Short answer: AI is a powerful tool for checking ESG claims against evidence — but it is also capable of generating confident-sounding claims that have no evidentiary basis. The honest framing for AI in sustainability communications is verification-first, generation-second.
The Greenwashing Landscape Is Hardening
Sustainability claims in the food and beverage sector are under greater regulatory and reputational scrutiny than at any prior point. The EU Green Claims Directive, once enacted, will require that environmental claims be substantiated before they are made public — not after a challenge. The UK Competition and Markets Authority has already taken action against misleading green claims in consumer markets. The US Federal Trade Commission’s Green Guides, long a reference point for substantiation standards, are under revision.
For breweries, the claim categories most exposed are familiar: “carbon neutral,” “sustainably sourced,” “eco-friendly packaging,” and “water responsible.” Each can be a defensible claim when anchored to documented evidence. Each becomes a greenwashing exposure when it relies on aspirational language, unverified supplier assertions, or — increasingly — AI-generated content that sounds authoritative but is not traceable to real data.
AI’s Hallucination Problem Applied to ESG
Large language models generate text by predicting plausible continuations. When asked to produce an ESG report section, a blog post, or marketing copy about sustainability, they will produce fluent prose that includes statistics, study references, and comparative claims. The problem is that these outputs are generated, not retrieved — the model has no connection to the brewery’s actual measurement data, no access to the specific lifecycle study it may appear to be citing, and no mechanism for flagging when it is extrapolating beyond available evidence.
A sustainability professional reviewing AI-generated content for fluency and tone will often not catch a fabricated figure or a study reference that does not exist — particularly when the figure is plausible in magnitude. This is not a hypothetical failure mode: AI “hallucination” in professional documents is a documented phenomenon, and ESG communications are exactly the kind of domain where confident-sounding prose is produced routinely and fact-checked insufficiently.
The risk is compounded by the verification gap: many sustainability claims are made in contexts (marketing materials, investor presentations, CSR reports) where the drafting team is not the same as the data team, and where the evidentiary chain between the disclosed number and the source measurement is not systematically checked.
Using AI Constructively: The Verification Role
The appropriate role for AI in ESG communications is not zero — it is redirected. The tasks where AI adds genuine value without creating greenwashing risk are analytical and checking functions, not generation functions:
Consistency checking: AI can parse multiple documents — prior ESG reports, current draft disclosures, underlying data tables — and flag where figures are inconsistent, where methodologies have changed without disclosure, or where claims in one document are not supported by data in another.
Regulatory alignment testing: Given a proposed claim (“our packaging is recyclable”), an AI tool with access to the relevant regulatory guidance (EU Green Claims Directive, FTC Green Guides) can assess whether the claim as stated would meet stated substantiation requirements — not to approve the claim, but to surface the gaps before publication.
Data extraction from source documents: AI can parse supplier sustainability certificates, utility bills, and logistics invoices to extract structured data that feeds into an ESG platform — reducing the manual effort of data collection without generating claims from thin air.
Comparative language review: AI can flag superlatives, unqualified comparatives (“lower carbon”), and category-wide claims in draft communications and prompt the drafter to identify the evidence base for each flagged phrase.
What AI should not be asked to do unsupervised: draft quantitative ESG claims, generate specific emission figures or reductions, or produce study citations — any of which may be hallucinated and which could appear in disclosures that carry legal and reputational consequences.
A Practical Protocol for Breweries
The operating model that reduces greenwashing risk in an AI-assisted ESG function looks like this: all quantitative claims trace to a named source document and a specific calculation methodology; AI tools are used to check that trace rather than to generate it; all external disclosures are reviewed by someone who has access to the underlying data and not just the drafted text.
This is not a counsel of excessive caution about AI — it is a recognition that the tool’s strength (fluent, fast, broad) is not the same as the task’s requirement (accurate, traceable, defensible). See also the broader honest-limits framing at /2026/the-honest-limits-of-ai-in-brewing/.
Honest caveat: This article itself was written with AI assistance. Every specific claim about regulations, risk categories, and AI behavior has been reviewed against the author’s working knowledge and flagged for verification — but readers should independently confirm the current status of specific regulatory instruments (Green Claims Directive, FTC Green Guides) before relying on them for compliance decisions.
Part of the ESG track — browse all.
Frequently asked questions
What is greenwashing and why is it an increasing regulatory risk for breweries? Greenwashing is making environmental or sustainability claims that are misleading, unsubstantiated, or materially incomplete. For breweries, the risk categories include unverified carbon-neutral claims, packaging sustainability assertions without lifecycle evidence, and sourcing narratives that cannot be traced to documented supplier data. Regulatory scrutiny is increasing: the EU Green Claims Directive requires pre-substantiation of environmental claims before public disclosure, with enforcement teeth that include fines and mandatory corrective advertising.
Can AI tools generate greenwashing even when that is not the intent? Yes — and this is one of the most underappreciated risks in AI-assisted sustainability communications. Large language models are trained to produce fluent, confident-sounding text. When asked to draft ESG content, they will produce plausible-sounding statistics, studies, and comparisons that may not correspond to real evidence. A sustainability report or marketing claim drafted with AI assistance and not verified against source data can contain fabrications that pass a casual editorial review. The risk is not malicious; it is structural to how current LLMs work.
How can AI be used constructively in ESG verification rather than just content generation? AI’s verification value lies in structured consistency checking, not in generating new claims. Useful applications include: cross-checking disclosed figures against source data in connected systems; flagging discrepancies between current disclosures and prior-year reports; identifying claim language that lacks a documented data source; and parsing regulatory guidance to test whether a proposed claim would meet substantiation standards. These are analytical tasks where AI augments human review rather than replacing the need for primary evidence.