Here's something I find odd about this moment in corporate sustainability: the same companies now deploying AI to improve their ESG reporting are, in many cases, the ones that spent the previous decade producing ESG reports that didn't reflect their actual environmental behavior. AI didn't create that problem. But it's worth asking whether it solves it, or just produces more convincing versions of the same thing.
That's the question this essay is actually about.
The Infrastructure of Disclosure
Corporate sustainability reporting has become its own industry. There are consulting firms, software platforms, disclosure frameworks (GRI, SASB, TCFD, CDP) and now a growing category of AI tools designed to help companies collect, organize, and communicate sustainability data. McKinsey estimated the market for ESG data and analytics at over $1 billion globally and growing, with demand accelerating as regulatory pressure increases on both sides of the Atlantic.
On the surface, this looks like progress. More data, better organized, more consistently reported. If AI can reduce the manual burden of collecting scope 1, 2, and 3 emissions data (cross-referencing supply chain records, flagging inconsistencies across reporting frameworks) that's useful. I'm not dismissing it.
But there's a structural problem hiding inside all this activity, and it has nothing to do with AI.
Reporting and performance aren't the same thing. A company can produce an increasingly sophisticated sustainability report (accurate, consistent, AI-verified) that documents an inadequate level of environmental performance. The report improves. The underlying operations don't. This is the basic mechanism of greenwashing, and it doesn't require anyone to lie. It requires only that the emphasis fall on disclosure quality instead of on what the business has actually changed about how it operates.
AI applied to reporting infrastructure makes the report better. That's not nothing. But it doesn't change the thing being reported on.
What "Greenwashing" Actually Describes
The term gets used loosely enough that it's worth being precise. Greenwashing operates through the structural relationship between a company's sustainability claims and its actual operations, not primarily through deception (though deception sometimes appears).
When a company installs solar panels on its headquarters while its supply chain runs on coal, then produces a report leading with the renewable energy percentage without adequate context about scope 3 emissions, that's greenwashing in the structural sense. Nobody necessarily lied. The solar panels exist. But the communication creates an impression of sustainability that the operational reality doesn't support.
What I've come to think of as the Sustainability Removal Test, developed in Michael Kovnick's paper Existential Sustainability: A Structural Approach to Anti-Extractive Tourism, offers a clean diagnostic for this. The test asks one question: can the sustainability practice be removed without causing immediate business failure?
Kovnick developed this framework in the context of tourism, but its logic extends much further. A carbon offset program can be removed without affecting operations. A renewable energy certificate purchase can be discontinued. A supplier code of conduct can sit in a PDF, unenforceable, and the business continues exactly as before. These are all removable. In Kovnick's terminology, they're performative sustainability: sustainability that exists in the communication layer, not in the operational structure.
Structural sustainability operates by a logic. When the sustainability practice is the business model, when removing it would cause the operation to fail, you have something entirely different. Interface's Mission Zero program, which I'll return to shortly, approached this in manufacturing. The tourism operator Kovnick examines has built it into every constraint defining how the business runs.
The question AI raises is: which version of sustainability is it actually improving?
AI as a Reporting Tool vs. AI as a Design Tool
The current conversation about AI and sustainability splits into two very distinct discussions that often get conflated.
The first is about AI as a reporting tool. This is where most of the investment and attention currently sits: natural language processing to extract emissions data from invoices, machine learning models to flag anomalies across large datasets, automated frameworks to cross-map data against GRI or TCFD requirements. These tools make reporting faster, more consistent, and in some cases more accurate.
The second discussion (much quieter, and considerably more important) is about AI as a design tool. Not for designing better reports, but for designing better businesses. For identifying, at the architecture level, where a business is structurally dependent on extractive practices, and what it would take to change that.
These aren't the same thing. The first discussion is about communication. The second is about structure. The sustainability community has spent far too much time in the first while the second barely gets started.
Interface, Again: Because It's Still the Right Example
Interface merits returning to here, not because it's the only example of structural sustainability in manufacturing, but because it's the most documented and the most honest about both its achievements and its shortcomings.
