An intro to AI and sustainability: practical steps for responsible AI use

Sustainable AI is becoming an increasingly important consideration for organisations exploring artificial intelligence. While AI can improve productivity and support decision-making, it also raises environmental, social and economic questions that organisations should understand before adopting it more widely.

In our recent webinar, we explored AI’s environmental, social and economic considerations, alongside practical steps organisations can take to make more informed decisions about AI use.

What is the relationship between AI and sustainability?

AI is not inherently good or bad. It is a tool, and its sustainability depends on how it is developed, powered, managed and used.

The environmental considerations discussed during the webinar included the hardware needed to operate AI systems, the electricity required by data centres and the water used in some cooling systems. We also explored wider social and economic questions, including changing workplace skills, the role of AI across different industries and the ongoing challenge of measuring productivity gains.

There are still significant evidence gaps. Providers do not always publish enough information to calculate the footprint of an individual AI request confidently, and reporting methods are not yet standardised. Rather than searching for perfect answers, organisations can focus on avoiding unnecessary use, asking better questions and reviewing available evidence.

Start by asking whether AI is needed

Before opening an AI tool, define the task and consider whether another approach would be simpler.

A conventional online search, a calculation, a manual folder search or an existing template may sometimes accomplish the task without AI. In other cases, AI may offer genuine value by helping to organise information, identify themes or create a useful first draft.

Consider:

• What problem are we trying to solve?
• Could an existing tool complete the task?
• What value will AI add?
• How much checking and revision will the output require?
• Will the result support a measurable improvement?

The lowest-impact AI request may be the one that does not need to be made. This does not mean avoiding AI altogether. It means using it with a clear purpose.

Use fewer, better prompts

A vague request can lead to an unsuitable response and several rounds of correction. Giving the AI tool sufficient context at the beginning may produce a more relevant result with fewer interactions.

An effective prompt should explain:

1. The role or perspective the tool should adopt
2. The task it needs to complete
3. The relevant background and source material
4. Any limits, such as budget or scope
5. The required format and level of detail
6. How the response will be used

For example, “analyse this survey” leaves many decisions open. A more useful instruction might ask the tool to group responses into common sustainability themes, identify the five most frequently mentioned topics, provide a concise summary and present the results in a table.

The goal is not to make every prompt unnecessarily long. It is to provide enough relevant information to reduce avoidable revisions.

Match the AI tool to the task

Different AI activities can require different levels of computing resource. Sorting information, producing text, generating images and creating video are not equivalent tasks.

Where a platform offers a choice, organisations can consider using the lightest suitable model. A smaller or older model may be sufficient for a straightforward task, while more complex work may require a more capable option.

A practical approach is to:

• Use AI only where it creates genuine value.
• Choose the least resource-intensive suitable tool.
• Use focused text-based tasks where appropriate.
• Provide approved source material.
• Review outputs for accuracy and relevance.
• Monitor whether the benefits justify continued use.

This is about progress rather than perfection. Organisations can start with the choices available to them and improve their approach as better information becomes available.

Use AI agents selectively

Some platforms allow organisations to create agents or predefined working environments containing instructions, reference materials and organisational context. These can support repeatable processes because users do not need to enter the same background information for every request.

An agent could, for example, contain brand guidance, project information or approved reference documents. This may help produce more consistent and relevant outputs.

However, an agent is not necessarily the best option for a simple question. If it searches multiple files or sources when that context is not required, it may introduce unnecessary processing. Agents are most useful when their stored knowledge genuinely improves the result.

Include sustainability when assessing AI providers

Cost, privacy, security and functionality will remain central to AI procurement. Sustainability can sit alongside these factors rather than being treated separately.

Useful questions to ask a provider include:

• Do they publish information about energy, carbon, water and infrastructure?
• Are their reporting methods clear?
• Can users choose between different models?
• Can the organisation monitor or export usage data?
• Where is organisational data stored and processed?
• What action is the provider taking to reduce its impacts?

Broad claims such as “green AI” or “carbon neutral” should be supported by clear evidence. The objective is not to find a provider claiming to have no impact. It is to find one that offers sufficient transparency for an informed decision.

Take a measured approach to AI

AI technology, regulation and supporting evidence continue to develop. Organisations do not need to have every answer before taking sensible first steps.

Begin by identifying tasks where AI may provide clear value. Set expectations for responsible use, involve relevant colleagues, monitor outcomes and review your approach regularly.

AI can support people rather than simply replace existing roles or processes. Achieving that balance requires clear governance, appropriate skills and an honest assessment of where the technology genuinely helps.

Five practical actions to take next

1. Define the need. Be clear about the problem before selecting an AI tool.
2. Improve the input. Provide useful context, limits and an output format in the first prompt.
3. Choose proportionately. Match the model and tool to the complexity of the task.
4. Ask for evidence. Include environmental transparency in procurement and provider reviews.
5. Monitor and adapt. Check whether AI is producing measurable value and update your approach as evidence develops.

Frequently asked questions

Is AI sustainable?

AI cannot be described as simply sustainable or unsustainable. Its impact depends on factors including the task, model, hardware, electricity source, data centre infrastructure and how frequently it is used. Organisations should consider both the resources involved and the value created.

How can an organisation reduce unnecessary AI use?

Define the task before opening the tool, use AI only where it adds clear value, provide sufficient context in the first prompt and select the least resource-intensive suitable model where possible.

Can the footprint of an individual AI prompt be calculated?

Not reliably in every case. Providers do not always disclose enough information, and reporting approaches vary. Organisations should be cautious about precise figures that do not explain their assumptions and boundaries.

What should organisations ask an AI provider?

Ask about energy, carbon, water, infrastructure, model choice, data storage and usage reporting. Look for transparent evidence rather than broad environmental claims.

If you’re developing your sustainability strategy, explore our iiE case studies to see how organisations across different sectors are making credible sustainability progress and addressing practical environmental challenges.