Using AI for sustainability work: what it's good for and when you need a specialist

LLM tools can save you hours on sustainability questionnaires and tender responses. They can also hand a buyer a polished answer that scores nothing. This guide covers how to get the first without the second.

Most people answering a sustainability question at work now have an LLM open - ChatGPT, Claude, Gemini etc - in another tab. That's not a bad thing. If you've been handed an EcoVadis request or a tender with a carbon section and nobody in the business owns sustainability, asking AI is a sensible first move.

Lauren Rabbitte of Go Green Now made the point when she interviewed our co-founder, Jon Wragg, in August. People can now ask ChatGPT for the information, which "almost removes that entire department". What they still lack, she said, is "the information on where to start or how to use the information to make it useful".

That gap is what this piece is about. We use AI tools ourselves, and we also read a lot of tender responses and supplier submissions. The ones written by AI with nothing behind them are easy to spot. Below we go through where AI helps and where it goes wrong, then how to get better output from it and when to bring in a specialist.

Now, we'll caveat all of this before going any further. We don't think sensitive or non-public information should go into an AI tool, whether that's a customer's tender pack or your own commercial data. Jon's first piece of advice is not to use AI at all unless you understand how it works and know the tool you're using is secure. That means knowing what happens to anything you upload, including whether it's stored or used to train the model. You also need an AI policy that says what can and can't go in, and someone reviewing how people across the business are using these tools. If all of that is in place, AI can help with much of what follows. If it isn't, that's the place to start.

Where AI helps

Getting your bearings

If you've never come across Scope 3, or you want to know what EcoVadis assesses before you pay for it, AI is a decent place to start. It will give you a plain explanation in seconds and you can keep asking until it makes sense. Plenty of people now find specialists that way too. One of our recent enquiries came from someone who asked ChatGPT what EcoVadis was, then asked it who could help (yes, we're so good at what we do that AI is citing us as the experts!).

Treat what it tells you as a starting point. The rules in this area change often, and we'll come back to why that matters.

Reading long documents

Tender packs are long, and the sustainability questions aren't always where you'd expect them. In Jon's experience they turn up in different places: "some will be in quality, some will be scored as sustainability, some will come as part of operations".

AI is good at going through a long pack and listing every question that touches carbon, energy, waste, social value or ethics, along with the weighting and scoring criteria for each. You still need to check that list against the original. It's a much quicker start than going through the whole pack with a highlighter.

Tidying your own words

Jon is blunt about where AI belongs in writing: "Claude and ChatGPT are great at tidying up and helping you reach word count, but they are not great at spewing out the correct answers, especially scorable answers."

The first half of that is a real strength. If you've written an answer that's accurate but rambling, AI can tighten it, bring it under a word limit and make the structure easier to follow. The substance has to come from you.

Organising your evidence

Buyers increasingly want proof attached to each answer, and they won't go hunting for it. As Jon says, "no scorer is ultimately going to read a 50, 60 page document." The strongest responses tell the scorer which page and section of a report backs up each claim. AI can help you build that index from your own documents, so every answer points straight to its evidence.

The technology that isn't a chatbot

A lot of the most useful technology in sustainability has nothing to do with LLMs. Carbon accounting platforms do the maths with the right conversion factors and keep your figures consistent from one year to the next. On the buyer side, supply chain tools show large businesses which of their suppliers are high, medium or low risk. Our partner tools show buyers where the suppliers without an EcoVadis rating sit.

For many smaller businesses the alternative is a spreadsheet kept by someone for whom sustainability is a third job. Moving that onto a proper platform will do more for the quality of your answers than any LLM.

Buyers use it too

Procurement teams are using AI to read what you send them. Jon has seen it used "as part of the scoring matrix" when a buyer is dealing with a lot of suppliers, and says it can "cut through a lot of the fluff". That works in your favour if your answer is specific. It works against you if it's padding.

UK public sector buyers now have guidance on this. PPN 017, which covers procurements starting on or after 24 February 2025, says suppliers' use of AI "is not prohibited", but that buyers should understand the risks in the same way they would if a bid writer had been used. It allows them to add questions to a tender asking whether and how you used AI, so it's worth assuming the question may come up.

Where it goes wrong

It doesn't know your business

This is the big one. A good sustainability answer doesn't stop at what you have. It goes on to say what that has changed in your business and what it means for the buyer. Jon calls that last step the "which means what", and it's the part AI can't write.

"Because it doesn't know your business inside and out, it cannot give you the which means what," he says. "And that's where procurement has changed now."

An accreditation such as ISO 14001 or an EcoVadis rating used to be enough to tick the box. Now most of your competitors have one, so buyers treat it as the starting line and score the evidence behind it. Jon gives the example of a question about how you're reducing energy, answered with a paragraph about donations to a tree planting scheme. "It's a great story, but it's not scorable." The buyer wants to know what you've done in your own operations, whether that's switching energy supplier, installing solar or finding easy wins through carbon accounting, and what each of those achieved.

AI has no idea what your electricity use was last year or what you've changed since. It fills the gap with general statements that sound right and say very little.

It can be confidently out of date

Jon's warning here is specific. "AI is terrible at pulling in buzzwords from other sources that aren't legit," he says, and at "trying to link you to legislations that either have changed or don't exist or are outdated."

