Prompt engineering best practices: Top 10 tips

Prompt engineering is all about writing effective instructions to get the most accurate, helpful, and relevant outputs from AI models. As AI models become central to more workflows, knowing how to use them effectively becomes a game-changing advantage.

We’ve created this prompt engineering guide to help you write better queries, structure your prompts, and guide the AI toward your desired results. You’ll learn how to lead with clear instructions, use strategic prompt formatting to your advantage, provide examples, and iterate your requests.

From assigning personas to setting output formats, each prompt engineering tip we feature will improve your interaction with language models. Additionally, we’ll explore best security practices when using AI, showcase top prompt engineering tools and libraries, and discuss how to apply these methods in real-world AI development.

Whether you’re a beginner or an aspiring prompt engineer, this guide will be beneficial for getting practical insights and prompt examples to help you get the most out of AI.

1. Put instructions at the beginning of the prompt

Leading with instructions helps AI models understand your intent before processing any context or data. This sets a clear direction for the response.

So, whenever you’re prompting AI, make sure to put the task you want to be done first. Then, you can follow up with details or context.

❌ Bad example:

Hey, GPT. This is an email I wrote last night. Can you revise it? 

✅ Good example:

Revise the following email to sound more professional.

2. Use delimiters

A clean prompt structure often relies on delimiters to separate your instructions from the content. Whether you use quotation marks, triple backticks, hashtags, or XML tags, the goal is the same: reduce ambiguity.

Delimiters help the model understand where your instructions stop and the referenced data begins, making your prompting more effective and less prone to misinterpretation.

❌ Bad example:

Revise the following email to sound more professional, but don't change anything else about it: I wanted to check if you had a chance to review the document I sent last week. Let me know what you think.

✅ Good example:

Revise the following email to sound more professional, but don't change anything else about it.

<email>I wanted to check if you had a chance to review the document I sent last week. Let me know what you think.</email>

3. Be very specific

Think of any AI model as someone on their first day doing their job. AI and large language models (LLMs) specifically have no context of what they should do unless you specify the task when prompting.

Spell out precisely what you want from AI and define the format, tone, or any constraints. The more concrete your request, the better the result.

❌ Bad example:

Write a blog post about prompt engineering.

✅ Good example:

Write a 500-word blog post about prompt engineering tips. The audience will be beginners, so use a friendly tone and give concrete examples for easier understanding.

4. Give the model a persona

You can improve your prompts simply by assigning the AI model a role or persona. It helps tailor the response style and depth, making outputs feel more aligned with a specific point of view or expertise level.

❌ Bad example:

Explain blockchain.

✅ Good example:

<role>You are a tech educator who specializes in explaining complex topics in simple, relatable language. Your audience is made up of adults with no technical background – teachers, business owners, and everyday readers curious about new technology.</role>

<task>Write a 500-word blog post explaining what blockchain is, how it works at a basic level, and why it matters. Avoid jargon terms.

Use a friendly, engaging tone with real-world analogies to keep the explanation easy to follow and interesting.</task>

5. Provide relevant examples

Including examples in the prompt format gives the AI model a reference point to mimic. In other words, including an example helps guide both the tone and structure of the response, which is especially useful for creative or complex tasks.

❌ Bad example:

Write a product description for a new smartwatch.

✅ Good example:

<instruction>Create a product description for a new smartwatch with these specifications: fitness tracking (steps, heart rate, sleep monitoring), GPS navigation, smartphone notifications, music control, contactless payments, and health monitoring (ECG, SPO2).

Use the following example as a style guide.</instruction>

<example>Stay fit and connected with our sleek fitness tracker that monitors your heart rate, tracks your steps, and syncs seamlessly with your devices.</example>

6. Ask the model to explain the chain of thought

Requesting a step-by-step explanation encourages the AI model to reason through a problem rather than jump to a conclusion. This prompt structure is especially beneficial for analysis, logic, or decision-making tasks.

❌ Bad example:

What’s the best marketing channel for a small bakery?  

✅ Good example:

Help me choose the best marketing channel for the small bakery that I just started. First, list your best picks and explain why they’re good, divide them into <best-picks> and <least-recommended> XML tags. After that, give your final answer.

