When exploring prompt engineering best practices, we learned a lot from the Anthropic’s Prompt Guidelines.
This authoritative guide offers valuable insights into crafting effective, clear, and structured prompts to improve AI outputs. Many of the recommended strategies serve as general best advice for any large language model, including ChatGPT — although they are designed with Claude AI in mind.
We discovered that with some adjustments, these techniques can be effectively applied to ChatGPT as well.
In this blog post, we’ll discuss the prompt engineering best practices for ChatGPT inspired by Anthropic’s strategies. There are 9 key recommendations.
Table Of Content
1. Crafting Effective Prompts
The hardest part in getting AI models to work effectively is in crafting the right prompts. Good prompts are crucial because they directly impact the quality of the responses you receive from AI tools.
Anthropic attempts to address this by offering a “Prompt Generator” that automatically creates initial templates based on prompt engineering best practices. Although ChatGPT doesn’t offer similar convenience (unless we’re talking about third-party plugins), there are two strategies you can use to ensure your prompts work great:
(a) Follow a Prompt Framework
A prompt framework provides a structured approach to crafting prompts. It ensures that you include all the key elements for a good prompt.

An example of a simple framework is the TAP method: Task, Audience, Purpose.
- Task: Clearly state what you want ChatGPT to do.
- Audience: Specify the intended audience for the response.
- Purpose: Explain why the task needs to be done or what the response should achieve.
Your prompt might look like this when you apply this framework:
Summarize the following article in 150 words. The summary should provide a quick overview of the article's main points for readers who want a brief understanding of the topic, intended for a general audience with no prior knowledge of the topic.
Are you able to identify which parts of the sentence correspond to Task, Audience, and Purpose respectively?
(b) Ask ChatGPT to Check and Improve Your Prompts
Meet the often-overlooked prompt engineering master — ChatGPT itself!
You can use ChatGPT to expertly refine your prompts. For instance, you might ask:
How can I improve this prompt to get more accurate results?
or
What’s missing in my prompt that could make the output better?
ChatGPT would provide suggestions to tweak your prompts, add necessary details, or rephrase your request for clarity.

And do you know you could even ask it to write prompts for you? Just share your requirements and request it to ask clarifying questions. It will produce top-notch prompts for you.
2. Be Clear and Direct
The second technique recommended by the Anthropic’s Prompt Guidelines is “be clear and direct”.
Clarity and directness in your prompts are key no matter which AI models you’re working with. Clear prompts help the AI understand exactly what you’re asking for, and thus minimize the chances of vague, irrelevant, or incorrect responses.
“Clarity and specificity” appears at the top of our list of best practices for prompt engineering in an earlier blog post. To illustrate why it’s so important again, we would like you to compare what kind of results you should be expecting if one issues the following two prompts to ChatGPT:
(a) A vague prompt that doesn’t provide guidance for ChatGPT to frame its answer:
Explain the function of a cat's whiskers.
(b) A more specific prompt to get ChatGPT to simplify its responses, probably because the asker doesn’t want to wrap his head around complex scientific concepts:
Explain the function of a cat’s whiskers in simple terms for a 10-year-old.
3. Use Examples (Multishot Prompting)
Examples are a powerful tool in prompt engineering because they guide ChatGPT by providing clear references for the type of response you want.
This technique, known as multishot prompting, involves embedding one or more examples in your prompt to ChatGPT. It is especially helpful when you find it difficult to craft a prompt to fully capture your requirements in words. Just show ChatGPT what you want, instead of struggling to tell it.
To illustrate, let’s say you are a writer who’s seeking ChatGPT’s help to develop an article on cat care. You could include examples in your prompts to make sure ChatGPT is on the same page with you, like so:
Suggest 10 tips to keep a cat clean that every average cat owner can do without much effort. For example: trim the cat's nails regularly, use shampoo made specifically for cats, and keep the cat's teeth clean with dental treats.

