Evergreen Assessment

Yes, absolutely. There are some very useful jobs I would happily give it.

But completing your NHS Evergreen Assessment on autopilot? That is where I would draw the line.

Used sensibly, a generative AI tool can save you some time. It can help you organise existing information, summarise documents, identify inconsistencies, turn rough notes into a first draft and build a list of questions for colleagues. Those are useful jobs, particularly when responsibility for Evergreen has landed with somebody who also has plenty of other things going on.

The problem begins when AI is treated as the source of the answer rather than a tool for working with information that the business already owns. And I am seeing this regularly now.

AI can give you an answer that can sound highly convincing. But it can still be wrong, out of date or unsupported.

A polished paragraph is not evidence.

I have seen beautifully written policies produced by AI just minutes before a conversation. Every sentence was polished, but the business had not yet understood or implemented the policy. That is where I pause. A draft can be a useful starting point; it still needs to reflect what the business actually does.

My working rule is simple: AI can be a useful assistant, but it cannot be the witness. Your organisation still has to know whether the answer is true and be able to show the evidence behind it.

What the Evergreen Assessment is actually asking you to do

NHS England describes Evergreen as a self-assessment and reporting tool through which suppliers share sustainability information with the NHS and receive a maturity score. It is not an exercise in who can make the fanciest documents. It is a structured account of where you are now and how you claim to align with NHS sustainability priorities.

At the time of writing, the Evergreen guidance does not set out a separate rule for AI-assisted drafting. However, using AI does not change the underlying job. The supplier remains responsible for the information submitted, the evidence selected and the commitments described.

If the assessment says that a policy is approved, a target has been adopted, emissions have been calculated, or that an action has been implemented, somebody in the company should be able to verify that claim. AI cannot supply that accountability for you.

Where AI can genuinely help with Evergreen

AI can genuinely help with several parts of Evergreen preparation, provided it works from reliable source material and a human reviews the result.

1. Understanding the question

Some Evergreen questions use sustainability, procurement or reporting language that may be wholly unfamiliar to the person completing the assessment. AI can translate the wording into plain English, explain key terms and produce a list of information that may be needed.

Treat that explanation as orientation, not as the final interpretation of the NHS requirement. Check it against the current Evergreen guidance and the wording shown in Atamis.

2. Organising existing evidence

AI can help classify documents and map them against requirements. For example, it can create a draft register showing which questions the Carbon Reduction Plan, environmental policy, modern slavery statement, social value records, training logs, or emissions calculations may support.

This is most useful when you instruct your AI tool to work only from the documents supplied and to identify gaps rather than filling them with invented content.

3. Finding inconsistencies

AI can compare documents and flag differences in emissions figures, reporting years, target dates, company names, review dates or net zero commitments. That can be a valuable early check before an assessment or tender submission.

A competent person still needs to decide which figure or statement is correct. The tool can point to the discrepancy, but it cannot resolve an accounting boundary or governance decision by guessing on your behalf.

4. Creating a first draft

If the business has supplied accurate notes and evidence, AI can help turn them into a clearer first draft. This may help when the subject owner knows exactly what’s going on but doesn’t enjoy writing these formal tender responses.

The final response should always still be edited into your own language. Generic phrases such as ‘we are deeply committed to sustainability’ add little unless they connect to a specific action, owner, target, or result. Without support, those sentences may be considered greenwash.

5. Building a gap and action list

AI can turn identified weaknesses into a structured list of questions and actions: what is missing, who needs to answer, what document needs approval and when the work is due. Used this way, it supports the evidence process rather than pretending the gaps do not exist. It can help you build that Evidence Register.

What AI cannot safely do for you

AI can support summarising supplied documents, drafting wording from approved facts, flagging missing evidence, comparing figures and dates, and suggesting a response structure. A human must still own confirming that the summary is accurate and current, approving the claim and accepting accountability for it, creating and implementing what is missing, resolving the correct boundary, methodology and final figure, and selecting the maturity claim that the evidence genuinely supports.

AI should not be asked to invent evidence, calculate a carbon footprint without controlled data and methodology, claim that an action has taken place, select the most flattering maturity level or produce a policy that nobody in the business has reviewed.

It also cannot sign off the Carbon Reduction Plan, approve a policy, accept a target on behalf of directors or confirm that a response meets a specific tender condition. Those are governance decisions, not writing tasks.

