AI Briefing for Managers Prompts Guardrails and Reviews

AI Briefing for Managers: Prompts, Guardrails, and Reviews. Managers who brief AI effectively can significantly increase their impact. The key isn’t just using AI; it’s engaging with it thoughtfully. Effective managers provide clear context, define desired outcomes, set explicit boundaries for what AI should and shouldn’t do, and critically review AI output before acting. This disciplined approach ensures AI acts as a force multiplier, not a source of risk or confusion. AI can help managers turn incomplete information into structured plans, highlight issues needing attention, prepare for sensitive conversations, summarise complex material, and draft communications or decision documents.

Despite these benefits, AI is a support tool, not a substitute for human judgement. Managers remain responsible for decision quality, fair treatment, information accuracy, and the consequences of actions based on AI output. This requires ongoing oversight, adherence to professional standards, and readiness to challenge or override AI suggestions.

Treat each interaction with AI as a professional briefing, similar to briefing a team member or advisor. For instance, a vague prompt like “Help me prepare for this meeting” leaves too much open to interpretation. An effective briefing should include the meeting’s purpose and goals, participants’ roles and interests, relevant background, desired outcomes, known constraints, potential sensitivities, and clear instructions on format and tone. The more specific and complete the briefing, the more relevant and actionable the AI’s support will be.

A manager’s briefing clarity and thoroughness directly determine how useful AI responses are. Detailed context, clear objectives, and explicit boundaries enable AI to deliver tailored, relevant outputs. A well-crafted briefing reduces ambiguity, prompts clarifying questions, makes assumptions explicit, and provides a solid draft for review. This approach saves time, reduces cognitive load, and surfaces risks or gaps before they become issues. Treating the briefing as a key management task ensures robust, actionable AI contributions and more consistent, high-quality decisions.

The briefing book is a practical toolkit of prompts, review questions, and routines that managers can reuse, adapt, and improve. It is designed to foster deliberate, consistent, and responsible AI use, not to shortcut every workplace issue. The briefing book encourages managers to think critically about when and how to use AI, building organisational confidence and capability over time. It can include standard prompts for core managerial activities, tailored to common scenarios such as planning projects, initiatives, conversations, or work cycles.

Summarising reports, notes, policies, feedback, or meeting transcripts. This includes condensing lengthy or complex documents into concise overviews that highlight key decisions, unresolved issues, risks, and recommended actions. A well-structured summary helps managers quickly grasp what matters most and identify next steps without getting overwhelmed by detail.

Scanning for risks, assumptions, dependencies, and unanswered questions. For example, a manager can use AI to systematically surface potential risks in a project plan, identify information gaps, reveal underlying assumptions to test, and highlight critical dependencies that could affect timelines or outcomes. This proactive approach helps prevent issues from being overlooked until they become problems.

Preparing for meetings, presentations, performance discussions, and stakeholder conversations. AI can help develop meeting briefs, anticipate participant concerns, draft opening statements, and suggest fair ways to keep discussions focused and productive. For performance or stakeholder conversations, it can provide direct, respectful language tailored to organisational values.

Turning broad objectives into actions, milestones, owners, and review dates. This includes breaking down strategic goals into practical steps, assigning responsibilities by role, setting timelines, and establishing regular review checkpoints. Such structured planning improves accountability and makes progress easier to track.

Comparing options and identifying the likely consequences of each. AI can help outline the pros and cons of different choices, flag potential unintended consequences, and suggest criteria for evaluating alternatives. This structured comparison supports better, more transparent decision-making.

Improving the clarity, tone, and structure of a draft communication. AI can review emails, memos, or reports for professionalism, readability, and alignment with audience expectations. It can also suggest ways to simplify complex language, highlight key messages, and tailor the tone for sensitive topics.

Creating questions that encourage deeper thinking and better team discussion. For example, AI can generate open-ended questions that prompt critical reflection, uncover assumptions, or invite diverse viewpoints. These questions help managers lead more insightful and inclusive conversations.