Providing examples is a powerful prompt engineering best practice. We’ve covered the “Zero-Shot, One-Shot, and Multi-Shot” prompt engineering technique in a separate article. Be sure to check that out.
4. Let the AI Think (Chain of Thought Prompting)
Chain of Thought (CoT) prompting compels ChatGPT to work through tasks step-by-step and break down complex problems into manageable parts. Instead of shooting a direct response to you, the AI takes the extra step to relook at the process it follows to arrive at the answer. This increases the opportunity for it to provide you with a more thoughtful response.
What’s more, ChatGPT will explain every step it takes in a detailed and logical manner, so you can scrutinize them to spot any potential errors.
Here’s how it works. Add this ‘magic phrase’ at the end of your main prompt: “Let’s do it step-by-step” or “Let’s go step-by-step”. For example:
Explain to me the best modes of transport to get from London to Edinburgh. Let's do it step-by-step.
We have a blog post that explains in detail how “Chain of Thoughts” prompting works too! Take a look.
5. Use XML Tags
Anthropic’s guidelines suggest using XML tags like <instructions>
, <example>
, and <context>
to guide Claude’s outputs more precisely. While ChatGPT doesn’t natively require or respond specifically to XML tags, you can still use them to visually separate and clarify different sections of your prompt.
Let’s use back the example under “Crafting Effective Prompts” to illustrate. This time, we added tags to the prompt to clearly separate the 3 elements in the TAP framework:
<task> Summarize the following article in 150 words. </task>
<purpose> Provide a quick overview of the article's main points for readers who want a brief understanding of the topic.</purpose>
<audience> General audience with no prior knowledge of the topic.</audience>
6. Give ChatGPT a Role
The next technique is also one of the prompt engineering best practices we recommend in various places across this website: giving ChatGPT a role.
Ask ChatGPT to tackle your request as a Math teacher, experienced car mechanic, successful business owner, or any role that corresponds to your scenario. This way, it will respond with relevant information and style, mimicking the specialized knowledge and skills associated with those roles to serve you better.

Like what the Anthropic prompting guidelines suggest, when you request the AI to act in a specific role, you turn it “from a general assistant into your virtual domain expert”.
Here’s an example to show you how this works:
Imagine you are an experienced car mechanic with more than 5 years of experience. My car engine is making a loud knocking noise when I accelerate, and it gets worse the faster I go. What could be causing this issue?
We have an article that provides an overview of the top 10 prompt engineering techniques. “Role-Playing and Persona” is one of the recommended techniques.
7. Prefill the AI’s Response
This is the practice of starting the AI’s response with a few words or phrases to guide its output in a specific direction. Unfortunately, this technique is not applicable to ChatGPT, so we won’t discuss it.
8. Chain Complex Prompts
If you have a complex task for ChatGPT, instead of cramming everything into a 1000-word prompt, it’s better to break the task down into smaller and more “digestible” parts and prompt ChatGPT sequentially. This is exactly what the technique “Chaining complex prompts” is about.
When you apply this technique, you guide the AI through each step of the task, ensuring that it handles each part methodically and thoroughly before moving on to the next step. Repeat this process until the task is accomplished.
“Chain Complex Prompts” is best suited for tasks that involve multiple steps, detailed reasoning, or require structured problem-solving. You need to design your sequence of prompts carefully before you apply it.

Let’s ask ChatGPT to act as a car mechanic again so that we can demonstrate this technique. Your first prompt might be:
Imagine you are an experienced car mechanic with more than 5 years of experience. List common reasons why a car might not start.
After you get the list from ChatGPT, your next prompt will be:
For each reason listed, provide a brief troubleshooting step.
Followed by:
Explain what to do if the car still doesn’t start after checking these common issues.
9. Long Context Tips
At the end of Anthropic’s prompting guidelines, several tips on handling long-context prompts were provided. These include:
(a) Prioritize the Most Important Information
Place the most critical information at the beginning of your prompt. Since ChatGPT pays the most attention to the start of the prompt, this helps it focus on the most relevant details.
(b) Break the Content into Smaller Chunks
Similar to what we’ve discussed under “Chain Complex Prompts”, it’s not necessary (or advisable) to stuff all the information into a single prompt. Length documents or instructions should be broken into manageable chunks and prompted to ChatGPT sequentially.
The above steps will ensure ChatGPT handles long-context instructions properly without losing information.

Conclusion
Give these prompt engineering best practices a try, and see how they enhance your interactions with ChatGPT.
We’d love to hear about your experiences—what worked well for you, and what challenges you faced. Share your results and tips in the comments below. Happy prompting!
No Comment! Be the first one.