How to Write a Prompt: Effective Techniques

01What Is a Prompt and Why Does It Matter?

A prompt is the instruction, question or task description you give an AI model. Assistants such as ChatGPT, Claude or Gemini produce their answers based on the text you write. Even when the model stays the same, the quality of the answer changes a great deal when the quality of the prompt changes.

The AI does not know you, does not know your business, and has to guess what you want. When you give incomplete information, it fills the gaps with the most likely but most ordinary options. The heart of the question how to write a prompt is that you fill those gaps yourself.

This discipline is called prompt engineering. It may sound technical, but at its core it is the skill of writing a good brief: whatever you pay attention to when briefing an agency, an intern or a designer, you need to explain to the AI in the same way.

02The 5 Parts of a Good Prompt

Thinking of an effective prompt as a five-part template makes your work easier. You will not need all five for every task, but the more parts you make clear, the more accurate the answer becomes.

03Giving a Role and Context

Giving a role determines the expertise and voice the model speaks with. Phrases such as “as an SEO specialist” or “as an editor with the attention to detail of a certified accountant” change the focus and terminology of the answer.

Context describes the reality in which the role will operate: the type of your business, your target audience, the market you are in, and the standout features of your product. A request without context returns a generic answer; rich context returns an answer made for you.

If your brand voice is clear, adding it to the context makes a big difference. Details such as how you address readers, the words you like and avoid, and a description of your tone are already defined in a brand management project; you can give the same guide to the AI.

04Showing an Example and the Output Format

One of the strongest ways to tell the AI what you want is to show an example. Sharing a product description you like, an email tone or a table structure is more effective than long explanations. This technique is called few-shot prompting.

Specifying the output format is also the most neglected step. A bullet list, a table, a 150-word paragraph, JSON? If you state the format up front, you will not waste time reformatting the answer.

Write 3 product descriptions in the tone of the example below.

Example:
"Handmade olive oil soap. Additive-free, simple and gentle on your skin. Brings the scent of the Aegean to your bathroom."

Products:
1) Lavanta kokulu sabun
2) Bay laurel oil soap
3) Goat milk soap

Format: At most 3 short sentences per product.

05Advanced Prompt Techniques

Once you have the basic template in place, you can use some advanced techniques for more complex work. They make a difference especially in analysis, planning and long-form content production.

Making it think step by step

For work that requires calculation, comparison or a decision, ask the model to explain its approach first and then answer. A structure like “First list the evaluation criteria, then score each option against them” reduces hasty, superficial answers.

Splitting the task into parts

Instead of asking for a long blog post in one go, ask first for the outline, then for each section separately. This is called prompt chaining. You stay in control at every step and catch mistakes early.

Using delimiters

Use headings, dashes or tags to separate the instruction, the document and the examples from one another. When you paste in a long text, this prevents the model from confusing the instruction with the content.

Asking it to critique itself

After the first answer, saying “List the weak points of this text and write a corrected version” noticeably raises quality in the second round. Asking the model to ask you questions also surfaces missing information.

06ChatGPT Prompt Examples

The following ChatGPT prompt examples work on the same logic in Claude and Gemini. Fill in the places in square brackets with your own details. We covered which assistant stands out at which task in our ChatGPT, Claude and Gemini comparison.

Social media plan

Drafting the plan with AI saves time, but visual quality, community management and measurement still require human effort. We handle this whole in our social media management service.

Role: You are an experienced social media manager.
Context: We run a [business type] in [City]. Our target audience is [audience].
Task: Prepare an Instagram content plan for the next 2 weeks.
Format: Table; columns: day, content type (Reels/post/story), topic, short caption.
Constraint: At most 4 posts per week; keep the sales language light.

SEO content outline

However good the outline, what determines rankings is original information, technical infrastructure and the right keyword strategy. For that, SEO and analytics work should be your foundation.

Role: You are an SEO-focused content editor.
Task: Draft a blog post outline for "[keyword]".
Format: H1, 6-8 H2s, 2-3 content notes under each H2, 4 FAQ questions at the end.
Constraint: Do not use made-up statistics; flag information you are unsure about as needing verification.

07Common Prompt Mistakes

The skill of writing effective prompts starts with knowing what does not work. The mistakes below are the most common causes of weak answers.

• One-word requests: instead of “Write campaign copy,” say for whom, for what and in which tone.
• Cramming several tasks into one sentence: number the tasks or split them into separate prompts.
• Relying only on negative instructions: instead of “Don't write long,” say “at most 100 words.”
• Treating the first answer as final: improve it with feedback.
• Including confidential or personal information in the prompt: mask customer names and contact details.
• Using the answer without verifying it: even when the model sounds sure, it can be wrong.

08Building a Prompt Library for Your Business

If your team uses AI, collecting the prompts that work well in one place saves a lot of time. Prepare approved templates for recurring tasks such as product descriptions, customer replies, campaign emails and report summaries.

This library keeps your brand voice consistent and helps new team members become productive quickly. If hundreds of product descriptions will be written on your store's e-commerce platform, a single good template sets the quality across the whole catalog.

At a more advanced stage, these templates can be built into custom software solutions that connect AI to your own systems through an API. That way the team only enters the necessary information instead of writing a prompt each time, and the system handles the rest. For Google users, the Gems feature we describe in our What is Gemini article offers a similar capability.