Experimenting with prompts

Warning This is not a comprehensive guide, but a starting point to support your AI journey. This is not intended to be any form of legal guidance or advice, and should not be used as a substitute for independent legal advice.

AI assistants use written, visual or audio information and instructions (a prompt) and create the most likely response. The best way to learn is by experimenting. Find out how to use AI at work. 

Always check AI output for accuracy, as it can give false information, outdated details or biased outputs. 

Use conversation to get better results from AI

You don't need the perfect prompt to get the results you want. Working with AI is a conversation - you can improve results both before and after the AI responds.

Ask AI to improve your prompts before you run them, for example:

  • "before you answer, review this prompt and suggest ways to make it clearer"
  • "before you answer, ask me three questions to make sure you understand what I need"
  • "I'm using this prompt with [AI tool name] - review it and suggest ways to improve it for better results"

Ask the AI to show its work so you can assess the quality of responses, for example:

  • "explain your reasoning step by step"
  • "highlight anything you're uncertain about"
  • "list all sources and direct quotes you used"

Refine the output if it isn't quite right, for example:

  • "revise the second paragraph to focus on costs"
  • "add more detail about the timeline"
  • "suggest three alternative approaches"
  • "explain why you recommended this option"

Example: iterating a draft for a briefing

First prompt:

"Draft a briefing on the new flexible working policy for department heads."

[AI shares initial draft]

Follow-up prompt:

"Make the tone more formal and add a section on implementation timeline."

[AI revises]

Follow-up prompt:

"The implementation section is too detailed. Condense it to three key milestones."

Write simple prompts for everyday tasks

Most everyday tasks just need a simple, clear instruction. To write a good prompt:

  • be specific about what you want - use direct language and action words like 'write', 'analyse', 'create' or 'list'
  • be specific about who it's for - describe your audience and their background so the AI uses the right tone and level of detail
  • be specific about the format - word count, structure (email, bullet points, table) and tone
  • give relevant background - include details the AI needs to give you a useful answer

Example: simple prompts

A bad prompt

Draft an email about flexible working 

A good prompt

// What I want

Draft an email explaining the new flexible working guidance

// Who it's for

Line managers who will be implementing it

// Background

Policy applies from April 2025, covers hybrid working and compressed hours

// Format

Email under 200 words, conversational but professional tone

Write detailed prompts for complex tasks

Some tasks need more information to get good results. These are usually tasks that involve multiple steps, need specific expertise, or produce longer outputs. 

Use detailed prompts when you need to: 

  • analyse data and draw conclusions 
  • create reports or strategies 
  • work with multiple documents 
  • produce detailed plans

You can structure detailed prompts using frameworks to help you remember what information to include. 

 Use the RACE framework for focused tasks

RACE works well for tasks that need one clear output, like identifying risks or drafting summaries. Write you prompt with these sections: 

  • Role - what role (profession) to take 
  • Action - what you want it to do 
  • Context - any necessary background 
  • Expected output - specify the format and approach 

Example: RACE prompt

// Role

You are a risk manager on a construction project

// Action

Identify the top 5 delivery risks

// Context

For a £200 million road upgrade in a busy city centre

// Expected output

Under 150 words, use bullet points, write in plain English

 Use the GCSE framework when you have source documents

GCSE helps when you're working with existing documents or data, on tasks that need accuracy and compliance. Write you prompt with these sections: 

  • Goal - what you want AI to achieve 
  • Context - any relevant background information 
  • Source material - list documents, links or data to use 
  • Expected output - format and tone 

Example: GCSE prompt

// Goal

Create a project schedule for our website redesign

// Context

The project serves 50,000 users. We have 3 developers and a 6-month deadline

// Source material

Use the project brief dated March 2025 and the technical requirements document

// Expected output

Table with dates, milestones, owners and dependencies

Use the CRIT framework for strategic or creative tasks

CRIT works well for tasks where structure and tone matter, like drafting frameworks or evaluation criteria. Write your prompt with these sections: 

  • Context - any relevant background information 
  • Role - what role (profession) to take 
  • Intent - what you want AI to achieve 
  • Type of output - format and tone

Example: CRIT prompt

// Context

Department for Transport needs to evaluate bids for electric vehicle charging infrastructure

// Role

You are a procurement advisor specialising in sustainability assessment

// Intent

Draft a sustainability scoring matrix for supplier bids

// Type of output

Table with criteria, weightings, and evaluation notes. Use clear, neutral language

 Use the PICSE framework for iterative workflows 

PICSE works well when you want the AI to follow a complex or structured approach, such as creating strategies, or planning consultations. It helps the AI check it's working, and follow a specific process. 

  • Persona - the role, expertise, job title 
  • Intention - What do you want to achieve and how  
  • Criteria - key criteria you need for this output (length, tone, format?) 
  • Sources – what information you want to the AI to use 
  • Examples – show what good looks like 

Example: PICSE prompt

// Persona

You are a senior policy advisor in the Department for Transport with expertise in stakeholder engagement and infrastructure policy.

// Intention

Design an expert stakeholder workshop to gather input on the government's Electric Vehicle Infrastructure Roadmap for ministerial recommendations.

// Criteria

Create an engaging format that maximises expert input, ensures balanced representation, and generates actionable policy insights.

// Sources

Reference the Green Book guidance on stakeholder engagement and Cabinet Office consultation principles (attached).

// Examples

Here are some examples of previous workshop agendas and facilitation guides.

Break down complex tasks into steps

Some tasks are too complicated for a single prompt. Split them into separate steps, where each prompt builds on the output of the last (known as 'prompt chaining'). Use this approach when you need to: 

  • produce a major report or strategy 
  • do detailed analysis that needs validation 
  • work through multiple stages of a process 

How it works: 

  1. Each prompt has one clear purpose.
  2. You refine each output through conversation before moving to the next prompt.
  3. Each refined output feeds into the next prompt.
  4. You maintain control at each stage.

This approach produces better results than requesting everything at once. You catch issues early, refine each stage, and maintain control over quality throughout.

Example: Creating a project implementation plan:

Rather than asking for a complete plan in one prompt, break it into stages:

Step 1:

"Review these project requirements and extract the five main deliverables and their dependencies."

You review and refine the deliverables through follow-up prompts

Step 2:

"Based on the deliverables above, create a timeline showing key milestones and resource requirements for a 6-month project."

You refine the timeline through conversation

Step 3:

"Draft an implementation plan including the timeline, risk mitigation steps, and communication approach. Format as a structured document with clear sections."

You review and refine the final plan

What to do next

Find out more about:  

Back to top