Reviewing a proposed decision before communicating or implementing it. AI can help ensure recommendations are evidence-based, risks are considered, and the reasoning is clear. It can also suggest additional review questions to ensure decisions are fair, proportionate, and aligned with organisational standards.

Documenting the reasoning behind a decision and recording follow-up actions. This involves capturing not just what was decided, but why—the evidence, rationale, and perspectives considered. Recording clear follow-up actions, responsibilities, and review dates helps ensure accountability and enables future learning from past decisions.

Treat each prompt as a reusable starting point, not a rigid script. Managers should adapt the wording to their organisation, industry, team, risk level, and preferred communication style.

The Anatomy Of A Strong Prompt

A strong management prompt usually contains six elements, each serving a distinct function in guiding the AI:

• Role: Specify the perspective or function the AI should take, such as project adviser, meeting facilitator, risk reviewer, or communications editor. This ensures the output aligns with the manager’s expectations and the task’s context.
• Context: Provide the relevant background, including the team’s purpose, the current situation, prior attempts, relevant stakeholders, and any important constraints (e.g., budget, policy, deadlines). The more context supplied, the more tailored and relevant the AI’s response will be.
• Objective: Clearly state what the manager wants to achieve. Is the goal to solve a problem, prepare a summary, generate options, identify risks, or draft a communication? A well-defined objective focuses the AI’s effort and prevents vague, generic responses.
• Inputs: Supply the information the AI should examine, such as reports, background notes, meeting transcripts, or policy documents. Remove unnecessary personal or confidential details to protect privacy and keep the AI focused on what matters most.
• Boundaries: Set clear limits on what the AI must not assume, invent, disclose, or decide. For example, instruct AI not to infer motives, fabricate data, or make final decisions on sensitive matters. Boundaries prevent overreach and keep the AI’s contribution within safe and appropriate limits.
• Output: Specify the desired structure (e.g., headings, bullet points), tone (formal, neutral, supportive), length, intended audience, and any specific next steps. Being explicit about output requirements yields more actionable, usable responses.

Practical Example:

Instead of writing a vague prompt such as:

“Help me plan our next team meeting.”

A strong briefing might read:

“Act as a meeting-planning assistant. Help me prepare a 45-minute team meeting about delays in a cross-functional project. The meeting should identify the main causes of delay, clarify ownership, and agree on three practical next steps. Use the background notes below. Separate confirmed facts from assumptions, identify outstanding questions, and propose an agenda with approximate timings. Do not assign blame or infer motives. Present the response under the headings: purpose, agenda, questions, risks, decisions required, and follow-up actions.”

This expanded prompt gives AI a clear role, detailed context, a specific objective, relevant inputs, boundaries to prevent error or bias, and a defined output format, tone, and structure. By including all six elements, managers enable AI to deliver precise, safe, and practically useful results.

Standard Prompt: Planning

Managers can use a planning prompt when an objective is clear, but the path forward is uncertain. Effective planning prompts guide the AI to create structured, actionable plans rather than generic task lists by incorporating critical elements:

• Outcome: Clearly define the result you want to achieve. This gives direction and helps AI focus its recommendations.
• Context: Briefly explain the background—such as the current status, prior attempts, or any external factors influencing the work. This avoids irrelevant suggestions.
• Timeframe: Specify deadlines, milestones, or sequencing so the plan is realistically scheduled.
• People and Roles: List the teams or job functions involved. Avoid unnecessary personal details; focus on responsibilities and capabilities.
• Constraints: Identify limits such as budget, capacity, policies, technical requirements, timing, stakeholder expectations, or external dependencies. These help AI avoid impractical advice.
• Plan Structure: Ask AI to present the plan in sections: major stages, specific actions, dependencies (what must happen before something else can start), risks, suggested owners by role, decision points, and review dates. Ask it to identify assumptions and clarify which are fact versus speculation.
• Critical Questions: Instruct AI to finish by listing five key questions to address before finalising the plan. These might include: What resources are missing? Which milestones are most at risk? Are there regulatory or compliance issues? Who needs to be consulted before certain steps? What alternatives exist if a dependency fails?

