How to Give Clear and Effective Instructions to OpenAI Models

Learning how to write the best AI prompts for OpenAI models starts with understanding that some prompt formats consistently perform better than others due to how the models are trained.

The official OpenAI prompt engineering guide is usually the best place to start, but the rules of thumb below cover the patterns we see work most often.

---

How Prompt Engineering Works

Large language models infer intent from the structure of your input. Clear instructions, clean separation between instruction and context, and explicit output requirements make the model's job easier and the output more consistent.

---

Rules of Thumb and Examples

Note: {text input here} is a placeholder for real context.

1) Use the latest model

For best results, use the latest, most capable model that fits your latency and cost constraints. Newer models are typically easier to prompt.

Note: There are meaningful differences between reasoning models and GPT-style models. When in doubt, review model-specific prompting guidance.

2) Put instructions at the beginning of the prompt and use ### or """ to separate instruction and context

Less effective:

Summarize the text below as a bullet point list of the most important points.

{text input here}

Better:

Summarize the text below as a bullet point list of the most important points.

Text: """
{text input here}
"""

3) Be specific, descriptive, and as detailed as possible about the desired context, outcome, length, format, style, etc.

Domain-specific prompts like Ethereum Developer benefit most when you spell out scope, constraints, and security expectations.

Less effective:

Write a poem about OpenAI.

Better:

Write a short inspiring poem about OpenAI, focusing on the recent DALL-E product launch
(DALL-E is a text-to-image ML model) in the style of a {famous poet}.

4) Articulate the desired output format through examples

Strict-format prompts like Commit Message Generator show why explicit schemas reduce ambiguity.

Show, and tell - the models respond better when shown specific format requirements. This also makes it easier to programmatically parse out multiple outputs reliably.

Less effective:

Extract the entities mentioned in the text below. Extract the following 4 entity types:
company names, people names, specific topics and themes.

Text: {text}

Better:

Extract the important entities mentioned in the text below. First extract all company names,
then extract all people names, then extract specific topics which fit the content and finally extract general overarching themes

Desired format:
Company names: <comma_separated_list_of_company_names>
People names: -||-
Specific topics: -||-
General themes: -||-

Text: {text}

5) Start with zero-shot, then few-shot. If neither works, then fine-tune.

Zero-shot:

Extract keywords from the below text.

Text: {text}

Keywords:

Few-shot (provide a couple of examples):

Extract keywords from the corresponding texts below.

Text 1: Stripe provides APIs that web developers can use to integrate payment processing into their websites and mobile applications.
Keywords 1: Stripe, payment processing, APIs, web developers, websites, mobile applications
##
Text 2: OpenAI has trained cutting-edge language models that are very good at understanding and generating text. Our API provides access to these models and can be used to solve virtually any task that involves processing language.
Keywords 2: OpenAI, language models, text processing, API.
##
Text 3: {text}
Keywords 3:

Fine-tune: Use fine-tuning best practices when you need consistent performance across many similar tasks.

6) Reduce "fluffy" and imprecise descriptions

Less effective:

The description for this product should be fairly short, a few sentences only, and not too much more.

Better:

Use a 3 to 5 sentence paragraph to describe this product.

7) Instead of just saying what not to do, say what to do instead

Less effective:

The following is a conversation between an Agent and a Customer.
DO NOT ASK USERNAME OR PASSWORD. DO NOT REPEAT.

Customer: I can't log in to my account.
Agent:

Better:

The following is a conversation between an Agent and a Customer.
The agent will attempt to diagnose the problem and suggest a solution,
while refraining from asking any questions related to PII.
Instead of asking for PII, such as username or password, refer the user to the help article:
www.samplewebsite.com/help/faq

Customer: I can't log in to my account.
Agent:

8) Code generation specific: use "leading words" to nudge the model toward a pattern

If you are testing code-style nudges, Python Interpreter is a good way to verify concise, execution-ready outputs.

Less effective:

# Write a simple python function that
# 1. Ask me for a number in mile
# 2. It converts miles to kilometers

Better:

# Write a simple python function that
# 1. Ask me for a number in mile
# 2. It converts miles to kilometers

import

9) Use the Generate Anything feature

If you can describe the task or the expected output in natural language, Generate Anything can create a tailored prompt to start from. Use it as a draft and iterate.

---

Best Practices for Prompt Engineering with the OpenAI API

• Use roles intentionally: keep stable behavior in system or developer messages and put task-specific input in user messages.
• Separate instructions from context with clear delimiters and avoid mixing multiple unrelated tasks in one request.
• Enforce structure with explicit templates or JSON schema when you need reliable parsing.
• Control randomness with parameters: use low temperature for extraction or scoring tasks and higher values for creative work.
• Keep context tight: include only what the model needs, and summarize or retrieve long documents instead of pasting them wholesale.
• Add evaluation loops: test prompts with real examples, store prompt versions, and measure outputs before shipping.

Example message structure:

[
  { "role": "system", "content": "You are a concise assistant that follows the format exactly." },
  { "role": "user", "content": "Task: Summarize the text in 5 bullets.\\n\\nText: \\\"\\\"\\\"\\n{text input here}\\n\\\"\\\"\\\"" }
]

---

Parameters

Generally, the most commonly used parameters to alter output are model and temperature.

• model: Higher performance models are generally more expensive and may have higher latency.
• temperature: Higher values increase randomness and creativity. For factual extraction or QA, start near 0.
• max_completion_tokens: A hard cap on token generation. The model may stop earlier if it thinks it has finished.
• stop: A set of tokens that stops generation when encountered.

---

Next step

If you want ready-to-run prompt templates and structured frameworks, head to AI Prompt Studio and apply these best practices faster.