A Practical Guide to AI Communication
As Wisconsin’s Polytechnic University, we provide hands-on, applied learning that prepares students for real-world challenges. As artificial intelligence (AI) and generative artificial intelligence (GenAI) reshape industries, understanding how to effectively communicate with AI systems is a valuable skill that can enhance productivity, creativity, and problem-solving across disciplines.
Whether you are a student, researcher, educator, or industry professional, learning how to engage with AI effectively can accelerate your work, enhance critical thinking, and open new opportunities for innovation. This guide provides a practical approach to prompt engineering, helping you build confidence in working with AI and leveraging its capabilities to support your goals.
What is Prompt Engineering?
Prompt engineering is the skill of designing clear, structured inputs that enable AI tools to generate accurate, useful, and relevant outputs. Crucially, it is also the ongoing process of reviewing the AI response and refining your inputs to guide continued success.
AI models, such as ChatGPT, do not think like humans. They generate responses based on patterns in vast amounts of data. Therefore, your ability to communicate effectively with AI technology directly determines the quality of the responses you receive.
What is the Prompting Cycle?
The prompting cycle is the iterative process of engaging with an AI to continuously improve the output.
Input → The initial prompt you provide to the AI
Output → The AI generates a response based on your input
Review → Critically evaluate the AI's response for accuracy and relevance
Revise → Refine your original prompt to improve the next output
How to Write Effective Prompts
Combine your answers into one clear prompt using bullets to stay organized. Review the prompt from a bird’s eye view and consider adding additional information. Skip any element that does not apply. When you are done, paste it into the AI tool. After the AI responds, use the CRAB key at the bottom to evaluate what you get back.
01
Role
Tell the AI who it is acting as and what expertise it should apply.
02
Task
State exactly what you want done and why it matters.
03
Context
Provide the relevant background, audience, data source, definitions, and situation.
04
Format
Define structure, length, tone, and level of detail.
05
Constraints
Tell the AI what to include, exclude, avoid, or prioritize.
06
Evidence & Assumptions
Tell the AI how to handle facts, uncertainty, citations, and missing information.
The CRAB Framework for Evaluation
AI tools produce fluent, confident, well-formatted text. Fluency is not accuracy. Every output is a draft that needs a human check before it becomes your work. CRAB is a four-part check you can run in under a minute. It reflects how we teach evaluation across the university, applied, repeatable, and useful in a real workflow rather than in theory.
Before you start:
Never paste protected or sensitive information into a tool that has not been approved for it. Data handling is a separate decision you make before prompting. CRAB evaluates what comes out.
Credibility
Can you trace where this came from?
AI systems generate believable sources as easily as real ones. Citations, statistics, and quotations are all fair game for invention.
Open the link. Confirm the author, publisher, and date exist. If the model cannot point to a source, treat the claim as unsupported.
Relevance
Does this answer the question you asked?
Models will respond to a request even when they have misread or misinterpreted it. The result looks complete and lands next to your goal instead of in alignment.
Reread your original prompt, then reread the output. Cut what does not serve the task. If the gap is large, revise the prompt and start again in a new chat rather than editing the output.
Accuracy
Do the specifics hold up?
Numbers, dates, names, technical steps, and policy details are the highest-risk elements in any AI output and the most likely to be wrong in ways that pass a casual read.
Verify every specific against a source you trust. In your own discipline, look for a traditional error you would catch in a student or apprentices work.
Balance
Whose perspective is missing?
AI outputs reflect patterns in training data. They tend to present a mainstream position as settled, flatten disagreement, and skip the counterargument.
Ask what a credible critic would say. If the topic is contested, the output should say so.
Scale the check to the stakes:
A brainstorm needs a light pass. Anything published, sent to students, submitted to a committee, or attached to your name needs all four. You own what you send, regardless of what produced the first draft.