The main risks of using AI for an Evergreen submission

Plausible but incorrect answers. AI tools can produce confident statements that are factually wrong, sometimes described as hallucinations.
Generic evidence. A polished policy may say everything expected of a sustainability policy while describing very little about the organisation that is supposed to use it.
Outdated requirements. Evergreen, Carbon Reduction Plan rules and procurement policy change. A tool may rely on old material unless it is given and instructed to use the current source.
Contradictions. AI-generated wording may introduce a new date, boundary or commitment that conflicts with the Carbon Reduction Plan, website or previous tender responses.
Loss of traceability. If nobody records the source used for an answer, it becomes difficult to explain where the statement came from or update it next year.
Confidentiality and data protection. Staff may upload commercially sensitive, personal or client information into a tool without understanding how that information will be processed, retained or accessed.
Overclaiming. AI tends to produce complete-sounding answers. It may smooth over uncertainty and turn an ambition into something that reads like an established practice. The National Cyber Security Centre warns that large language models can produce incorrect statements as facts. The Information Commissioner’s Office also makes clear that where AI use involves personal data, data protection requirements still apply. These are not reasons to avoid AI altogether. They are reasons to use it with basic controls.

I would not work on the assumption that nobody will notice generic AI wording. But I have not seen anything in the published Evergreen guidance saying submissions are screened for AI-generated text.

One pharmaceutical company gave me an AI-written draft of its Modern Slavery Statement. It took three meetings and several rewrites to make sure it reflected the business and was understood by the people responsible for it. That work mattered. The final version avoided overclaiming and set out a clear path for improvement.

Do not upload everything into a public AI tool

Before uploading policies, spreadsheets, tender documents, or staff information, check which AI tools your organisation permits and what data you may enter. The answer depends on the tool, account settings, contract, and internal policy.

At minimum, avoid entering personal data, client-confidential information, commercially sensitive prices, unpublished tender material, security information or detailed employee records unless the organisation has assessed and approved that use.

Where possible, remove names and unnecessary detail, use an approved business tool, limit the material to what is needed for the task and retain the source documents in the organisation’s controlled evidence system.

A closed or enterprise AI system may offer stronger controls, but that does not make every use safe. Check the tool’s configuration, contract and your own internal policy before uploading sensitive information.

Prompts that are safer and more useful

A good prompt limits the job and makes gaps visible. For example:

“Using only the documents supplied, create a table showing which source may support each Evergreen question. Do not draft evidence that is not present.”
“Compare these documents and list every inconsistency in dates, targets, emissions figures, reporting periods and company names. Quote the source location for each discrepancy.”
“Based on this requirement and the supplied evidence, list the questions I need to ask finance, HR, procurement and the managing director. Do not answer the questions yourself.”

A less useful prompt is: ‘Complete my Evergreen Assessment.’ It gives the tool no reliable boundary, no controlled sources and no way to distinguish fact from fiction.

Should you disclose that AI was used?

The public Evergreen guidance does not currently describe a separate disclosure requirement for AI-assisted drafting. That does not remove the need for transparency inside the organisation. It should be clear who prepared the response, who checked it and who approved the final submission.

If AI has materially shaped a calculation, analysis or decision rather than simply helping with wording and filing organisation, there may need to be more formal review. Follow the organisation’s AI, information security, data protection and quality procedures.

AI can speed up the work, but it cannot create maturity

AI is useful when you already have knowledge and evidence that needs organising. It is much less useful when the real problem is that nobody owns the policy, the emissions calculation is inconsistent, the target has not been approved, or the actions haven’t proceeded.

In those cases, better wording alone will not solve the problem. You need to close the gap. That may involve collecting data, agreeing a boundary, training staff, changing a process or recording evidence of implementation.

AI should not make the business appear more mature than it is.

Need a human check before you submit?

If AI has helped you prepare an Evergreen response but you are unsure whether the evidence genuinely supports it, Evergreen Assessment can provide an independent review before submission.

The Pre-Submission Check is suitable where the assessment and evidence are substantially complete, and you want a focused review. If you have received No Level Awarded or the supporting documents need more substantial work, the Evergreen Assessment and Evidence Review or Evidence Pack may be more appropriate.

You can explore the support options or contact me to discuss the most proportionate next step. If you do not need paid support yet, I will tell you that too.

Can You Use AI to Complete the NHS Evergreen Assessment?