This approach encourages the AI to produce a plan that goes beyond generic tasks, surfaces hidden risks and dependencies, and clarifies who is responsible for what and when. The manager should then test the plan against operational reality—checking the availability of suggested owners, the achievability of deadlines, unresolved decisions, and any legal, financial, safety, cultural, or people-related factors the AI may not have recognised. This final review ensures the plan is both actionable and robust.

Standard Prompt: Summarising

A summary should help a manager understand what matters without creating false confidence that it has preserved every detail. A strong summarising prompt does more than condense information—it guides the AI to extract, structure, and clarify what is most important for decision-making. To achieve this, the prompt should:

• Ask for a concise overview at the start, ideally in five sentences or less, to give managers a rapid sense of the main points.
• Request a breakdown of key facts, decisions already made, unresolved issues, risks, competing viewpoints, and recommended follow-up questions, each in its own section. This structure helps managers quickly scan for what they need.
• Instruct the AI to clearly distinguish between information directly stated in the source and any conclusions or inferences it draws, so assumptions aren’t disguised as facts.
• Explicitly prohibit adding information not present in the material, and require the AI to flag areas where the source is unclear, ambiguous, or internally contradictory.

Highlight the value of requesting multiple summary types from the same material: an executive summary for senior leaders, an action summary for those responsible for next steps, a question summary to clarify uncertainties, and a communication summary for broader sharing. This tailored approach ensures each audience gets relevant information in the right format and level of detail.

Remind managers to compare the summary with the original material—especially when sensitive topics, dissenting views, exceptions, or qualifications are present. Even a well-written summary can miss crucial nuance, so human review is essential to catch omissions or misrepresentations that could lead to poor decisions or misunderstandings.

These additional steps make the summarising prompt more robust, producing clearer, more actionable summaries that are less likely to mislead through omission or oversimplification.

Standard Prompt: Risk Scanning

A risk scanning prompt should help managers cast a wide net, surface hidden dangers, and avoid false confidence. To make the most of AI’s capabilities while guarding against its limitations, a strong risk scanning prompt should:

Specify the types of risks to consider—operational, financial, legal, privacy, security, safety, reputational, stakeholder, and people-related—so the AI’s scan is comprehensive, not superficial.

• Ask for a structured analysis of each risk: the possible cause (what might trigger the risk), likely consequence (if it happens, what is the impact?), early warning sign (how would you know it’s emerging?), existing controls (what’s already in place to manage it), and practical mitigation (what could reduce the likelihood or consequence?).
• Instruct the AI to rank risks only when the available facts support it, and to label uncertain cases for human assessment. This prevents the AI from presenting a guess as a factually grounded judgement.
• Require the AI to identify missing information and explicit assumptions, so managers understand the limits of the analysis.
• Prohibit the invention of incidents, regulations, evidence, or probabilities to keep the scan anchored in reality.
• Emphasise that a risk scan is a starting point for further investigation, not a substitute for formal risk assessment or expert review.

A useful follow-up prompt is to challenge the initial scan, encouraging a second look for blind spots and overlooked issues:

• Ask what important risks a busy manager might overlook.
• Probe for risks that seem minor but could have outsized consequences.
• Question the practicality of proposed controls—are they realistic given resources and constraints?
• Ask what evidence or new information could change the assessment of each risk.

This two-stage process prompts managers to think critically, examine their own blind spots, and use AI as a tool for structured thinking rather than a source of final answers. By layering these checks, managers can avoid overreliance on AI and ensure risk management remains thoughtful and robust.

