what we do
the AI prompt library
Every prompt from the digital bootcamp, collected in one place and grouped by the job you are doing. 59 prompts today, and each new episode adds its own.
Every prompt says which day it comes from, so you can go back to the exercise that explains it. They are starting points: give them your real context, and check what comes back against primary sources.
select a collection
Prompt basics
7Role, audience and context, and the assumptions a model makes wherever you leave a gap.
How models work
5See prediction, fluency and training data for yourself, in your own tool.
Model and effort
5Compare settings, make assumptions explicit, and stress-test an answer before you use it.
Big policy moments
3Summarise a speech, map it to your files, and turn it into an advocacy plan.
Research and evidence
3Organisational and issue research with sources, gaps, and context you can reuse.
Narrative and message
5Public narrative, and the same approved message carried across channels.
Stakeholders and audiences
4Verified stakeholder research, audience accessibility, and an inclusion check.
Visuals and video
6A creative brief first, then image concepts and short-form video prototypes.
Publish, measure and translate
4Rollout planning, measurement levels, and localisation that a fluent human approves.
Choosing your tools
2Draw up a shortlist for your own organisation, then compare products on your own work.
More from the bootcamp
15Prompts that have not been filed into a collection yet.
Sign in to copy the prompts
The bootcamp and its prompts are free. You need an account so we know who is working through them, and so your progress is saved.
Prompt basics
Role, audience and context, and the assumptions a model makes wherever you leave a gap.
Control prompt: no role, no audience, no constraints
Log in to read and copy this prompt.
Day 1 · Exercise 1: run the same question with three different roles →Role prompt A: communications trainer
Log in to read and copy this prompt.
Day 1 · Exercise 1: run the same question with three different roles →Role prompt B: EU civil servant
Log in to read and copy this prompt.
Day 1 · Exercise 1: run the same question with three different roles →Vague audience prompt: one word of context
Log in to read and copy this prompt.
Day 1 · Exercise 2: audit what the model assumes →Assumption-audit prompt
Log in to read and copy this prompt.
Run this in the same conversation, straight after the answer you want to audit.
Day 1 · Exercise 2: audit what the model assumes →Context-rich audience prompt
Log in to read and copy this prompt.
Day 1 · Exercise 3: replace assumptions with real context →The human version
Log in to read and copy this prompt.
Day 4 · Anatomy of a useful working prompt →
How models work
See prediction, fluency and training data for yourself, in your own tool.
Prediction test: next word
Log in to read and copy this prompt.
Run it three times, each in a new conversation, and compare the first word of each answer.
Day 2 · Large language models: prediction machines →Fluency test: persuasive language
Log in to read and copy this prompt.
Day 2 · Large language models: prediction machines →Representation test: which humans?
Log in to read and copy this prompt.
Day 2 · Training data: which humans are represented? →Dictionary test: explain the stack
Log in to read and copy this prompt.
Day 2 · The AI dictionary on one screen →Verification test: what needs checking?
Log in to read and copy this prompt.
Day 2 · Exercises →
Model and effort
Compare settings, make assumptions explicit, and stress-test an answer before you use it.
1. The same task, two settings
Log in to read and copy this prompt.
Run it once with a faster model on lower effort, then with a stronger model on higher effort, and compare.
Day 3 · Prompt pack: compare, verify and stress-test →2. Make the plan and assumptions explicit
Log in to read and copy this prompt.
Day 3 · Prompt pack: compare, verify and stress-test →3. Stress-test the argument
Log in to read and copy this prompt.
Day 3 · Prompt pack: compare, verify and stress-test →4. Decide how much AI this task deserves
Log in to read and copy this prompt.
Day 3 · Prompt pack: compare, verify and stress-test →5. Compare two model outputs
Log in to read and copy this prompt.
Day 3 · Prompt pack: compare, verify and stress-test →
Big policy moments
Summarise a speech, map it to your files, and turn it into an advocacy plan.
Summary prompt
Log in to read and copy this prompt.
Check every material commitment against the official transcript before circulating the summary outside your immediate working team.
Day 3 · Worked exercise: build your SOTEU AI workflow →Mapping prompt
Log in to read and copy this prompt.
Use a stronger model at high effort. Read the evidence and stated assumptions as carefully as the conclusion.
Day 3 · Worked exercise: build your SOTEU AI workflow →Action plan prompt
Log in to read and copy this prompt.
The answer may sound strategically convincing even where the underlying assumptions are weak. Treat it as a draft for human judgement, and keep the campaign decision with the team.
Day 3 · Worked exercise: build your SOTEU AI workflow →
Research and evidence
Organisational and issue research with sources, gaps, and context you can reuse.
Prompt 1.1 · Organisational research
Log in to read and copy this prompt.
Day 4 · Build a verified organisational brief →Prompt 1.2 · Stable context
Log in to read and copy this prompt.
Day 4 · Create a reusable organisational brief →Prompt 2 · Deep research
Log in to read and copy this prompt.
Day 4 · Build the issue evidence base →
Narrative and message
Public narrative, and the same approved message carried across channels.
Prompt 3 · Public narrative
Log in to read and copy this prompt.
Day 4 · Explain why the campaign exists →Sub-prompt 3.1 · Clarity, authenticity and evidence
Log in to read and copy this prompt.