Ray Anderson's 1994 decision to pursue Mission Zero, zero environmental footprint by 2020, was a design decision, not primarily a reporting decision. He asked his engineers and product designers to figure out how to make carpet tiles from biological and recycled materials, using renewable energy, with zero waste to landfill. The sustainability targets were baked into what the product was, not into how it was described.
Between 1994 and 2019, Interface cut waste sent to landfill by 92% across manufacturing sites. Its LaGrange, Georgia facility reached 100% renewable electricity. By 2019, renewable energy accounted for 89% of electricity use globally. The company's Live Tile product achieved carbon neutrality through material redesign: bio-based materials that stored more carbon than was emitted during manufacture.
What AI could have done for Interface during that period is an interesting thought experiment. Better data collection, certainly. Faster anomaly detection in energy use. More consistent scope 3 tracking across a complex supply chain. These would have been useful. But what made Interface's sustainability structural wasn't data quality. It was the decision to make zero-impact a design criterion instead of a reporting target. A human leader made that decision in 1994, and the commitment ran for 26 years.
No AI tool produces that kind of decision. And no AI tool applied only to the reporting layer would have turned a greenwashing operation into Interface. The structural commitment came first. The reporting followed.
What the EU Taxonomy Is Actually Testing
The European Union's taxonomy regulation, which came into full force in 2022 and is still being refined, is the most serious regulatory attempt yet to force the distinction between performative and structural sustainability.
The taxonomy establishes technical screening criteria for what counts as a "sustainable" economic activity. It's not a reporting framework in the traditional sense. It's a classification system that asks whether an activity, a thing a company does, meets defined environmental thresholds. To qualify, activities must make a substantial contribution to at least one of six environmental objectives without doing significant harm to the others.
This differs from ESG scoring, which aggregates company-level metrics that can be gamed in various ways. The taxonomy asks about activities. A gas-fired power plant doesn't become sustainable because the company running it also has a good governance score or a well-designed sustainability report.
The EU Taxonomy Regulation is imperfect, the debates around natural gas and nuclear inclusion were a political mess, but its underlying logic is exactly right. It's trying to create a regulatory structure that can distinguish between a company that has truly redesigned activities around environmental criteria and one that reports well on activities that remain fundamentally unchanged.
AI is now being deployed to help companies determine which of their activities qualify under the taxonomy and to document the evidence trail. That's useful. But it's worth noticing that the taxonomy itself is doing the structural work: creating the criterion. AI is helping with compliance documentation. Those are separate contributions.
The Reporting Sophistication Trap
Here's something I worry about, and I think it deserves more attention than it gets.
As AI makes sustainability reporting more sophisticated (more detailed, more consistent, more defensible against regulatory scrutiny) it simultaneously raises the floor for performative sustainability. Companies that are primarily greenwashing can now do so with better data, more rigorous methodology, and higher-quality disclosure than was possible five years ago. The report looks more like Interface's. The operations may look nothing like Interface's.
This is what I'd call the reporting sophistication trap. Better reporting tools don't discriminate between companies that have made structural changes and companies that haven't. They help both produce better reports. The gap between appearance and reality doesn't necessarily shrink; in some cases it may widen, because the appearance becomes more convincing.
The Sustainability Removal Test cuts through reporting quality entirely. You don't ask how good the report is. You ask: if you removed the sustainability practice tomorrow, would the business fail? If yes, the sustainability is structural. If no, it's performative: regardless of how well it's documented.
A company whose entire supply chain, product design, and operational infrastructure revolves around zero-waste principles can't remove those practices without dismantling its product. That's existential sustainability. A company that purchases offsets and produces quarterly ESG reports can remove both without affecting a single operational decision. The reports may look similar. The structures are completely different.
Tourism as a Design Laboratory
Kovnick's framework is worth examining in a corporate sustainability context because small-scale tourism operations have structural properties that make the distinction between performative and existential sustainability unusually visible.
The operation he studies runs at maximum 250 guests per year, 18 guests per group, 14 weeks of operation per destination per year. These aren't voluntary aspirations that could be revised upward if revenue pressure mounted. They're structural constraints embedded in the operating model. Local revenue retention runs at 72%, compared to an industry norm of 20-30%. After 20 years, partner retention is at 100%.
