Carbon-intelligent hiring starts with a simple idea: sustainable recruitment does not need less AI. It needs smarter, more proportionate AI.
In my previous AMS article, Small Steps, Big Impact: Redesigning Recruitment for a Carbon-Neutral Future, I focused on the parts of recruitment’s environmental footprint that we can see.
Candidate travel. Printed materials. Large hiring events. Repeated journeys into an office for interview stages that could just as easily have happened online.
That work still matters. There is no good reason to reintroduce waste simply because the sustainability conversation has moved on. But it has moved on, and recruitment’s growing use of AI makes the next stage far less straightforward.
Every AI-generated job description, CV summary, interview transcript, chatbot response or skills match relies on an infrastructure that most recruiters will never see. Behind the screen are cloud platforms, data centres, servers, electricity grids, cooling systems and water.
It is easy to describe a recruitment process as more sustainable because it has become digital. That assumption is becoming harder to defend.
I do not believe the answer is to retreat from AI. Used well, it can reduce administration, improve matching, surface internal talent and give recruiters more time for the work that still depends on judgement, context and empathy. In some cases, it could remove far more waste than it creates.
The real test is whether we are using it to improve the hiring process or simply to produce more activity, faster.
Why sustainable recruitment’s footprint is becoming harder to see
The first phase of sustainable recruitment dealt mainly with visible choices. We could calculate the journeys avoided by moving early-stage interviews online. We could reduce printed materials, rethink large events and challenge whether candidates really needed to travel several times before a decision was made.
Virtual and hybrid hiring remain among the clearest opportunities. Where an interview can be conducted well online, particularly at an early stage, there is little value in asking a candidate to spend hours travelling simply because that is how the process has always worked.
That does not require every hiring experience to become remote. There are moments when being together matters. Candidates may want to experience a workplace, meet future colleagues or get a proper sense of the organisation before making a significant career decision.
The point is to make those moments count, rather than building travel into a process by default.
AI adds a different type of impact. A recruiter may experience it as a faster shortlist. A hiring manager receives a clearer summary. A candidate gets a quicker response. The environmental cost sits somewhere further down the technology chain, shaped by the model being used, the amount of processing required, the location of the data centre, the local energy mix and the efficiency of the supplier’s infrastructure.
According to the International Energy Agency, data centres consumed around 415 TWh of electricity in 2024. It projects that demand could rise to approximately 945 TWh by 2030, with AI among the main drivers of that increase.
Recruitment technology is clearly only one small part of that picture. Even so, talent acquisition cannot adopt AI at scale while treating the infrastructure behind it as someone else’s responsibility.
We are already asking harder questions about bias, explainability, security and candidate trust. Environmental impact needs to start entering the same conversation.
More activity is not the same as greater efficiency
This is where I think recruitment needs to be honest with itself.
AI can remove waste from hiring, but it can also make waste easier to create. An outreach platform that allows a recruiter to contact thousands more people is not automatically efficient. It may simply produce more irrelevant messages, more disengaged candidates and more data for somebody else to process.
The same is true across the hiring journey. More applications create more screening. More automated content creates more material to review. More tools create more integrations, workflows and duplicated data. A process can become faster at each individual step while becoming less coherent overall.
We should be careful not to confuse the ability to do more with evidence that more needed to be done.
A skills-matching tool that finds someone already working within the organisation could prevent an external search, reduce onboarding effort and retain valuable knowledge. An interview intelligence platform might remove hours of administration and help a recruiter stay present during a conversation. Better market insight could stop a team launching an unrealistic search in the wrong location.
Those uses of AI have the potential to remove genuine friction.
Sending thousands of weakly targeted messages because automation has made them cheap is a very different proposition. Both might appear on a technology roadmap under the heading of AI, but their value to the candidate, recruiter and organisation is nowhere near the same.
This is also why I am cautious about the current fascination with calculating the carbon cost of a single prompt. There is no universal figure. Published estimates vary depending on the model, the task, the length of the response, the infrastructure and what the calculation includes.
Google and Mistral, for example, have published very different figures for individual AI interactions because they are measuring different systems in different ways. Attaching a single carbon number to an AI-assisted hire would create an impression of accuracy that the evidence does not yet support.
For talent acquisition, the more useful unit of measurement is the workflow.
Did the technology remove an unnecessary interview stage? Did it help find internal talent before an external campaign was launched? Did it improve the relevance of outreach? Did it reduce administration while keeping the candidate experience personal? Did it help the organisation make a better decision, or did it just increase the volume moving through the process?
