ChatGPT - Prompt Engineering Tips

This article explains how to write clear, structured prompts that help ChatGPT produce more useful, accurate, and appropriately formatted responses.

What Is Prompt Engineering?

Prompt engineering is the practice of writing clear, structured instructions that help ChatGPT produce the response you want.

ChatGPT generates its output based on the information and instructions you provide. The more precise and intentional your input is, the more useful and relevant the result is likely to be.

You do not need a technical background to write effective prompts. The basic principles are straightforward:

• Be specific
• Provide relevant context
• Describe the task clearly
• Explain what kind of output you expect

The Fundamentals

Be Specific

One of the most effective things you can do is be specific about what you want. Vague prompts usually produce vague results.

Vague:

Tell me about enrollment.

Specific:

Summarize fall 2025 enrollment trends, broken down by college. Focus on year-over-year changes and highlight any colleges with enrollment growth or decline greater than 5%.

The second prompt defines:

• Scope: Fall 2025 enrollment
• Structure: Results broken down by college
• Focus: Year-over-year changes
• Threshold: Changes greater than 5%

This level of detail is more likely to produce a useful response.

Assign a Role

Telling ChatGPT what role it should take can help define the tone, depth, and perspective of the response.

This is especially useful when you need domain-specific expertise or a particular communication style.

Examples:

• "You are an experienced academic advisor. A first-generation college student asks you how to choose a major. Explain the factors they should consider and the questions they should ask themselves."
• "You are a technical writer. Rewrite the following error message so that a non-technical user can understand it."
• "You are a grant reviewer for the NSF. Evaluate the following abstract and identify its strengths and weaknesses."

Define the Output Format

If you have a specific format in mind, include it in your prompt. ChatGPT can usually match the structure you describe.

Examples:

• "Respond in a numbered list."
• "Format the response as a table with columns for Name, Department, and Status."
• "Write this as a professional email with a subject line."
• "Give me a one-paragraph summary followed by three bullet points with key takeaways."
• "Respond in plain language at an eighth-grade reading level."

Without format instructions, ChatGPT will choose its own structure, which may not match what you need.

Provide Context

ChatGPT does not automatically know your internal systems, institutional processes, team preferences, or the details of your situation.

The more relevant context you provide, the more useful the output is likely to be.

Instead of:

Write a response to this complaint.

Try:

I work in a university registrar's office. A student emailed because their transfer credits have not yet appeared in their degree audit. Our policy is that transfer credit evaluation takes four to six weeks after official transcripts are received. Write a professional and empathetic response that acknowledges the student's frustration, explains the timeline, and directs them to the appropriate transfer credit contact.

This gives ChatGPT the:

• Role
• Situation
• Policy
• Tone
• Information to include

Break Complex Tasks Into Steps

For multi-part tasks, divide the request into sequential steps instead of asking for everything at once.

This gives ChatGPT a clearer path to follow and reduces the chance that an important part of the task will be skipped.

Instead of:

Create a presentation about our new employee onboarding process.

Try:

I need to build a presentation about our new employee onboarding process. Work through it in these stages:

1. Suggest an outline with 8 to 10 slide titles covering the full onboarding journey from offer acceptance through the 90-day check-in.
2. After I approve the outline, write the content for each slide.
3. Then suggest talking points I can use as speaker notes.

This approach gives you checkpoints where you can review the work and redirect ChatGPT before it continues.

Use Examples

If you want ChatGPT to match a specific style, tone, or format, provide an example.

This is one of the most effective ways to show ChatGPT what you expect.

Example:

Here is an example of how we write knowledge base article summaries:

Multi-Factor Authentication adds a second layer of security to your university account. After entering your password, you will be prompted to verify your identity using an approved authentication method. MFA is required for eligible university users.

Now write a similar summary for our VPN service.

ChatGPT can use the example to match the length, tone, and structure more closely.

Specify What to Include and Exclude

Telling ChatGPT what not to do can be as important as telling it what to do.

Examples:

• "Explain how DNS works. Do not use analogies. Describe the technical process directly."
• "Summarize this research paper. Focus only on the methodology and findings. Do not include the literature review or future work sections."
• "Write a job description for a systems administrator. Do not include salary information."
• "List the advantages and disadvantages of migrating to a cloud-hosted learning management system. Do not recommend a specific product."

Set Constraints

Constraints help keep responses focused and appropriately scoped.

Examples:

• "Explain this concept in three sentences or fewer."
• "Limit the response to 200 words."
• "Provide exactly five recommendations."
• "Use only information from the uploaded document. Do not use general knowledge."

