Better Prompts, Better Public Service: A Practical Framework for Leaders Using AI

AI is not a mind reader. A vague or incomplete request often produces a generic, incomplete or poorly targeted response. Structured prompting requires the leader to clarify the problem, the audience, the constraints and the desired outcome before asking AI to respond. 

This can also create an opportunity for what the late psychologist and Nobel Prize winner Daniel Kahneman called “slow” thinking. Instead of reacting immediately, a leader can use the prompting process to examine assumptions, identify missing information, compare alternatives and sharpen the decision question. AI can support that reflection, but it should not replace professional judgment. Leaders remain responsible for protecting sensitive information, checking facts, applying organizational policy, and considering ethical and equity implications.

A simple way to structure an effective prompt is CRAFT: 

Context: Explain the situation. Include the organization’s purpose, the problem, relevant facts, stakeholders, constraints and any prior decisions the AI must understand. 

Role: Identify the expertise or perspective the AI should use, such as a budget analyst, program evaluator or public-engagement advisor. 

Action: State the task with a direct verb: analyze, summarize, draft, critique or recommend. If the work is complex, divide it into stages instead of placing several unrelated tasks in one prompt. 

Format: Describe the desired output — an executive summary, briefing memo, table, agenda or presentation outline. Include a word limit or required sections when helpful. 

Tone: Name the audience and desired voice: concise for an executive team, accessible for residents, empathetic for clients, or neutral and evidence-based for a policy audience. 

CRAFT is a starting point. Ask the AI to identify missing information or assumptions, then review the first response, add context, challenge weak reasoning and request revisions. 

CRAFT is easiest to understand in practice. Consider a county health department seeking resident input on clinic hours. The comparison below shows that a useful prompt is not simply longer; it clarifies purpose, stakeholders, deliverables, constraints, audience and uncertainty. 

Weak prompt: “Write a community engagement plan.” 

Stronger prompt: “Act as an experienced public-participation consultant. Our county health department is considering changes to evening clinic hours and wants input from residents who rely on public transportation, people with disabilities, working parents and clinic staff. Develop a 90-day engagement plan that identifies stakeholder groups, three accessible engagement methods, responsible staff, a timeline, risks and measures of success. Present the plan as a one-page executive summary followed by a table. Use clear, neutral language for department leaders. Do not invent local facts; identify information we still need.” 

The weak prompt leaves AI to guess the organization, audience, timeline, stakeholders and success measures. The stronger prompt makes those choices visible. It cannot guarantee accuracy, but it produces a response that is easier to assess, verify and revise. 

Many prompting mistakes resemble the habits of an ineffective supervisor: vague priorities, missing context, too many assignments at once and unstated expectations. Common mistakes include being too broad, assuming the AI understands the organization, ignoring the audience and tone, and accepting the first response without review. 

Before using an output, ask: Are the facts accurate and current? Can important claims be verified with authoritative sources? Did the response confuse fact, inference and recommendation? Are legal, privacy, accessibility or equity concerns missing? Treat the output as a draft that requires accountable human review. 

In his book “Stick with It: A Scientifically Proven Process for Changing Your Life — for Good,” Sean Young’s SCIENCE model offers a useful starting point: Take small steps, learn with peers, connect the practice to important goals, reduce friction, treat AI as support, keep experimentation engaging and engrain effective uses in regular workflows. 

Begin with one low-risk task, such as turning nonconfidential meeting notes into a draft summary. Compare the result with your own work, refine the prompt and save approaches that consistently help. Over time, move to more complex tasks while maintaining appropriate safeguards.  

Effective prompting is not about surrendering leadership to AI. It is about giving better instructions, asking better questions and using human judgment to turn a fast draft into responsible public value. 

Todd Suddeth, founder and director of Equity Leadership Consulting, is a senior lecturer for the John Glenn College of Public Affairs, where he also conducts leadership training and professional development. A career and personal coach, he has also provided consultation to higher education institutions with a specific emphasis on cultural centers and student engagement. Suddeth previously served in multiple leadership and administrative roles in various departments at The Ohio State University. He has a bachelor’s degree in psychology from the University of Akron and a master’s degree in public affairs and a doctoral degree in higher education and student affairs at The Ohio State University.