Can AI Help You Avoid Greenwashing, or Just Do It Faster?

In June 2026, researchers at the University of Auckland published the results of an experiment that should give anyone using AI for sustainability content a moment's pause. They asked postgraduate students to produce corporate sustainability reports using generative AI, mixed those reports in with real ones, and had people assess them blind. The AI-generated reports were rated more credible than the real thing. They also scored high on established greenwashing metrics. Both results at once. More convincing, and less honest.

That is the AI greenwashing problem in a single experiment. The tools most businesses now use to draft their sustainability content are very good at producing exactly the kind of language regulators are moving against, and very good at making it sound trustworthy.

So the question in the title is worth taking seriously. Can AI help you avoid greenwashing? Yes, it can. But not the way most businesses are currently using it.

The Default Setting of Generative AI Is Greenwash

Large language models learn from what has already been written. And the average sustainability sentence on the internet is not a carefully evidenced claim. It is marketing copy. Committed to a greener future. Passionate about the planet. Driving positive change. The model has read millions of these sentences, and when you ask it to write about your environmental efforts, that register is what comes out.

This is not a bug you can prompt your way around with one clever instruction. It is the statistical centre of the training data. The Auckland study, published in Business Strategy and the Environment, found that when participants used ChatGPT to write sustainability reports, the output glossed over difficult issues and ranked high on greenwashing measures, and that participants did little to correct it. The polish did the persuading. Nobody felt the need to check underneath.

Think about what that means for a small business without a sustainability team. You ask an AI tool to write your product page. It produces fluent, confident, plausible copy. Nothing about it looks wrong. And unless you already know what a compliant environmental claim looks like, you have no reason to question it. The tool has not lied to you exactly. It has done something more subtle. It has written a claim that no one in your business ever decided to make.

The Claim Nobody Decided to Make

That last point deserves its own section, because there are actually two different failure modes here, and the second is worse than the first.

The first is vagueness. The AI writes "eco-friendly packaging" when your packaging is 60% recycled cardboard. Annoying, risky under the new rules, but at least tethered to something real. You can fix vagueness by adding the specifics.

The second is fabrication. Ask a general-purpose AI to make your copy "more credible" and watch what happens. It adds credibility markers, because that is what credible text looks like in its training data. A percentage here. A certification there. "Certified carbon neutral." "Made with 100% organic cotton." If those details are not in what you gave it, it will supply them anyway, and they will be invented. Not maliciously. Statistically. Credible sentences contain numbers, so the model adds numbers.

For an SME, this is the nightmare scenario, because a fabricated specific is more dangerous than a vague generality. A regulator reading "we care about the planet" sees puffery. A regulator reading "certified by the Soil Association" checks the register. When the certification does not exist, you have moved from a questionable claim to a demonstrably false one, and you may never have noticed the sentence going in. The Auckland researchers found participants did little to correct AI output. That is the pattern to fear: not bad intent, but unexamined fluency.

AI Greenwashing Meets the New Rules

The timing of all this could hardly be worse, because the regulatory direction of travel is the exact opposite of the AI default.

From 27 September 2026, the EU's Empowering Consumers Directive bans generic environmental claims such as "eco-friendly", "green" and "climate neutral" unless they can be backed by recognised excellent environmental performance. That is precisely the vocabulary generative AI reaches for first. In the UK, the CMA has held direct fining powers since April 2025 under the Digital Markets, Competition and Consumers Act, with penalties of up to 10% of global turnover for misleading claims.

And here is the part that closes the loop. The regulators are using AI too. The ASA's Active Ad Monitoring system scanned nearly 60 million online ads during 2025, with green claims named as a priority area. Its recent rulings against recycled-content claims from Adidas, Calvin Klein and Uniqlo came through AI-assisted monitoring, not consumer complaints. Ads from Nike, Superdry and Lacoste have been flagged the same way.

So the machine writes the claim, and another machine reads it. The unqualified "sustainable" that an AI tool generated in three seconds gets picked up by a regulator's model trained to find unqualified "sustainable". There is something almost comic about it, except that the enforcement notice arrives with your company's name on it, not the software's.

The Speed Problem Nobody Prices In

There is a second problem, and it has nothing to do with the quality of any single output. It is volume.

Before AI, writing marketing copy took time, and that time was an accidental control. When a product description took someone an hour, that person read it. Claims passed through a human brain on their way out the door. Slowly, imperfectly, but they passed through.