I would rather see TA work towards an impact-per-quality-hire mindset than chase a theoretical carbon-per-prompt figure that tells us very little about whether the process improved.
The biggest model should not always be the default
There is a tendency in technology to assume that the most advanced model must also be the best choice. That makes sense in a product demonstration. It makes far less sense as an operating principle.
Recruitment includes a huge range of tasks, and they do not all need the same level of intelligence.
Scheduling, basic data extraction, structured updates and simple workflow triggers may be handled perfectly well through conventional automation or a smaller specialist model. More capable generative models may add real value when the work requires synthesis, contextual understanding, complex language or interpretation across several sources.
Using a smaller model is not settling for an inferior solution when it achieves the same outcome. It is choosing technology that is proportionate to the work.
Research from UCL and UNESCO suggests that practical design choices, including smaller specialist models, shorter prompts and responses, and more efficient processing, can substantially reduce energy use. In some of the scenarios tested, the combined reduction was as high as 90 per cent compared with using a large general-purpose model.
That should matter to buyers of recruitment technology. It should also influence how products are designed.
Using the most powerful model available for every task may sound innovative, but it can be the technological equivalent of using a removal van to deliver an envelope. It works, but it is difficult to call it intelligent.
There is also growing interest in carbon-aware computing, where non-urgent workloads can be processed at times or in locations with lower grid-carbon intensity. Candidate-facing interactions will often need to happen immediately, but not every recruitment task does. Bulk reporting, data processing, interview summarisation and some forms of analysis may have more flexibility.
This capability is still developing, and I would not pretend that most TA teams can switch it on tomorrow. It does, however, show where expectations are heading. Recruitment platforms will increasingly need to explain not only what their AI can do, but how efficiently it does it.
Sustainable hiring still needs to feel human
There is a danger that this becomes a discussion about servers, models and procurement questionnaires. Sustainable recruitment is bigger than its technical footprint.
A hiring process can use less energy and still be a terrible experience. It can be fast but opaque. It can be efficient for the employer while placing more effort on the candidate. It can automate communication so heavily that people no longer know whether anybody has genuinely considered their application.
Environmental and social sustainability cannot be separated that neatly.
Recruiters are already reporting meaningful time savings from AI. What matters now is what happens to that time. Does it create more space for proper candidate conversations, better advice to hiring managers and more consistent communication? Or does it simply raise the expectation that each recruiter should process a greater volume of work?
The second outcome may improve a productivity measure while doing very little for recruitment.
For me, the strongest case for AI is not that it can remove the recruiter from more stages of hiring. It is that it can remove the repetitive work that prevents recruiters from adding value when people actually need them.
Human judgement still matters when a requirement needs challenging, when a candidate’s experience does not fit a standard profile, when there are trade-offs in a hiring decision or when somebody needs an honest conversation about what happens next.
A sustainable process should preserve those moments, not automate them away.
What TA leaders should ask next
Talent acquisition leaders do not need to become experts in data-centre engineering. They do need to become more demanding customers.
When a supplier says its product is AI-enabled, we should understand what that actually means. Which models are being used? Are they appropriate for the task? Where is the processing taking place? Can usage be measured at a customer or workflow level? What is included in the supplier’s environmental reporting, and just as importantly, what is left out?
Water consumption deserves attention alongside energy use. So does the supplier’s approach to fairness, explainability, privacy and human oversight. These are not separate conversations. They are all part of understanding the full cost of a technology choice.
Procurement teams will need to help. So will technology, risk, sustainability and data leaders. TA should not attempt to solve this alone, but it should have a clear enough view of the hiring workflow to challenge whether the technology is being applied in the right places.
We also need to be more precise in the language we use. Claims about “green AI” or “carbon-neutral technology” are difficult to stand behind without a clear methodology and reporting boundary. A process may be lower impact than the one it replaced. A platform may be designed to use resources more efficiently. A workflow may reduce unnecessary hiring activity.
Those claims may sound less dramatic, but they are more credible.
This is not about creating another layer of governance that slows every decision. It is about asking better questions before AI becomes so embedded in hiring that changing direction becomes difficult.
By 2030, leading TA functions may be measured on far more than time-to-hire, cost-per-hire and quality-of-hire. They may also need to understand the wider resource intensity of their operating model, including candidate travel, recruiter effort, events, cloud usage and supplier infrastructure.
That sounds complicated, but the principle behind it is familiar.
Use the right tool for the right work. Understand what it improves. Be honest about the trade-offs. Stop doing the things that add activity without adding value.
The future of sustainable recruitment is not less AI. It is less waste, better judgement and much clearer accountability for how the technology is used.
