Comprehensive guide to prompt engineering in Persian

What Makes a Good Prompt?

Language models are not mind-readers. The clearer, more complete, and more structured your request is, the better response you will get. Remember a simple rule:

If you gave this exact request to a smart but newcomer colleague, would they understand precisely what you want?

If the answer is "no," make the prompt more complete.

Six Components of a Complete Prompt

Component	The Question It Answers	Example
Role	From what perspective should the model think?	"You are an experienced Persian editor"
Context	What is the background of the matter?	"This text is for a dental clinic website"
Task	What exact task needs to be performed?	"Suggest three engaging titles"
Audience	Who is the output for?	"For parents with young children"
Format	What should the output look like?	"In a two-column table"
Constraint	What is forbidden or required?	"Maximum 100 words, no medical jargon"

You do not always need to write all six components, but the more important the task, the more components are necessary.

An Example: From Weak to Strong Prompt

Weak:

Write a text about exercise.

Strong:

You are a fitness coach writing for a general audience.
Write a 200-word blog post on the topic
"Five simple exercise moves for office workers who sit at a desk all day".
- Each move in a short paragraph with how-to and its benefit
- Friendly and motivational tone
- At the end, a short safety advice

The second prompt specifies the audience, exact topic, length, structure, and tone; consequently, the output is usable with almost no editing.

Professional Techniques

1. Provide Examples (Few-shot)

If you want a specific style, instead of describing it, show it. Put two or three examples of the desired output in the prompt and say "continue in this same style."

2. Ask It to Think Step-by-Step

For logical, mathematical, or decision-making problems, write: "First, analyze the problem step by step, then give the final answer." Reasoning models do this themselves; read more about them in this guide.

3. Separate Data from Instructions

When you provide text for summarization or translation, place it inside specific markers:

Summarize the text inside the quotation marks in three sentences:
"...text..."

This prevents the model from confusing potential instructions inside the text with your command.

4. Give the Model Permission to Say "I Don't Know"

A phrase like "If you do not have enough information, say so instead of guessing" is one of the most effective ways to reduce hallucination.

5. Iterate, Don't Start Over

The first response is usually not the best. Instead of writing a new prompt, provide specific feedback within the same conversation: "The second paragraph is too general; add a real example" or "Make the tone more formal."

6. Request the Output Format Precisely

If the output is going to be used elsewhere, state its format explicitly: a table, a numbered list, JSON, or "only the final text without extra explanation."

Special Persian Writing Tips

• If you want natural text, write: "Fluent and natural, like something written by a Persian speaker, not a translation."
• For formal texts, ask the model to observe zero-widths spaces (نیم‌فاصله) and punctuation marks.
• For specific technical terms, clarify whether you want the Persian equivalent or the original Latin.
• If the quality of the Persian response is poor, sometimes writing instructions in English and asking for a "response in Persian" yields a better result.

Common Mistakes

1. Too short prompt: "Write an article" almost always yields a generic and lifeless result.
2. Multiple unrelated requests in one message: Ask for each important task separately.
3. Blind trust in numbers and sources: Always verify statistics, dates, and quotes.
4. Entering confidential information: Never enter passwords, banking info, and customer data in public chatbots. Read the Security and Privacy Guide.

Practice

The best way to learn is by practicing with real examples. In the Prompt Library, you will find dozens of ready-made Persian prompts, and in the Prompt Lab, you can see how each model responds to the same prompt.