an ESG label alone tells you very little without understanding which methodology produced it. Two funds both marketed as "ESG" can hold meaningfully different companies, because their AI screening tools are optimizing for different definitions of what counts as responsible. Before assuming a fund matches your personal values, it's worth checking the actual holdings and the specific rating methodology the fund uses to screen them, rather than relying on the fund's name or marketing category.
The Self-Reporting Problem
Much of the raw data feeding these AI models comes directly from the companies being scored, in the form of voluntary sustainability reports and disclosures. That creates an incentive structure where companies with well-resourced sustainability communications teams can score better simply by disclosing more thoroughly and favorably, regardless of whether their underlying practices are actually stronger than a smaller competitor that discloses less. Larger companies, which can afford dedicated ESG reporting staff, have historically tended to score better on average than smaller companies for this reason alone, independent of actual environmental or social performance.
Regulators in a number of markets have been pushing toward more standardized, mandatory ESG-related disclosure requirements specifically to reduce this gap, but the disclosure landscape still varies a lot by country and industry as of now. Until reporting becomes more uniform, AI-generated ESG scores will continue to reflect a mix of genuine performance and quality of self-reported paperwork, and it's difficult for an outside investor to know exactly how much of the score is which.
A Framework for Using ESG Scores Without Over-Trusting Them
Treat an AI-generated ESG score as one input rather than a verdict. A practical approach is to identify the two or three ESG factors that actually matter most to you personally — carbon emissions, labor practices, board diversity, weapons or tobacco exclusions, whatever it is — and check a fund's actual exclusion criteria and top holdings against those specific priorities, instead of relying on an aggregate score that blends dozens of factors you may not equally care about.
It's also worth reading which rating provider a given fund uses and doing a quick sanity check on a company you know well: if a fund's methodology rates a company you have specific knowledge or concerns about surprisingly favorably, that's a signal about the methodology's blind spots, not necessarily proof the model is wrong. ESG investing built on AI screening can still be a legitimate way to align a portfolio with personal values, but it requires the same due diligence as any other investment decision, applied to the screening methodology itself rather than assumed away by a green label.