How to design a great prompt

There are a lot of ways to use LLMs. The information below is adapted from OpenAI’s guide on designing a prompt to best work with their system.

Each of these elements can help create an excellent prompt, but when used together they give your LLM a framework for the current project you’re working on. Pick and choose which elements to include to best train the AI for you, but keep them in this order.

1. Role and Objective

Define what the model is and what it’s trying to do – if the model knows what the goal is, it’ll be more focused.

You are a helpful tutor improving my writing skills.
Your goal is to provide feedback on the samples I give you highlight ways to improve without doing it for me.

2. Instructions

High-level behavioral guidance. Be specific: what to do, what to avoid. Include tone, formatting, and restrictions.

Always respond concisely and professionally.
Avoid speculation, just say “I don’t have enough information” if unsure.
Format your answer using bullet points.

3. Sub-Instructions (Optional)

Add focused sections for extra control. Examples:

Sample Phrases:
Use “Based on the document…” instead of “I think…”

Prohibited Topics:
Do not discuss politics or current events.

When to Ask:
If the input lacks a document or context, ask:
“Can you provide the document or context you’d like summarised?”

4. Step-by-Step Reasoning / Planning

Encourage structured thinking and internal planning.

“Think through the task step-by-step before answering.”
“Make a plan before taking any action, and reflect after each step.”

5. Output Format

Specify exactly how you want the result to look.

Respond in this format:
Summary: [1-2 lines]
Key Points: [10 Bullet points]
Conclusion: [Optional]

6. Examples (Optional but Powerful)

Show GPT what “good” looks like.

# Example
## Input
Source 1 reveals a great deal about the lack of knowledge people had about disease and medical practices. Even by the 16th century, over 300 years after the outbreak of the Black Death, bloodletting was still a regular practice, as shown in this woodcut. The woman receiving treatment likely trusted this medical professional to cure her of an ailment (which is not disclosed) and likely had little to no understanding of the procedure herself. The woodcut is, however, a valuable documentation of part of this procedure, allowing historians to better understand the ways in which medical professionals engaged with their patients. If it is part of a larger series of prints, such as a manual or guide for medical students, then it can also reveal a great deal about the ways in which medical professionals were taught during this period and the lessons they had learned from the outbreak of plague in the 13th century.

## Output
Summary:
This paragraph analyses a woodcut depicting bloodletting in the 16th century, focusing on historical medical understanding and its educational value.

Key Points:

 > Clarify the source’s title or context upfront to ground the analysis.
 > Avoid vague phrases like “reveals a great deal”; be specific about what is revealed.
 > Be cautious with speculation (“likely trusted”, “likely had little understanding”)—consider qualifying or balancing these statements.
 > Maintain consistency in referencing time periods (the Black Death began in the 14th century, not the 13th).
 > Refine sentence structure—some are overly long and could be split for clarity.
 > Use more precise academic language (e.g., “bloodletting persisted” instead of “was still a regular practice”).
 > Address the purpose and limitations of the source more directly—what can and can’t it tell us?
 > Consider integrating historiographical insight—how have historians interpreted such sources?
 > The final sentence introduces a new idea (training manuals) without sufficient context or evidence—expand or trim.
 > Maintain a more formal tone throughout; “this woodcut” could be rephrased as “the source” for academic style.

Conclusion:
With stronger structure, clearer evidence, and more precise language, this paragraph can provide a sharper and more compelling source analysis.

7. Final Instructions

Repeat key parts at the end to reinforce the model’s behavior, especially in long prompts.

“Remember to stay concise, avoid assumptions, and follow the Summary → Key Points → Final Thoughts format.”

8. Bonus Tips from the Guide

• Put key instructions at the top and bottom for longer prompts
• Use Markdown headers (#) to structure input
• Break things into lists or bullets to reduce ambiguity
• If things break down, try reordering, simplifying, or isolating specific instructions