What's interesting about these numbers is what they reveal about design intent. A 72% local retention rate doesn't emerge from a report. It emerges from who gets paid, in what amounts, under what contractual arrangements, for what activities. It's a structural property of how money moves through the operation. Remove it (shift to lower-cost suppliers, reduce local partner margins, bring logistics in-house) and the operation no longer works as designed. The guests don't return. The partners leave. The whole thing unravels.
That's existential sustainability in practice. The sustainability isn't in the disclosure. It's in the architecture.
Most corporate sustainability programs are nowhere near this. They're designed to perform sustainability to stakeholders (investors, regulators, customers) while leaving the underlying operational architecture largely intact. AI can help them perform better. It can't make the architecture structural.
What Would AI-Enabled Structural Sustainability Actually Look Like?
I want to push on this, because there's a version of AI application in sustainability that's genuinely different from what's currently dominant.
Imagine AI deployed not to improve disclosure, but to model the structural dependencies in a business. Which supplier relationships could be redesigned to eliminate extractive dynamics? Where in the supply chain does environmental harm occur that could be addressed through operational redesign rather than offset purchase? What would it actually cost to restructure material flows so that sustainability was a design criterion instead of a mitigation strategy?
That's design work, not reporting work. It requires different inputs (operational data, not just financial data) and different outputs. The goal is a different business.
Some companies are beginning to use AI this way. Life cycle assessment tools that model the full environmental footprint of product design decisions before manufacturing begins. Supply chain mapping tools that identify concentration risk and extraction points. Scenario modeling that lets designers test whether a given operational structure can achieve zero-impact targets without compromising financial viability.
These tools are less visible than ESG reporting platforms, partly because they don't produce the disclosure documents that regulators and investors currently reward. They're upstream of disclosure. But they're doing structurally more important work.
The 18% net margins documented in Kovnick's paper, earned while retaining 72% of revenue locally and operating at strict capacity limits, come from that kind of structural design work. Financial viability and sustainability are the same thing. That's what structural design produces when it's done well.
The Regulatory Incentive Problem
One reason performative sustainability persists even as AI improves reporting quality is that the regulatory incentive structure rewards disclosure instead of structural change.
If your ESG score improves when your report improves, and your report improves when your data collection improves, and AI helps your data collection improve, then AI helps your ESG score improve without necessarily changing anything about how your business operates. The reward mechanism runs through the communication layer, not the operational layer.
This isn't a criticism of regulators exactly: disclosure requirements are easier to enforce than operational requirements, and they're a reasonable starting point for accountability. But the limitations deserve naming. A disclosure regime that doesn't distinguish between Interface's operational transformation and a company that has purchased carbon offsets and improved its data infrastructure will generate better reports from both companies without closing the gap between them.
The EU taxonomy attempts to break this pattern by making classification criteria operational instead of disclosure-based. Whether it succeeds depends on enforcement quality and the integrity of the technical screening criteria over time. I'm cautiously optimistic about the direction but uncertain about the execution.
What I'm more confident about: AI tools applied within a disclosure-reward regime will primarily improve disclosure. AI tools applied within an operational-redesign regime could do something more interesting. The constraint isn't the technology. It's the incentive structure the technology is being asked to serve.
The Honest Version of What AI Can Do
I want to be fair to what AI is actually achieving here, because I've been critical and the picture isn't entirely negative.
Scope 3 emissions tracking (the hardest category, covering indirect emissions from supply chains and product use) has always been difficult to measure with any accuracy. AI tools that can ingest supplier data, logistics records, and product lifecycle information to construct better scope 3 estimates are doing real work. Better measurement is a prerequisite for structural change, even if it isn't sufficient on its own.
Anomaly detection matters too. AI systems that flag inconsistencies between reported emissions and actual energy consumption data, or between supplier certifications and actual practice, can surface the gap between performance and disclosure in ways that manual auditing often misses. That's a genuine contribution to accountability.