Standard Prompt: Meeting Preparation

AI can support meeting preparation by helping a manager clarify the purpose, anticipate different perspectives, and structure the discussion for maximum effectiveness. To get the most value from AI, a meeting preparation prompt should:

• Clearly state the meeting topic, desired outcome, and participants’ roles. This ensures the AI understands the focus, goals, and relevant viewpoints.
• Provide concise, de-identified background information—such as recent events, decisions, or challenges that make the meeting necessary. This context helps the AI generate relevant talking points and anticipate potential issues.
• Ask for a structured meeting brief that includes: the meeting’s purpose, required decisions, key information to confirm, likely concerns from each participant or role, targeted questions to ask, potential areas of disagreement, and a proposed opening statement. This ensures the manager enters the meeting with a clear roadmap and is ready to address both planned and unexpected topics.
• Ask for recommendations to keep the conversation focused and respectful, such as setting ground rules, using neutral language, and giving each participant time to share their perspective. This is particularly valuable in meetings where disagreement or tension is expected.
• Instruct the AI to avoid inferring personal motives or making unsupported claims about individuals to reduce the risk of bias or misinterpretation.
• For difficult or sensitive conversations, the prompt can be extended to:
• Request sample language that is direct, calm, and respectful, focusing on observable behaviour or results rather than assumptions or labels.
• Ask for suggested open-ended questions that invite the other person’s perspective and encourage dialogue rather than confrontation.
• Explicitly prohibit diagnosing, labelling, threatening, or making unapproved promises about outcomes to keep the conversation fair and within policy.

Finally, remind managers that while AI can help structure and prepare for meetings, it cannot replace genuine human connection and professional judgement. Authenticity, organisational policy, and each conversation’s unique circumstances should always guide the final approach and wording.

The Guardrails

The most important guardrail is that AI output is a draft for review, not an authority. Treating AI-generated content as a starting point for human consideration—rather than a final answer—helps prevent errors, bias, and overreliance. NIST’s generative AI risk guidance emphasises the need to define roles and responsibilities for human-AI configurations and oversight. In practice, this means every organisation should clearly define what AI may assist with, what requires human approval, and what must not be delegated.

To operationalise these guardrails:

• Establish written policies clarifying which types of decisions AI can support (e.g., drafting, summarising, brainstorming), which require mandatory human review (e.g., hiring, promotion, performance assessment, legal advice), and which are off-limits to AI entirely (e.g., final disciplinary action, sensitive HR matters, regulatory submissions).
• Ensure an appropriately qualified human reviews every AI-assisted output before taking any action, especially where legal, financial, or reputational risk is present.
• Train managers and users to recognise the limits of AI tools, including their potential for hallucination, bias, and lack of context. Reinforce that users, not the AI, are accountable for outcomes.
• Require documentation of key decisions, including when AI was used, how outputs were checked, and who approved the final version.
• Implement technical controls where possible—for example, limiting access to certain data types, requiring double sign-off for high-impact uses, or using watermarking to identify AI-generated content.
• Encourage a culture of questioning, where staff challenge AI output, check for missing perspectives, and escalate concerns when in doubt.

By embedding these safeguards into workflows and culture, organisations can harness AI’s productivity while maintaining accountability, fairness, and trust.

Data Privacy

Data privacy is a foundational requirement for responsible AI use in management. Never enter personal, confidential, commercially sensitive, legally privileged, health-related, disciplinary, or security-sensitive information into an AI tool unless the organisation has specifically approved that tool and put robust controls in place for its use. This includes not only the input data but also any output that could be linked back to individuals or sensitive matters.

This is particularly important in Australia, where the Office of the Australian Information Commissioner (OAIC) makes clear that the Privacy Act 1988 and the Australian Privacy Principles apply to all uses of AI involving personal information. Inaccurate, incomplete, or artificially generated information about an identifiable person can also be considered personal information, meaning privacy obligations extend to both fact and fiction.

