For public servants, heavy workloads with endless admin can slow progress on the things that truly matter. Apolitical’s AI Skills Lab is a series of short sessions dedicated to helping public servants save time and deliver for the people they represent.

The most recent session focused on AI for Policymaking. During the event, James Ancell, Head of Foresight at the UK Cabinet Office, walked through how generative AI can strengthen policy thinking, speed up grunt work and help you communicate with more punch — as long as you do it with your eyes open.

Below are the most useful tips from the session.

1. Treat AI as a policy sidekick for speed, sense-making and stress-testing

James moved quickly through examples showing how AI can support different stages of the policy pipeline:

  • Drafting and summarising complex material into clear submissions.

  • Breaking down legislation or memoranda into stakeholder impact tables, scoring who is affected and how much.

  • Wind-tunnelling policy options across scenarios to see what breaks, and where.

  • News scanning tools that trawl global headlines, spot emerging risks and compare how different countries are responding, all powered by simple prompts and spreadsheets.

His bigger point: AI is very good at finding needles in haystacks and giving you a rapid first pass. That’s gold for busy teams, as long as you don’t skip the human work of consultation, judgement and consensus-building.

2. Use “meta-prompting” to get better prompts, then better policy

One of the most practical techniques James shared was meta-prompting: asking AI to write the prompt you actually need. Instead of jumping straight to “write me a two-page policy submission”, start with:

“Write me a prompt that will generate a two-page submission from a longer report, with the right tone, format, brevity, and audience focus.”

This extra step helps you bake in structure, context, and constraints before the AI starts drafting. Think of it as prompt planning, and it pays off fast.

3. Start with the risks, then design your prompts to reduce them

AI can be brilliant, but it comes with baggage. AI outputs can contain misinterpreted or sometimes fabricated facts – and you don’t want to include these in your work. To avoid this, ask the model to provide citations with links so you can verify sources yourself.

There are two other hazards to be aware of:

  • Narrow or biased outputs, because AI reflects what it was trained on, not necessarily the full policy world.

  • Misinformation risks, including deliberate manipulation of the information environment.

Interrogate outputs like you would any other evidence. AI is a tool, not a truth machine.

4. You’re still the policymaker, so keep the critical thinking switched on

Speaker James Ancell urged participants to stay sharp and to be intentional about when not to use AI. Carefully evaluate which stages of policy development AI is appropriate to use in. You will be held accountable for what it puts out!

Make sure you hold onto those critical thinking skills. There’s a real risk of skill fade if AI becomes a shortcut for everything. The sweet spot is using AI to accelerate your thinking, not replace it.

Finally, the most emphatic tip of the session: don’t put sensitive information into external AI tools. If you’re prototyping, use dummy data to build the structure first, then transfer the output into secure systems for real-world use.

It’s a small habit that keeps experimentation safe and sustainable, especially in high-trust policy environments.

Want more? Watch the event highlights!

If you missed the session, watch the highlights video below for a quick hit of James Ancell's best demonstrations and prompting tricks. You can also visit the event page to watch the full recording and access all resources.

How are you using AI for policymaking?

The session ended on an upbeat challenge: go experiment responsibly. As you experiment, make sure to share your insights and emerging questions with thousands of peers in Apolitical’s Global Policymaking Community.

Go to the Community

This article was written with the help of AI using an anonymised transcript and edited by a human before publishing.

Make sure to share your own thoughts with the author by leaving a comment below