There's also a version of predictive modeling, still early but real, where AI helps companies understand the financial implications of different sustainability targets before committing to them. If you can model the P&L impact of a genuine zero-waste commitment before making it, you're better positioned to make the commitment structural instead of aspirational. That's closer to the design application I described earlier.
None of this is trivial. But I keep returning to the Sustainability Removal Test as a way of calibrating how much weight to give these tools. If an AI-generated sustainability report can be removed from a company's operations without affecting those operations, the report is still performative: regardless of how good the AI is.
What Existential Design Requires That AI Can't Provide
Kovnick's Existential Sustainability: A Structural Approach to Anti-Extractive Tourism makes an argument I find increasingly relevant outside tourism as AI reshapes sustainability infrastructure across industries.
His argument is that sustainability requires structural commitment, decisions about what a business is and how it operates, that precede and constrain every subsequent operational choice. The capacity limits, the local retention requirements, the partner relationships that have held for 20 years constitute the model itself. Remove them and there's no business.
AI can help document that structure. It can verify that the structure is performing as designed. In more sophisticated applications, it can help model what structural changes would produce outcomes.
But the decision to build a business structurally instead of cosmetically, to make sustainability the architecture instead of the marketing layer, is a human decision. It requires the kind of long-range thinking that Anderson embodied when he committed Interface to a 26-year transformation in 1994. It means accepting constraints that limit short-term revenue in exchange for structural integrity that compounds over time. The 31% guest return rate documented in Kovnick's paper doesn't appear in year one. It appears after years of structural consistency.
AI doesn't produce that kind of commitment. And as AI makes performative sustainability more sophisticated and more defensible, the pressure to make structural commitments may actually decrease. Why redesign your business when your AI-generated report scores well enough to satisfy investors and regulators?
That's the trap I worry about most.
An Observation About What Gets Measured
Something in the Interface story keeps catching my attention every time I return to it. The company found, repeatedly, that sustainability investments generated financial returns: through waste reduction, energy savings, better product performance, and customer loyalty. Sustainability was a source of competitive advantage.
Kovnick's data shows something similar in a completely different industry context. Eighteen percent net margins while operating at strict capacity limits and retaining 72% of revenue locally. The structural design produced both financial performance and sustainability at once.
The implication is that the assumption underlying most ESG frameworks, that sustainability is a cost to be managed and reported against, may itself be wrong. Or at least, it may be wrong for businesses that have made sustainability structural. For those businesses, sustainability is the source of durable competitive advantage.
AI reporting tools, operating within frameworks that treat sustainability as a cost to be disclosed and offset, will tend to reinforce that assumption. They're built to measure compliance with sustainability requirements, not to reveal the structural properties that make sustainability financially generative.
I suspect the businesses that look most interesting in 20 years will be the ones that figured out how to make the Sustainability Removal Test produce a failing grade. The ones where removing the sustainability means removing the business.
Where This Leaves Us
The question I opened with was whether AI solves the greenwashing problem or produces more sophisticated versions of it. Honestly? It depends on how it's deployed, and most current deployment is in the reporting layer instead of the design layer.
That's not a reason to dismiss AI's contribution to sustainability. Better measurement matters. More consistent disclosure matters. Scope 3 tracking that was previously impossible is now possible. These are improvements.
But the distinction between performative and existential sustainability, what Interface demonstrated in manufacturing and what the tourism operation in Kovnick's paper demonstrates in its sector, doesn't close because reporting improves. It closes because businesses make structural decisions to embed sustainability in what they are, not just in what they say.
AI can help with that second kind of work. It isn't currently being asked to do it at scale.
Watch the companies using AI not to produce better sustainability reports but to ask harder questions about their own architecture. Where are we structurally dependent on extractive practices? What would it cost to change that? What would a business look like where the Sustainability Removal Test fails every time?
Those questions don't have easy answers. They probably don't have AI-generated answers, at least not yet. But they're the right questions: and the fact that AI is making it easier to avoid asking them, by making performative sustainability more convincing, seems worth naming clearly.
The report isn't the thing. The structure is the thing.
Comments (0)
No comments yet. Be the first!