To Operationalise Privacy Safeguards, Managers Should:

• Use only AI tools that are formally approved by the organisation and regularly reviewed for compliance and security.
• Limit prompt information to only what is necessary for the task. Remove or generalise names, locations, specific dates, or case details wherever possible. For example, use “a team member in a customer-facing role” rather than “Sarah from the Sydney office.”
• Consider the data’s full lifecycle: who will access the prompt and generated output, how long the information may be retained, and whether it could be reused for training or other purposes.
• Ask: Does the task require a decision about an identifiable person? If so, apply the highest level of caution and consult with legal or privacy experts when unsure.
• Embed privacy questions into every AI workflow: Is this tool approved? Is this information necessary? Can it be de-identified or generalised? Who will see the output? What are the long-term risks?

For Additional Protection:

• Require a human to review any AI output containing personal information before use, especially when the output may inform decisions about people.
• Keep a record of when and how AI was used, and whether it included or excluded sensitive information.
• For organisations in New South Wales, pay attention to state-specific privacy laws and risks associated with generative AI tools, as these may place stricter requirements on public sector teams.

The OAIC recommends a proactive approach: embed privacy and security risk assessments into AI projects from the outset, establish clear lines of human oversight, and verify outputs before using them in decision-making. By treating privacy as an ongoing, shared responsibility—not a one-off compliance box—managers protect their teams, their organisations, and the people whose information they hold.

Hallucination Checks

AI can produce statements that sound confident but are inaccurate, incomplete, outdated, or unsupported—a phenomenon known as “hallucination.” Managers must take proactive steps to detect and address these risks before acting on AI-generated output.

To strengthen hallucination checks:

• Require the AI to distinguish facts from assumptions and to highlight areas of uncertainty. Prompts should explicitly instruct the AI to cite or reference its sources where possible, or to state when no source exists.
• Use layered verification prompts, such as:
• “Review your response and identify every statement that depends on an assumption, external fact, interpretation, or calculation. Label each item as confirmed, likely, uncertain, or unsupported.”
• “Recheck your response against the source material. List any claims that cannot be verified from the source. Do not repair missing information by guessing or inventing details.”
• Ask the AI to flag information that may be outdated or context-dependent, prompting a human reviewer to check for recent changes or external developments.
• Remind managers to independently verify all critical details—including names, dates, figures, calculations, quotations, and policy, legal, regulatory, and compliance claims. This is particularly important for:
• Information about employees, customers, suppliers, or other individuals
• Recommendations involving safety, health, finance, security, or employment
• Statements that could damage a person’s reputation or influence a formal decision
• Any content intended for publication, external use, or senior decision-makers
• Escalate scrutiny in proportion to the potential consequences of error. For high-impact decisions, use a step-by-step fact-checking process and seek input from subject-matter experts where appropriate.
• Document the verification process—note which facts have been cross-checked, which uncertainties remain, and what additional review steps are required before action.

By embedding these robust hallucination checks into routine workflows, managers can reduce the risk of acting on flawed or misleading AI output, protect organisational integrity, and reinforce a culture of critical thinking and accountability.

Human Review Points

Don’t leave human review until the very end. For robust governance and risk management, it should occur at multiple, clearly defined stages of the workflow:

• Before entering information: Assess whether the data you are about to provide is safe to share. This includes screening for personal, confidential, or sensitive content, and ensuring compliance with organisational policies and privacy laws. Consider whether the information is truly necessary for the task and whether you can de-identify, generalise, or redact it.
• After AI generation: Review the draft output for accuracy, completeness, tone, clarity, and relevance. Check that the AI has not introduced errors, made unsupported assumptions, or omitted important context. Scrutinise whether the draft aligns with the original briefing and objectives and is suitable for further consideration.
• Before deciding: Evaluate the fairness, proportionality, and potential consequences of acting on the AI-generated material. Consider whether you have explored alternative perspectives, properly assessed risks, and whether the recommendation aligns with ethical and legal standards. If the stakes are high, seek a second opinion or escalate for expert review.
• Before communicating: Review the message for audience appropriateness, clarity, sensitivity, and alignment with organisational tone and brand. Ensure that no private or protected information is inadvertently disclosed and that the message cannot be misinterpreted or cause unintended harm.
• After implementation: Post-action review is critical. Evaluate outcomes, identify what worked and what didn’t, and document lessons learned. This helps refine future prompts, improve workflow, and foster continuous learning.