Two current examples show how easily this happens. Carbon reduction plans for central government contracts now come under PPN 006, which replaced PPN 06/21 for procurements starting from 24 February 2025, yet plenty of what's online still refers to the old note. And in the EU, the Omnibus changes that came into force in March 2026 narrowed the Corporate Sustainability Reporting Directive to companies with more than 1,000 employees and more than €450m turnover. A tool that learnt from older material may still describe the original thresholds, which reached companies with as few as 250 staff.

Neither mistake looks like a mistake. The wrong version reads just as fluently as the right one, and that's the danger.

It reaches for the wrong words

Most greenwashing isn't deliberate. Jon's view is that a lot of it is accidental: "They have something good and they want to shout about it or talk about it and they're just doing it in the wrong way because they're using the wrong buzzwords. AI is terrible for that."

Terms such as carbon neutral, eco-friendly and net zero carry specific meanings, and a claim you can't support is a risk whether a person or a tool wrote it. If a sentence makes a claim, you need the proof to hand. Organisations such as the Anti-Greenwash Charter, which we've signed, specialise in how businesses communicate green claims and can review how you talk about your work.

It can't produce evidence

What buyers ask hasn't changed as much as how they ask it. In one tender Jon saw, the questions were familiar ones about carbon footprint and energy management, but under pretty much every question the buyer now asked "can you prove it?" They wanted certificates, case studies and a contact who could vouch for past work, right down to the site's energy use from 18 months earlier set against today's. The supplier could answer because they had a carbon report showing the same figures year on year.

No AI tool can produce that for you. It can only work with what you give it, and if it starts filling in numbers you didn't supply, you have a far bigger problem than a weak answer. We cover what a report like that needs in our guide to carbon footprint reports.

People can tell

"I see a lot of tender responses and you can spot an AI written one a mile off," Jon says. The giveaways are familiar. There's plenty of general commitment and very few figures, in language nobody in your business would use. Lauren Rabbitte made a related point in the same session. AI struggles in particular with the social value side of a bid, and the person scoring it is still a person deciding who they want to do business with.

It's easy to paste in the wrong thing

Tender documents often come with confidentiality terms, and your own energy, staffing and supplier data is sensitive. Before anyone pastes either into an AI tool, check what the customer's terms allow and what your own policy says. If you don't have a policy, that's worth fixing.

The same thing happens at board level. Jon held several board seats before setting up Ltt, and remembers how AI arrived: "AI was just flung across everybody's desk. We need AI, but it's not a strategy." His view is that the strategy and the policy come first, with proper governance behind them.

How to make AI useful

1. Read the question yourself first. Jon's first piece of advice to anyone facing sustainability questions is to "read what you are being asked and understand the scoring criteria behind it. Not a lot of people do." AI can help you break a question down, but you should decide what it's asking.
2. Give it your facts, not a blank page. Paste in your figures, policies, certificate dates and what changed, and tell it to use only what you've supplied. If it needs something you haven't given it, it should ask rather than invent.
3. Ask it for the gaps. A useful prompt is: "What evidence would a scorer expect for this question that isn't in what I've given you?" Treat the answer as a to-do list for you rather than as content for the response.
4. Check every name, date and figure. Any regulation, standard or scheme it mentions needs checking against the source, such as GOV.UK or the scheme's own guidance. Look at the date on what you find, too.
5. Take the buzzwords out. If you can't point to the evidence behind a claim, cut the claim.
6. Point to your evidence. Give the page and section of each supporting document so the scorer can find it in seconds.
7. Get the right people to sign it off. Someone who knows the business must read the final version. Jon's advice is to treat a tender as the set of specialisms it is. Each section goes to the team that owns that area, sustainability included, and one person pulls the whole response together. One of the most successful businesses he has worked in handled tenders exactly that way.

When to bring in a specialist

AI will get you further than a blank page. What it can't do is create evidence you don't have, and that's usually the point where specialist help pays for itself.

When a lot rides on the answer

If the sustainability section carries real weight, a strong answer changes the rest of the bid. In the tender mentioned earlier, the sustainability section had gone from 3 to 5% of the marks to 30%. "You might not need to be the cheapest," Jon says. He points out that you could come in a few percentage points more expensive than a competitor and still win if you beat them in that section.

When the evidence doesn't exist yet

If you have no carbon baseline, your policies don't line up with what the buyer is assessing or your management systems aren't written down, that's work to be done rather than words to be written. Better prompts won't fix it.

When the rules are specific

Carbon reduction plans under PPN 006, NHS supplier requirements, EcoVadis methodology and requests driven by your customers' own reporting duties all come with details that change. Getting them wrong costs marks.

When claims are going public

Anything you say publicly or put in front of a customer should be checked by someone who knows what you can back up.

When it's a strategic decision

Decisions such as whether to go for EcoVadis, or which customers to pursue, depend on where the business is heading. A chatbot doesn't know that.

It doesn't have to be a big commitment

Help doesn't have to mean a retainer. "Not every consultant is going to charge you a retainer fee and charge you masses of money," Jon says. Sometimes a short education or training session is enough to understand what's being asked and how it will be scored.

When you do look for help, check that the consultant can help you answer the question as well as produce the evidence. "Not every sustainability consultant out there is commercial," Jon says. "They're really good at getting you to a place and providing you the evidence. But what they're not always good at is telling you how to answer that question correctly."

His closing advice from the Go Green Now session sums it up: "You would ask a specialist in other areas of your business. Sustainability is a specialism. Ask for help."

If you've got a tender, questionnaire or EcoVadis request in front of you and you're not sure what it's asking for, book a call with us.

Watch the full sessions: Sustainability as a growth strategy with Go Green Now, and Strategy in Practice.