The Middle Principle is a way to use AI at work without outsourcing your skill or your decisions: you set the direction, AI accelerates the middle, you make the judgement. Choose a role or team to see practical examples and copyable prompts. Do not paste confidential, personal or identifiable information.
Human Direction
You set purpose, audience, constraints and what must not be claimed — before generation.
AI Acceleration
AI drafts, structures and compares options in the middle of the work. Output stays provisional.
Human Judgement
You check, adapt and take responsibility. A tool cannot inherit your authority.
Direction is the brief: audience, offer, proof and what you will not claim. AI can expand options for campaigns, content, research summaries and proposals. You check accuracy, tone and whether you would put your name to it.
Human Direction→AI Acceleration→Human Judgement
Campaign planning
Campaign planning
Human Direction
Write who it is for, the one outcome, the proof you may use, and the claims you will not make. Do not paste customer lists or unpublished figures.
AI Acceleration
A realistic prompt. Fill the brackets with your own context. Do not paste confidential or identifiable information.
Help me plan a campaign. Do not write finished ads yet.
Audience: [who, in one sentence]
Outcome I want: [one sentence]
Offer in my words: [one sentence]
Proof I may use (public or already approved): [list]
Claims I will not make: [list]
Channel: [e.g. email, LinkedIn, print]
Constraints: [length, tone, brand notes]
Give me three campaign angles, five subject or headline options, and the questions I must still answer before anything is published. Flag anything that sounds like an unproven claim as [CHECK].
Human Judgement
Check every claim against approved proof. Cut anything you would not defend. You decide what is published and you remain responsible for it.
Content development
Content development
Human Direction
Decide the reader, the point, and the sources you trust. Ask for options and a labelled draft, not a finished piece you could post as-is.
AI Acceleration
A realistic prompt. Fill the brackets with your own context. Do not paste confidential or identifiable information.
Draft content from this brief. Stay inside the constraints. If a fact is missing, mark it [UNKNOWN] rather than inventing it.
Reader: [who]
Point I am making: [one sentence]
Must include: [points]
Must not include: [points]
Sources I will check myself: [titles or notes — no confidential files]
Tone: [plain / warm / firm]
Length: [n words]
After the draft, list (a) what you assumed (b) what I must check (c) what only I can decide.
Human Judgement
Fact-check. Rewrite into your voice. Remove anything invented. You own the published words.
Customer research and proposals
Customer research and proposals
Human Direction
Summarise the problem in your own words. Do not paste CRM notes, personal data or another organisation’s confidential material.
AI Acceleration
A realistic prompt. Fill the brackets with your own context. Do not paste confidential or identifiable information.
Help me structure a proposal. Do not invent the client’s situation.
Problem in my words: [one paragraph, no names if sensitive]
What we can actually offer: [list]
What we will not promise: [list]
Reader: [role]
Length: [e.g. one page]
Give me a structure, five questions I should still ask the reader, and three ways to state the offer without overclaiming. Do not invent case studies, results or testimonials.
Human Judgement
Check the offer against what you can really do. Add only evidence you hold. You remain responsible for the proposal.
Direction is purpose, audience and what must not be said. AI can draft emails, tidy notes and suggest process options. You check accuracy, tone and confidentiality before anything is sent.
Human Direction→AI Acceleration→Human Judgement
Emails
Emails that still sound like you
Human Direction
Write the purpose, the reader, and what they need to do. Leave out names or details you would not put in a shared inbox.
AI Acceleration
A realistic prompt. Fill the brackets with your own context. Do not paste confidential or identifiable information.
Help me draft an email. Confirm the brief first, then wait.
Purpose: [one sentence]
Reader: [role, not a private email address]
What they need to do: [action and deadline if any]
Tone: [plain / firm / warm]
Must include: [points]
Must not include: [points]
Give me a five-line brief to confirm. Only draft after I say the brief is right. Then give me three judgement questions before I send.
Human Judgement
Read it as if you were the recipient. Check facts and tone. You click send, so you own it.
Meeting notes
Meeting notes you can stand behind
Human Direction
Capture decisions and actions in your own words. Do not paste a recording or anything that identifies people unnecessarily.
AI Acceleration
A realistic prompt. Fill the brackets with your own context. Do not paste confidential or identifiable information.
Turn these anonymised notes into a clear action list. Do not invent who said what.
Meeting purpose: [one sentence]
Notes (no personal data):
[your bullets]
Produce: decisions, actions with owner role (not names unless I have included them), open questions, and anything that is still unclear marked [CHECK]. Do not add items I did not mention.
Human Judgement
Check every action against what was actually agreed. Add names only if you are sure. You remain responsible for circulating the note.
Process improvement and project coordination
Process improvement and coordination
Human Direction
Describe the current steps, the bottleneck, and what “better” means. Do not upload internal systems diagrams that are confidential.
AI Acceleration
A realistic prompt. Fill the brackets with your own context. Do not paste confidential or identifiable information.
Help me improve a process. Do not redesign the organisation.