7. Specify the desired output format

Telling the model how to structure its response ensures consistency and usability. Whether you need a list, table, code snippet, or sections with headers, spell it out clearly when prompting.

❌ Bad example:

List SEO tips for beginners.

✅ Good example:

List five beginner-friendly SEO tips in bullet points. For each tip, include a short explanation to help readers understand why it matters.

8. Supply the AI with relevant data

If the model needs to reference specific facts, figures, or text, include that data directly in the prompt. Don’t assume it has access to current or personal information; instead, provide what it needs to work with.

❌ Bad example:

Summarize our Q1 sales performance.

✅ Good example:

Summarize the following Q1 sales data in 3 bullet points:  

“Q1 revenue: $120k, up 15% YoY. Top product: SmartLamp. Customer growth: +8%.”

9. Ask for evidence

AI can sometimes hallucinate, making claims that are false or unsupported. The good news is that there are strategies to reduce this.

You can ask the AI to say “I don’t know” when it’s unsure. Or, prompt it to explain its reasoning before coming up with a final answer. This encourages more accurate, evidence-based responses.

❌ Bad example:

Who is the wealthiest man in the world?

✅ Good example:

Who is currently the wealthiest person in the world? Provide their most recent net worth and source (Forbes, Bloomberg, etc.) 

If up-to-date data is unavailable, say “I don’t know.”

10. Prompt in iterations

There are no perfect prompts that can guarantee an ideal output from AI on the first try. So, treat prompting as a conversation – start simple, review the output, and refine your prompt based on what’s missing or unclear. This iterative approach leads to better results.

If responses still fall short, consider implementing prompt tuning techniques.

How to ensure security in prompt engineering?

When we use AI to simplify our tasks, we often need to feed our proprietary data to the AI model. Without proper safeguards, this data can be exposed, misused, or leaked, especially when working with third-party models or cloud-based systems.

But don’t worry – there are ways to protect your data when using AI. Here are some prompt engineering security best practices you can follow:

• Use data masking. Hide your real data values by replacing them with fictional but structurally similar data. For instance, replacing a credit card number with 1234 5678 9101 2345 or with XXXX XXXX XXXX XXXX keeps the format intact while protecting sensitive information.
• Pseudonymize data. Remove or replace personal identifiers with placeholders. For example, write a customer name like Jane Doe as User A to lower the risk of data exposure.
• Generalize data. Broader categories reduce identifiability. Instead of listing exact ages (like 18, 19, 20), you can group them as 18-20. This makes it harder to trace data back to an individual while still preserving analytical value.
• Data swapping. Switching the values between records can obscure patterns. For instance, swapping customer zip codes between rows keeps statistical integrity but masks individual identities, which is useful when training models that don’t need exact matching data.
• Use synthetic data. Synthetic data is generated by algorithms to mimic real datasets without exposing actual user information. For example, AI can create realistic health records for model training, ensuring privacy while maintaining data quality.
• Audit and log AI interactions. Track how AI is used within your organization to detect misuse and ensure compliance with internal policies.
• Implement role-based access. Some AI tools can be used collaboratively across a team to increase productivity. Limiting access and interactions based on user roles minimizes exposure risks and maintains control over sensitive data.

Are there any tools and resources to make prompt engineering easier?

There are many tools available to help you test and optimize prompts for LLMs. Mainly, they fall into two categories: general-purpose prompt engineering tools and code-based libraries. Let’s take a closer look at both.

Best prompt engineering tools

Prompt engineering tools are typically no-code or low-code platforms that let users experiment with prompts. They are ideal for non-developers or rapid prototyping.

The following are some of the best prompt engineering tools you can try out:

• PromptPerfect. A tool that automatically optimizes prompt quality to help you achieve consistent and high-quality results from LLMs. It can refine prompts for both text and image models with features like prompt comparison and reverse prompt engineering.
• Promptist. An AI-powered assistant developed by Microsoft that helps craft better prompts for image generation models like Stable Diffusion. It provides prompt templates and a user-friendly visual editor to simplify prompt creation and refinement.
• OpenAI Playground. An interactive web-based tool from OpenAI that lets users experiment with OpenAI’s language models in real-time. It’s great for testing, tweaking, and refining prompts directly with immediate feedback.
• FlowGPT. Example of a prompt engineering tool that’s also a community platform for sharing, discovering, and curating AI prompts. It has a visual interface for AI interaction and encourages open-source collaboration.