Importantly, AI should never make final decisions about hiring, dismissal, promotion, discipline, performance ratings, workplace investigations, allocation of opportunities, or other matters that significantly affect a person’s rights or livelihood. In these cases, AI can assist by organising information or preparing questions. Still, the accountable human decision-maker must always examine the evidence, apply relevant policies, and make the final judgement. This preserves fairness, accountability, and trust within the organisation.

The Five-Stage Workflow

1. Brief

Define the outcome with precision before opening the AI tool. Explain the context, identify the audience, state any relevant constraints (deadlines, policies, sensitivities), summarise the source material, and specify the desired output format. Remove any unnecessary identifying or confidential information to protect privacy and compliance. At this stage, managers should also clarify the type of assistance needed—is the AI organising information, generating creative options, challenging assumptions, improving wording, or drafting a document? The more specific and focused the task, the more useful and relevant the AI’s response will be. Consider preparing a checklist of briefing elements to avoid overlooking anything.

2. Generate

Ask AI to produce an initial draft, set of options, meeting brief, summary, or risk assessment tailored to the task. Use the structured briefing from the previous step to guide the AI’s output. The primary purpose here is to expand thinking, reduce preparation time, and surface new angles or considerations. Treat the first output as raw material for review, not a finished product. Managers can—and should—ask for multiple alternatives or formats, fostering comparison and avoiding tunnel vision that comes from accepting the first plausible suggestion. This stage is iterative; managers may refine their brief and rerun the prompt if the initial result is off-target.

3. Critique

Critically review the AI-generated material, using both automated critique prompts and human judgement. Ask what key information is missing, what may be incorrect, which assumptions are unstated, and what objections or concerns stakeholders may have. Examples of effective critique prompts include: “Critique this response as a cautious senior manager. Identify unsupported claims, missing perspectives, risks of unfairness, impractical recommendations, and questions requiring human judgement. Suggest revisions only after listing weaknesses.” Supplement AI critique with a thorough human review—AI can highlight patterns and logic, but it lacks context on relationships, culture, and ethics. For high-impact decisions, consider peer review or input from subject matter experts.

4. Decide

Make the decision using only verified information, relevant policies, and sound professional judgement. Consult with others as appropriate, especially for decisions with wide-ranging or sensitive impacts. Use a decision framework to guide the process and ensure you answer key questions: What are we choosing? Why are we choosing it? What evidence supports it? What alternatives were considered? Who may be affected? What risks remain? What safeguards or review points are required? Who has authority to approve or modify the decision? This step is about accountability—AI may inform, but should never dictate, especially for critical or people-affecting matters.

5. Document

Thoroughly record the final decision, the rationale, the evidence and sources referenced, the responsible people or roles, the required actions, and a review date if applicable. Where AI was used, clearly note how it assisted and who reviewed and approved the output. Documentation ensures transparency, enables future learning, and creates a record that can be reviewed if similar situations arise. Consider using a standard template for decision documentation to maintain consistency and completeness across the team or organisation.

This five-stage workflow embeds best practices for safe, effective, and accountable AI use in management. Each stage provides a checkpoint for quality, fairness, and compliance, ensuring AI enhances decision-making without replacing experienced managers’ critical thinking, judgement, and oversight.