Day 4 · Explain why the campaign exists →Sub-prompt 3.2 · Online reading
Log in to read and copy this prompt.
Day 4 · Explain why the campaign exists →Prompt 5 · Channel adaptation
Log in to read and copy this prompt.
Day 4 · Turn the campaign framework into clear public communication →Follow-up · The clarity test
Log in to read and copy this prompt.
Output: first-draft channel adaptations ready for human editorial and compliance review.
Day 4 · Turn the campaign framework into clear public communication →
Stakeholders and audiences
Verified stakeholder research, audience accessibility, and an inclusion check.
Prompt 4 · Verified stakeholder research
Log in to read and copy this prompt.
Day 4 · Map the stakeholder landscape →Layer in deep research
Log in to read and copy this prompt.
Day 4 · Map the stakeholder landscape →Prompt 4.1 · Audience accessibility
Log in to read and copy this prompt.
Day 4 · Test whether your communication works for different audiences →Prompt 4.2 · Inclusion check
Log in to read and copy this prompt.
Day 4 · Inclusion check: who could our campaign be overlooking? →
Visuals and video
A creative brief first, then image concepts and short-form video prototypes.
Prompt 6 · Visual concepts
Log in to read and copy this prompt.
Day 5 · Build the visual brief →Follow-up · Generate a prototype
Log in to read and copy this prompt.
Day 5 · Build the visual brief →Prompt 7 · Short-form video prototype
Log in to read and copy this prompt.
Day 5 · Turn the visual brief into motion →The full image brief
Log in to read and copy this prompt.
There is nothing magic in the wording. It is a complete creative brief, which is why it works in almost any tool.
Day 7 · Put the brief together →Turn one message into three visual concepts
Log in to read and copy this prompt.
Run this before you generate anything. Three briefs cost five minutes and save an afternoon.
Day 7 · One message, three visual directions →Derive your organisation’s visual brief
Log in to read and copy this prompt.
This is the visual twin of the organisational brief from day four. Keep it with your AI usage policy.
Day 7 · Build your organisation’s visual recipe →
Publish, measure and translate
Rollout planning, measurement levels, and localisation that a fluent human approves.
Prompt 8 · Seven-day content plan
Log in to read and copy this prompt.
Day 5 · Build the rollout around campaign objectives →Prompt 9 · Measurement framework
Log in to read and copy this prompt.
Day 5 · Measure the campaign against its objective →Follow-up · Hypotheses to test
Log in to read and copy this prompt.
Day 5 · Measure the campaign against its objective →Prompt 10 · Translation and localisation
Log in to read and copy this prompt.
Day 5 · Adapt the campaign across languages →
Choosing your tools
Draw up a shortlist for your own organisation, then compare products on your own work.
Draw up your shortlist
Log in to read and copy this prompt.
Treat the shortlist as a starting point. The next prompt is where the actual test happens.
Day 6 · Before you buy, test it on ten of your own tasks →Compare two tools on the same task
Log in to read and copy this prompt.
Run this for each of your ten tasks and keep the tables. Together they are your benchmark, and they beat anybody else’s.
Day 6 · Before you buy, test it on ten of your own tasks →
More from the bootcamp
Prompts that have not been filed into a collection yet.
Monitoring agent: starter instruction
Log in to read and copy this prompt.
Replace every bracket before you run it. What matters here is the structure: role, mission, scope, sources, process, boundaries, output.
Day 8 · A starter instruction for a monitoring agent →What to write in the instructions box
Log in to read and copy this prompt.
The last line is the boundary. Write it before you are asked for it.
Day 9 · Microsoft 365 Copilot: Agent Builder →Project instructions for EU policy monitoring
Log in to read and copy this prompt.
Six lines. The sixth is the one that saves you.
Day 9 · Claude: Projects →Project instructions
Log in to read and copy this prompt.
The last line matters more than it looks. It is permission to go and check.
Day 9 · ChatGPT: Projects →The instruction
Log in to read and copy this prompt.
Paste this on its own first. Nothing else yet.
Day 10 · Layer 1. It is all about customisation →Test prompt, first audience
Log in to read and copy this prompt.
Day 10 · Layer 1. It is all about customisation →Test prompt, second audience
Log in to read and copy this prompt.
Day 10 · Layer 1. It is all about customisation →Add this to the instructions
Log in to read and copy this prompt.
The two steps are the whole change. Everything above them is layer 1, unchanged.
Day 10 · Layer 2. Flip the interaction →Test prompt
Log in to read and copy this prompt.
Day 10 · Layer 2. Flip the interaction →The instruction, now with both roles
Log in to read and copy this prompt.
Day 10 · Layer 3. Give it a persona, then a second one →Test prompt
Log in to read and copy this prompt.
Day 10 · Layer 3. Give it a persona, then a second one →Add the fact-checking step
Log in to read and copy this prompt.
Step 5 is the addition. The permission to say "I am not sure" is the important half of it.
Day 10 · Layer 4. Prevent the invention →Test prompt
Log in to read and copy this prompt.
Day 10 · Layer 4. Prevent the invention →The complete instruction set
Log in to read and copy this prompt.
This is the finished agent. Five layers, one block.
Day 10 · Layer 5. Ground it in the official documents →Test prompt
Log in to read and copy this prompt.
Day 10 · Layer 5. Ground it in the official documents →