Former senior civil servants Kate Sturdy and Clare Dobson share lessons from training policymakers as AI moves from pilots to everyday use. They explain why prompting is a discipline, critical thinking prevents hallucinations, and the smartest officials know when to think first and use AI second for better public outcomes.
Government use of AI tools is growing. Last year, a government study trialling the use of generative AI with 20,000 civil servants concluded that that the potential time savings were equivalent to nearly two working weeks per person, per year.
Thousands of civil servants are now using AI
This year the use of generative AI tools in government has shifted dramatically from pilots to large-scale operational deployment. Thousands of civil servants are now using AI in departments such as HMRC and MHCLG.
We are former Senior Civil Servants with an expertise in policymaking who now run executive coaching and consultancy businesses. Alongside this, we work as associates for Total Training (TJ’s sister company), designing and leading sessions to upskill civil servants in policymaking, strategic thinking and communications.
Over the past 12 -18 months, we have observed that training participants are increasingly reaching for AI tools (such as Copilot, ChatGPT or Claude) as the default starting point for completion of exercises and tasks. Others, by contrast, have barely touched them. We became rapidly conscious of a gap in capability not only between course participants, but also between our own skill and confidence levels compared to some of those we were training.
As we started to upskill ourselves and adapt our materials, it became increasingly clear that the challenge extended beyond learning new software. We now understand better how to use AI effectively and offer some key lessons here.
Is prompting the new discipline?
Our first lesson, shared by Tom Bryant of TFB Consulting, was to think of the AI tool as a ‘thought partner’ rather than a ‘vending machine’. The tool can produce a superficially plausible response to almost any question, but the real skill lies in prompting it to genuinely drive quality outputs. Learning how to ask clearly for what you need is critical – providing context, setting parameters around time to avoid outdated outputs, specifying source content to drive trustworthy outputs. This is what enables the AI tool to add real value.
And from this flows an additional realisation: that the act of prompting must now be a core discipline. Successful prompting requires open, clear thinking and commissioning. Nothing can be left to intuition, assumption or osmosis, although these can sometimes become default modes within the Civil Service. If we cannot clearly articulate what we need, AI is unlikely to give us the value-add so needed in our work serving the public.
Is critical thinking the antidote to AI hallucination?
The next insight emerged as we explored the opportunities and risks of AI tools. We encountered a widely reported US case involving Roberto Mata, who in 2022, filed a personal injury lawsuit after allegedly being injured by a metal serving cart on an international flight. His lawyers used ChatGPT to draft a legal motion, which included fictitious legal cases, fabricated quotations and entirely invented citations. Unsurprisingly, the judge dismissed the case which has since become a prominent example of an “AI hallucination”, where an AI system generates information that appears convincing but is actually false.
AI hallucination is to be expected when you consider the technology is designed to generate plausible responses, not to exercise judgement. But that makes critical thinking more important, not less. People must still interrogate sources, challenge assumptions and verify evidence. In many ways, the rise of AI increases the value of precisely those human skills, scepticism, judgement and intellectual rigour, that underpin good policymaking and sound decision-making.
Is successful use of AI about knowing when not to use AI?
The biggest shift for us came when we examined each stage of the policymaking process and considered where generative AI tools genuinely add value. At its core, policymaking is the search for solutions to problems, many of which societies have never confronted before. Hybrid war and new security threats, the rapid emergence of AI, and the accelerated decarbonisation of energy systems all present governments with challenges for which there is no established playbook. Addressing them requires original thinking from people who deeply understand the problem and the associated wider context.
People experience the world directly: we notice tensions, identify opportunities, and make conceptual leaps. Generative AI, by contrast, learns statistical patterns from existing data and recombines them in sophisticated ways. That capability is powerful, but it also creates a risk. Emerging research suggests that overreliance on AI can weaken independent reasoning and reduce cognitive effort. Consequently, at the ideas-generation stage of policymaking, our advice to officials to try to do the human thinking first. There is no substitute for wrestling with a problem yourself, or with a team, using combined experiences to tackle tricky questions. Once the human thinking has been done, however, AI can be used as a thought partner to test outputs or generate alternative options.
The same principle applies when training officials to brief Ministers. Our advice is to draft the briefing first without AI assistance. If you understand the narrative and logic of a problem deeply enough to explain it clearly, you are far better placed to respond under challenge. Put simply: if you cannot explain it, you do not properly understand it, and no AI tool will rescue you in front of a Minister!
Risk and reward
All advances in technology carry risk. Each generation faces a new set of threats and cynicism to overcome in the war against the machines. But there are also many reasons to be cheerful, and opportunities to learn and embed new skills that will enrich our working lives.
As Tom summarised, “AI works best when it challenges and extends human thinking, not when it replaces it. The future advantage will belong to people who can combine curiosity, judgement and critical thinking with these new tools.” Certainly, for government policymaking these human qualities are becoming more important not less.
The civil servant now and in the future will need to keep pace with AI tools, decide when to use them, how to challenge them, and when to prioritise independent thinking over machine assistance. Our task as an L&D community is to build the skills and confidence to apply them wisely in service of better policymaking and public outcomes.
Kate Sturdy is Director of Kate Sturdy Coaching
Clare Dobson is Director of The Policy Key