Process in my words: [steps]
Where it currently fails: [one sentence]
Constraint I cannot change: [e.g. two people, a weekly deadline]
What better looks like: [one sentence]
Give me three small changes, a simple checklist for the next run, and questions I should ask the people who do the work. Do not invent tools we do not have.
Human Judgement
Test the suggestions with the people who run the process. You decide what changes and you remain accountable for the result.
Direction is policy, fairness and care. AI can help you draft onboarding, staff communication and training outlines. You check legal fit, tone and whether you would be willing to stand behind the words. Never paste staff records or health information.
Human Direction→AI Acceleration→Human Judgement
Onboarding
Onboarding that is accurate
Human Direction
List what a new starter actually needs in week one. Use public or already-approved policy language. Do not paste personal files.
AI Acceleration
A realistic prompt. Fill the brackets with your own context. Do not paste confidential or identifiable information.
Help me outline a first-week onboarding plan. Do not invent policy.
Role (generic): [e.g. office coordinator]
What they must know by Friday: [list]
Sessions we can actually offer: [list]
Tone: [welcoming, clear]
Give me a day-by-day outline, a checklist the starter can use, and questions I should still confirm with managers. Mark anything that needs a real policy check as [CHECK].
Human Judgement
Align with your actual policies and who owns each session. You remain responsible for what new staff are told.
Staff communication
Staff communication
Human Direction
Write the purpose, what staff need to do, and what you will not say. Keep it general; do not include individuals.
AI Acceleration
A realistic prompt. Fill the brackets with your own context. Do not paste confidential or identifiable information.
Draft a staff message. Confirm the brief before drafting.
Purpose: [one sentence]
What staff need to do: [action]
What I will not say: [list]
Tone: [calm, plain]
Length: [short]
Give me a five-line brief, wait for confirmation, then a draft plus three judgement questions (clarity, fairness, what might be misread).
Human Judgement
Read it as a sceptical colleague would. Check it matches what leaders have agreed. You send it, so you own it.
Training and development
Training and development
Human Direction
Decide the skill, the audience, and how you will know it worked. Ask for a session outline, not a finished course you have not reviewed.
AI Acceleration
A realistic prompt. Fill the brackets with your own context. Do not paste confidential or identifiable information.
Outline a short staff development session. Do not write a full slide deck.
Skill: [e.g. giving feedback, using AI in admin]
Audience: [role]
Time: [mins]
What they should be able to do afterwards: [one sentence]
Non-negotiable: people remain responsible for their own work
Give me a session structure, two practice activities, and discussion questions. Do not invent legal or HR advice. Label uncertainty as [CHECK].
Human Judgement
Check activities against your culture and policy. Facilitate in your own words. You remain responsible for the session.
Direction is the decision, the constraints and who is affected. AI can organise options and questions. It cannot inherit your authority. You make the call and you remain accountable.
Human Direction→AI Acceleration→Human Judgement
Decision preparation
Decision preparation
Human Direction
State the decision, the options you already see, and the constraints. Do not paste confidential board papers or personal data.
AI Acceleration
A realistic prompt. Fill the brackets with your own context. Do not paste confidential or identifiable information.
Help me prepare a decision. Do not tell me what to decide.
Decision: [one sentence]
Options I already see: [list]
Constraints: [time, budget band in general terms, people affected]
What I must not ignore: [risks or duties]
What “good” looks like: [one sentence]
Give me: a clearer framing of the decision, questions I should still ask, arguments for and against each option, and what I would need to check before choosing. Do not invent data.
Human Judgement
Bring in the people and facts the tool cannot see. You choose. You remain accountable for the outcome.
Strategy
Strategy on one page
Human Direction
Write the purpose of the period, what will not change, and how you will know it is working. Keep commercially sensitive numbers out.
AI Acceleration
A realistic prompt. Fill the brackets with your own context. Do not paste confidential or identifiable information.
Help me structure a one-page strategy note. Do not invent a vision.
Purpose of this period: [one sentence]
What will not change: [list]
Priorities in my words: [three]
How we will know it is working: [signs, not fake metrics]
Audience: [e.g. team leads]
Give me a structure, tighter wording for the priorities, and questions a sceptical colleague would ask. Do not invent results or testimonials.
Human Judgement
Test the page with the people who must deliver it. You own the priorities.
Team development and change planning
Team development and change planning
Human Direction
Describe the change in plain language, who is affected, and what you will protect. Do not include individuals’ performance notes.
AI Acceleration
A realistic prompt. Fill the brackets with your own context. Do not paste confidential or identifiable information.
Help me plan a change conversation with a team. Do not write HR casework.
The change in one sentence: [plain language]
What will stay the same: [list]
Who is affected (roles, not names): [list]
Timeframe: [e.g. this term]
What I will not pretend: [e.g. that nothing will be different]
Give me a conversation structure, questions I should invite, and risks if I only send an email. Do not invent legal advice.
Human Judgement
Speak as yourself. Check the plan with whoever owns people policy. You lead the change and you remain responsible for how it is done.