Ask for a Step-by-Step Explanation

For tasks involving calculations, analysis, or problem-solving, ask ChatGPT to explain the process in clear steps.

Example:

A student has 45 transfer credits and needs 120 credits to graduate. They want to complete the degree in two years while taking a full course load. Show the calculations step by step and explain what course load would be required each semester.

This can make the response easier to review and help you identify incorrect assumptions or calculations.

Iterate and Refine

You do not need to write a perfect prompt on the first attempt.

Treat ChatGPT as a collaborative tool. Start with a reasonable prompt, review the response, and refine it as needed.

Useful follow-up prompts:

• "This is good, but make the tone more formal."
• "Shorten this to half the length."
• "Add a section about accessibility requirements."
• "Rewrite the third paragraph to focus more on cost savings."
• "This is too technical. Simplify it for a non-technical audience."
• "Try again, but structure it as a frequently asked questions section."

Iteration is a normal and effective part of working with ChatGPT.

Use Instructions in Custom GPTs and Projects

If you repeat the same instructions across many conversations, consider adding them to a Custom GPT or a Project.

For example, someone who frequently writes knowledge base articles could use instructions such as:

You are a technical writer for UMS IT. Write in plain language, use active voice, preserve the full source content, and follow the UMS knowledge base formatting standard. Use clear section headings and include an Environment section at the end.

Conversations within that Custom GPT or Project can follow those instructions automatically.

Provide Reference Material

When ChatGPT has access to the documents relevant to your task, it can produce more accurate and grounded responses.

You can upload materials such as:

• Policy documents
• Style guides
• Data files
• Templates
• Examples of previous work
• Drafts that need revision

Tell ChatGPT how to use the uploaded material.

Example:

I uploaded our department's style guide and the previous annual report. Use the style guide's formatting standards and match the tone of last year's report. Draft the executive summary for this year's report using only the data I provide.

Ask for Multiple Options

When you are exploring ideas or choosing between approaches, ask ChatGPT to provide several options.

Examples:

• "Give me three subject lines for this email, ranging from formal to conversational."
• "Suggest five possible titles for a workshop on data privacy in higher education."
• "Write two versions of this announcement: one for students and one for faculty."

This gives you alternatives to compare and can help clarify what you prefer.

Use ChatGPT to Improve Your Prompt

If you are unsure how to phrase a request, ask ChatGPT what information it needs.

Examples:

• "I want help writing a grant proposal, but I am not sure what information you need. What questions should I answer before we start?"
• "Here is my prompt: [paste prompt]. How could I improve it to get a more useful response?"
• "What additional context would help you provide a better answer about our network architecture?"

ChatGPT can often identify missing details that would improve the request.

Common Mistakes to Avoid

Being Too Vague

A prompt such as "Help me with a report" gives ChatGPT very little information.

Include the topic, audience, format, length, and purpose when those details matter.

Asking for Too Much at Once

A prompt that asks ChatGPT to research a topic, write a long report, create an executive summary, and build presentation slides may produce weaker results across all parts of the task.

Break large tasks into stages.

Accepting the First Response Without Review

ChatGPT can produce confident-sounding content that contains errors, leaves out important details, or misunderstands your intent.

Review the response critically and refine it when needed.

Not Providing Enough Context

ChatGPT does not know your institution's policies, your team's preferences, or the details of your project unless you provide that information.

When a response feels generic, additional context is often the solution.

Assuming ChatGPT Remembers Every Conversation

A new conversation may not include context from previous chats.

If earlier information is important, provide it again or work within a Project that contains the necessary files and instructions.

Using the Same Prompt Style for Every Task

Different tasks require different approaches.

A prompt that works well for drafting an email may not work as well for:

• Data analysis
• Technical troubleshooting
• Research
• Brainstorming
• Document revision

Adjust the prompt structure to match the task.

Prompt Checklist

Before sending an important or complex prompt, review the following checklist:

• Role: Did I explain what role or perspective ChatGPT should use?
• Task: Is the task clearly stated?
• Context: Did I provide the relevant background information?
• Format: Did I specify how the response should be structured?
• Constraints: Did I define limits on length, scope, or content?
• Examples: Would an example help communicate my expectations?
• Exclusions: Did I explain anything ChatGPT should avoid?

Not every prompt needs all of these elements. A simple question may not require a role, example, or specific output format.

For important or complex tasks, including the relevant elements will usually produce better results.

Environment

• University of Maine System
• ChatGPT Edu
• Prompt writing
• Custom GPTs
• Projects
• Supported desktop and mobile web browsers