Prompt engineering libraries

Prompt engineering libraries are code-based frameworks that give developers programmatic control over prompt design and interaction. They are especially valuable in professional or research environments, where they help streamline development, support advanced prompt architectures, and make AI integration more maintainable.

A common use case for libraries is integrating LLMs into software applications. Using their preferred library, developers can manage prompts dynamically, automate workflows, and maintain consistency across different tasks.

Most libraries also come with reusable templates or functions for tasks like summarization, classification, or translation. This not only speeds up development but also reduces the chance of errors, particularly in large-scale projects where efficiency and standardization matter.

Two examples of prompt engineering libraries are LangChain, which allows developers to chain prompts, tools, and data sources in a structured workflow, and PromptTools, which enables batch testing, side-by-side comparison, and evaluation of prompt performance across various models and metrics.

How to apply prompt engineering tips in real life?

In a nutshell, here’s how you can apply prompt engineering in everyday use:

✅ Start with clear instructions. AI models aren’t human, so there’s no need to worry about sounding too direct. Get straight to the point, state the task clearly, and avoid adding unnecessary fluff.

✅ Use structure and examples. If you want a response in a particular format, such as a table, list, or code snippet, specify it in the prompt. If possible, show a concrete sample.

✅ Iterate as you go. Don’t try to get the perfect result in one shot. Start small, review the output, then adjust your prompt for better results.

✅ Protect sensitive data. If you’re using real-life information (like work documents), apply data masking or pseudonymization to avoid exposing personal or confidential details.

You can use these prompts to simplify many of your tasks with any AI models, from writing and planning to coding and design. Additionally, as more businesses adopt AI to streamline workflows, the ability to improve prompts is becoming a valuable skill in the job market. With effective prompts, tasks that once took hours can now be completed in minutes.

Also, with AI, anyone can create a web app or a website, which is an essential asset for building a strong online presence. By implementing the prompting tips shared in this article, you can guide AI in generating functioning code or content tailored to your needs.

Hostinger Horizons is a beginner-friendly AI tool that lets anyone create a web app without writing a single line of code. Getting started with Hostinger Horizons is as simple as iterating your prompts to build the perfectly functioning web app you want. It works in a real-time sandbox environment, so you can preview changes instantly before publishing.

You can also upload images like logos or product shots directly into the editor to personalize the design. Just drag and drop or use the upload button, and the AI will use those visuals to match your brand style more accurately.

Implementing prompting tips is just as powerful for more advanced users, not just beginners. If you’re a developer, try harnessing AI coding tools, such as GitHub Copilot, Tabnine, or Claude, to generate, refactor, or debug code efficiently.

Additionally, with multiple other AI-powered platforms, there are many different ways you can streamline complex development workflows, automate repetitive tasks, and even prototype entire applications with greater speed and accuracy.

How to become a good prompt engineer?

As AI becomes central to creative work and the rise of vibe coding makes building digital products easier without coding skills, writing effective prompts is becoming a key skill, not just for better AI output, but for working faster every day.

A good prompt engineer always starts by giving clear instructions when prompting, defining the task, format, tone, or purpose. Sending vague requests like “Write a blog post” will lead to a generic response based on the AI model’s assumptions.

Remember, LLMs will only give the response that you want if you provide the context of the task. The more details you describe in your prompt, the better the output.

Avoid common mistakes like mixing instructions and content, sending overly broad requests, or skipping formatting guidelines. Instead, apply the prompting tips from this guide to improve precision, speed, and quality.

Ultimately, the best way to become a strong prompt engineer is through hands-on practice. Test different approaches, refine your prompts, and iterate often. If you have any questions or additional prompt engineering tips, feel free to share in the comments below.

All of the tutorial content on this website is subject to Hostinger's rigorous editorial standards and values.