AI removes that friction. A business can now generate forty product descriptions in the time it used to take to write one. If the tool's default register leans green, and nobody is checking each output against evidence, you have not made one risky claim. You have made forty, published across your website, your marketplace listings and your social channels, each one indexed, each one live, each one attributable to you.

The irony is that this does not even work as marketing. Research on consumer responses to AI-generated green content, published in the Journal of Retailing and Consumer Services, found that moderate environmental messaging improved purchase intention, but excessive green content triggered greenwashing scepticism and pushed customers away. Piling on green language is not just a compliance risk. Audiences have learned to read it as a warning sign.

Where AI Genuinely Helps

None of this means the technology is the problem. It means the direction of use is the problem.

Used as an unsupervised writer, AI generates claims. Used as an interrogator, it tests them. And it happens to be genuinely good at the second job. The same pattern-matching that produces greenwash by default is remarkably effective at spotting it when you point it the other way.

In practice, that looks like this. You give the tool your draft claim and the evidence behind it, and ask where the gap is. You ask it what a regulator applying the CMA's Green Claims Code would challenge. You ask it to list every absolute term in your copy, every "100%", every "fully", every standalone "sustainable", and to tell you which ones your evidence actually supports. You ask it whether the claim covers the full life cycle of the product or quietly stops at the factory gate.

These are questions most SMEs never get asked, because the people who ask them professionally cost more per hour than most SMEs can spend on a product description. AI makes the interrogation affordable. That is the genuine opportunity, and it is the opposite of the way the tools are mostly being used.

Here is what the difference looks like in practice. A generated claim: "Our eco-friendly water bottles are made from sustainable materials and help protect the planet." Three claims, zero evidence, all three now problematic in the EU and challengeable in the UK. The interrogated version starts from the evidence instead: the bottle is 75% recycled aluminium, verified by the supplier's documentation, and the lid is not recycled at all. So the claim becomes: "Bottle made with 75% recycled aluminium. Lid made from new plastic. We're working on that." Less impressive-sounding. Far more credible, and audiences increasingly know the difference. The second version survives a regulator, a journalist and a sceptical customer. The first survives none of them.

The test is simple. If AI is introducing claims into your content, it is working against you. If it is challenging claims you were already making, it is working for you.

Five Rules for Using AI on Sustainability Copy

If you use AI anywhere near your environmental claims, these five rules are the practical version of everything above.

Never let AI introduce a claim. Facts flow one way: from your evidence into the draft. If a claim appears in the output that you did not put in, it comes out. No exceptions, however good it sounds. Especially if it sounds good.

Ban the default vocabulary. Tell the tool explicitly not to use "eco-friendly", "green", "sustainable" or "planet-friendly" as standalone descriptions. These are the words the EU is banning and the ASA keeps ruling against, and they are the words AI reaches for first.

Give it the evidence, not the vibe. "Write something about our sustainability" invites invention. "We switched to 80% recycled packaging in March 2026, verified by our supplier's certificate, here it is" gives the tool something true to work with. The quality of the input decides whether the output is a claim or a story.

Use it as the antagonist. Before anything goes live, run the reverse prompt: "You are a regulator reviewing this claim under the Green Claims Code. What would you challenge?" It is the cheapest pre-publication check available to any business, and almost nobody does it.

Keep a named human sign-off. Someone in your business, with a name, approves every environmental claim before it publishes. Not because AI is untrustworthy in some abstract sense, but because when the ASA writes to you, "the AI did it" is not a defence. Liability does not transfer to the software.

The Honest Answer

So, can AI help you avoid greenwashing or just do it faster? Both. It will do whichever one you set it up to do, and the default is the wrong one.

I should be straight about where I stand here. My Green Comms is itself an AI tool. I built it because I watched general-purpose AI produce confident green copy that would not survive contact with a regulator, and I thought the same technology could be pointed the other way: at checking claims against evidence and current rules before they go out, rather than generating them from thin air. The technology was never the problem. The absence of discipline around it is.

My Green Comms helps SMEs communicate sustainability clearly, credibly, and in a way that stands up to scrutiny. If you'd like to see what that looks like in practice, start free at app.mygreencomms.com.

Download our free 10-point claim check.

This article is for informational purposes only and does not constitute legal advice. If you have concerns about your legal accountability for environmental claims, speak to a qualified solicitor.