Building A Prompt Library

A shared prompt library turns isolated experimentation into a scalable, team-wide capability. It provides a single, accessible resource that consolidates, tests, and improves best practices over time. To build a robust and effective prompt library:

• Include tested prompts for all core management tasks, not just the prompt text, but also sample inputs, annotated sample outputs, required review questions, and clear warnings about inappropriate or risky uses.
• Organise the library by real-world scenarios and workflows—such as planning and prioritisation, project updates, meeting preparation and follow-up, risk and assumption reviews, stakeholder communication, coaching and development, decision analysis, documentation, writing and editing, and quality assurance/fact checking. Each section should offer context-specific templates, guidance, and examples.
• Attach to each prompt a concise note that explains:
• When to use the prompt
• What information to include and what to exclude (especially for privacy or compliance)
• What the output should not be used for (e.g., final HR, legal, or disciplinary actions)
• Who is responsible for the review and approval of the output
• Implement a regular review and versioning process. Test prompts in practice, and flag any that produce vague, biased, or impractical results for revision. Document changes and lessons learned so you capture and share improvements.
• Distinguish between general-purpose prompts and those specifically approved for sensitive or regulated data, internal systems, or unique organisational workflows. This reduces misuse and helps users select the right prompt for their needs.
• Provide a feedback channel—such as a shared comment log, regular review meetings, or a digital form—where users can suggest improvements, flag issues, and share learnings. This keeps the library up to date and responsive to evolving needs.

A well-maintained prompt library saves time and reduces errors while building consistency, fairness, and compliance into AI-supported management. Over time, it becomes a living asset that raises overall prompt quality and reinforces a culture of learning and safe AI use across the organisation.

The 30-Minute Briefing Clinic

A briefing clinic gives managers a hands-on, interactive opportunity to develop practical AI briefing skills in a safe, structured environment. The clinic is not about abstract AI theory; instead, it focuses on improving real-world management outcomes by refining how prompts are written, evaluated, and shared. To deepen the impact and learning, the clinic can be enhanced as follows:

• Begin with a clear explanation of session objectives, the approved AI tool, data privacy expectations, and the importance of understanding the difference between using AI for assistance versus delegation. Emphasise that managers remain accountable for outcomes.
• Demonstrate a weak prompt, then lead a discussion about why it fails—does it lack context, clear objectives, or boundaries? Highlight common pitfalls (e.g., vagueness, missing constraints, or risks of bias).
• Guide participants through strengthening the prompt by explicitly adding context, objectives, required constraints, audience, and output requirements. Invite them to suggest refinements and discuss how each element improves the AI’s response.
• Ask participants to adapt the improved prompt to a real but de-identified management scenario from their own experience. Encourage small-group work or peer feedback so participants benefit from diverse perspectives and practical examples.
• Run a group critique of the AI-generated outputs, using a structured review checklist: Does the output address the task? Are assumptions and uncertainties clear? Is sensitive information handled appropriately? What important perspectives or risks might be missing? How actionable and practical is the result?
• In the final minutes, capture the strongest revised prompts and add them to the shared prompt library, discussing how versioning and feedback will help the library improve over time. Also, agree on one concrete improvement or experiment to try before the next clinic to foster a cycle of continuous learning.

Reinforce that clinics should always use fictional or carefully de-identified scenarios to safeguard privacy and comply with organisational policy. Explain how these sessions can build a culture of critical thinking, safe experimentation, and prompt literacy across the management team.

By expanding the clinic in these ways, managers gain not only better prompts but also the skills and confidence to use AI responsibly and collaboratively in day-to-day work.

