Why AI shouldn't write your sustainability report

There's a version of this article where I sound defensive. A writer warning you off AI-generated reports.

Of course, I would say that. I’m a writer.

But, honestly, I’m not anti-AI. I see its benefits (environmental impacts aside, which is a separate conversation people are still skimming past). Grammarly has saved my bacon more than once when I was bleary-eyed, and if you haven’t downloaded Merlin and taken it for a walk in a forest, go do that first.

You can thank me later.

I’m also realistic about where this is going. AI isn’t going away. It’s getting better. And most people working in sustainability reporting are already experimenting with it, whether openly or behind the scenes.

I should also preface this by saying that I don’t use AI to write any of my reports. In fact, most of my NDAs prohibit me from using such tools in the first place.

But sustainability report writers don’t “just write”. From CSRD to AI, I spend a lot of time watching how things intersect, where risks emerge, and how small wording choices can turn into big problems six months down the line when internal audit starts asking questions.

I don’t move fast on opinions.

(For context, the most impulsive thing I’ve done this decade was getting bangs for the first time in 25 years.)

All of that to say: I’ve spent a lot of time thinking about this, and here’s where I’ve landed, for now:

For many companies, your sustainability report is a regulated document that will be reviewed by external assurers, scrutinized by investors, and read by people trained to find gaps.

The wording is evidence, not decoration.

And AI is very good at producing something that looks like evidence. It is not good at knowing whether it is.

(Side note: Have you ever asked AI a question and its reply was “I don’t know?” In most cases, it would rather make a stab in the dark than admit failure.)

In a document where the difference between sounding right and being right can constitute a material misstatement, that's a liability.

Here's what I see in AI-drafted reports: Sentences that are technically defensible but don’t actually say anything. Claims that feel specific until you try to trace them back to a number. Language that signals ambition without committing to a measurable outcome.

It reads well.

It doesn’t survive scrutiny.

AI also doesn’t know what to leave out. It doesn’t know when a caveat quietly undermines the claim it’s attached to. It doesn’t know that your 2025 data gap is more credible than a polished 2026 projection you can’t support yet.

It doesn’t know that one word—“committed,” “aiming,” “exploring”—can change how an entire paragraph is interpreted.

You do. Or you should. That's the point of the report.

And this is where the idea that writers are “just writers” starts to fall apart. In this context, the writing process is where risks are surfaced (or buried).

A good sustainability writer is part wordsmith, part risk translator, part investigator. We’re listening for what’s missing as much as what’s there. We’re stress-testing language before anyone else does. We’re asking questions your data can’t answer yet. We’re flagging where a claim could be misinterpreted. We’re shaping a narrative that is both clear and defensible.

That work shows up as the absence of problems later.

The efficiency argument doesn't hold up either.

Yes, AI can produce a draft quickly.

But then someone has to verify every claim, strip out generic framework language, reinsert company-specific context, align it with legal, and fix the tone so it sounds like your company instead of a composite of everyone else’s.

And now they’re doing it without the context of having written it.

That time doesn't disappear. It just shifts to someone who's probably less qualified to catch the problems than the person who wrote it.

(Side note: I edited an AI-written report last year. The initial draft was fast. The cleanup was, um, not.)

There’s also an accountability problem people don’t talk about enough.

When something in your report gets challenged, you can’t point to the tool. I mean, you can, but good luck. Someone has to stand behind those words. Someone has to explain why that claim was made, why that wording was chosen, and why that omission was acceptable.

“AI wrote it” is not a defense.

There's a subtler issue, too.

Readers can feel the difference between a report that was generated and one that was written. We’ve all become attuned to what AI slop reads like. At this point, I can spot it a mile off.

A genuinely written report signals that leadership took disclosure seriously. That's not a soft benefit. In a regulatory environment that is actively tightening, it's the whole point.

I’m not arguing that AI has no place in the reporting process. It does. Use it to scan regulation updates. Summarize source material. Pressure-test your structure.

Take it for a walk in the woods, even. (Send me what you hear afterward.)

But writing the report? That's a disclosure decision. And disclosure decisions should be made by people who understand what's at stake and are prepared to be accountable for the result.

If that sounds like something worth getting right, I'd be glad to help