How to write an LLM prompt

LLM prompts are commands used to guide large language models (LLM) to precise answers. Well-formulated prompts can improve the quality of results and help you work more efficiently with generative AI. Approaches like CARE are useful for designing clear and precise LLM prompts.

What are LLM prompts?

LLM prompts are instructions or input given to a large language model (LLM) in order to generate certain answers or actions. They can include questions, tasks and context information in written or spoken language, as well as images and other data. The quality and structure of the prompt play a big role in how precise and useful an answer you get from the AI model. Formulating precise LLM prompts is called LLM prompt engineering and aims to take full advantage of the possibilities of generative AI.

What are some best practices for writing LLM prompts?

LLM prompting is essential for efficiently interacting with artificial intelligence, as poorly formulated instructions might not return the kind of answer you’re looking for. For the best results, make sure to pay attention to the following things when writing an LLM prompt:

• Understand the abilities of the AI model: Knowing the strengths and weaknesses of the LLM you’re using (as well as its training data) can help you to adapt a prompt to the capabilities of that AI.
• Formulate LLM prompts precisely: Unclear prompts usually result in imprecise or ambiguous answers. Clear and precise prompts ensure that the AI model can correctly interpret the task and deliver targeted results. It’s also important to keep LLM prompts concise and to use the same tone in the input as you’d like in the output.
• Provide context: Background information makes it easier for AI to understand the prompt. Providing clear context will significantly increase the relevance and accuracy of the output. If you provide additional sources, it’s also useful to specify which information the AI model should consider.
• Optimise prompts step by step: You’ll often need to adjust an LLM prompt to get the result you’re looking for. If the first prompt you try doesn’t produce the right output, modify it based on the answer you got or try out different prompts. -Use neutral formulations: Leading questions can sometimes influence a model’s answer. Make sure to formulate LLM prompts neutrally to avoid any extra bias in the results.
• Clearly define the role of the AI model: If you assign the AI a role, you’ll get more relevant results. Giving it a specific role allows you to tailor the context and get targeted answers.
• Use LLM prompt templates: Use tried-and-tested prompt templates and adapt them to your individual needs. You can find numerous LLM prompt examples for a variety of uses online.

How to optimise LLM prompts using the CARE approach

There are several frameworks for using large language models (LLMs) effectively. One popular method is the CARE formula, a simple system designed to help create highly effective LLM prompts. CARE is an acronym that stands for:

• Content
• Ask
• Rules
• Examples

Content

Define the subject matter or theme you want the AI to focus on. Be as detailed as possible when describing the context or background information. Strong content helps the AI understand exactly what domain or situation it is working within, improving the relevance and quality of its response.

Ask

Frame a clear and specific request. Instead of giving vague commands like ‘Write something about marketing’, be direct, such as ‘Create a social media post promoting a new product launch in the tech industry’. A strong ask gives the AI a defined goal, which leads to more targeted and useful outputs.

Rules

Establish any guidelines the AI should follow when generating the response. This might include setting the tone (e.g., formal or casual), formatting instructions (e.g., use bullet points or write in paragraphs), language restrictions (e.g., UK English), or output length (e.g., 150 words maximum). Rules act like boundaries that keep the AI’s creativity aligned with your expectations.

Examples

Provide sample outputs or model responses that match the style, tone, or structure you want. Examples serve as concrete references that guide the AI more precisely than abstract instructions alone. The more relevant and clear your examples are, the easier it is for the AI to mirror the desired result.