The manager’s review checklist

Before using AI-assisted work, the manager can ask the following, expanding each point for thoroughness and practical application:

• Is the output based on accurate and sufficient information? Double-check the sources, verify facts, and confirm the information you gave the AI was complete and up to date. If the data had gaps, clearly note any extrapolation or inference.
• Has the AI introduced facts, motives, figures, or conclusions not provided? Look for “hallucinated” details—information or reasoning that did not originate from your input or from authoritative, cited sources. Challenge anything not grounded in evidence.
• Are assumptions clearly identified? Review the output for any reasoning based on assumptions, implied context, or speculative statements. Call out each assumption and, where possible, validate it or flag it for further investigation.
• Have important alternative perspectives been considered? Check that the AI has surfaced or acknowledged relevant dissenting views, counterarguments, or stakeholder interests, rather than simply reinforcing a single viewpoint.
• Could the wording be unfair, discriminatory, misleading, or unnecessarily harsh? Review for tone, language, and framing. Remove or revise any statements that could be perceived as biased, exclusionary, or insensitive.
• Does the output expose personal, confidential, or commercially sensitive information? Scrutinise for privacy risks and compliance with organisational policies. Redact or generalise sensitive details as needed before sharing or acting on the output.
• Has the relevant policy, expert, or decision-maker been consulted? Cross-check the output against organisational policies or, for critical topics, seek input from subject matter experts or authorised decision-makers.
• Is the recommendation proportionate to the evidence? Assess whether the available data and analysis justify the recommendation’s strength and scope. Avoid overreaching conclusions or definitive statements on ambiguous matters.
• What happens if the output is wrong? Consider the risks and consequences of acting on incorrect, incomplete, or biased advice. For high-impact decisions, add extra review steps or run scenario tests.

What should you document before you act? Record the rationale, evidence, assumptions, decision-makers involved, and any limitations or review processes. Good documentation supports accountability and future learning.

This expanded checklist helps managers avoid a common pitfall: accepting professional-looking AI output at face value. Remember, clarity and sound reasoning—supported by critical review—matter more than polished formatting.

Creating The Top Five Prompts

The final exercise is to create and share your top five management prompts. This process helps individual managers clarify their approach while building a foundation for team-wide consistency, learning, and improvement. To make this exercise more practical and impactful, consider the following steps:

1. Identify the five management tasks you perform most often or find most mentally demanding—such as preparing a team meeting, turning objectives into a plan, summarising a project update, identifying risks, or drafting a difficult communication. Choose tasks that recur or have a significant impact on your work.

2. For each prompt, document a full set of details to guide safe, consistent, and effective use:

• The management task it supports (e.g., “Project planning for a product launch”).
• The precise context to supply to the AI, including relevant background, key objectives, constraints, and stakeholders.
• The information that must be excluded or de-identified for privacy and compliance—such as names, exact dates, confidential data, or sensitive business details.
• The exact output required: specify format, tone, level of detail, and any sections or headings needed (e.g., “Produce a summary with sections for decisions, risks, and open questions”).
• The review questions to ask afterwards, tailored to the task. For example, “Does the plan consider dependencies?” “Are all key facts in the summary supported by the source material?”, or “Have risks been ranked with evidence?”
• Situations where the prompt must not be used—such as legal, disciplinary, or high-stakes HR decisions, or any context involving sensitive or regulated data.
• The person or role accountable for checking and approving the result before any action is taken.
• Any additional notes or tips, such as how to adapt the prompt to different teams or projects, or reminders to update prompts as needs evolve.

3. A strong set of five prompts might include one for planning, one for summarising, one for risk scanning, one for meeting preparation, and one for critique. This covers a balanced routine: understanding the situation, planning the work, examining the risks, preparing the conversation, and challenging the draft before deciding. You can also add prompts for stakeholder communication, performance feedback, or other tasks central to your role.

4. Test your prompts on low-risk, real-world scenarios and refine them through regular use. Share successful prompts with your team and encourage feedback, so the library grows more robust and relevant over time.

When managers learn to brief AI clearly—and when they systematically build privacy, verification, critique, and strong human judgement into the process—they do more than save time. They help create a more consistent, thoughtful, and accountable way of working that benefits the whole organisation.

This Post is 19/20. Tomorrow’s post will be the last in this series: Innovation Leadership Principles