Best practices for writing AI prompts
10 min
By the end of this lesson, you’ll be able to:
- Recognize the key components of a great prompt
- Write effective AI prompts
- Use follow-up prompts to refine AI results
The basics of AI prompting
A prompt is a text instruction you give an AI to generate a response. Prompts can range from simple questions to detailed instructions or even code. The quality of the AI's response is directly tied to the quality of your prompt.
Prompt engineering is the process of designing and refining prompts to get the best results from an AI. It involves structuring your instructions clearly so the AI understands what you need. Like any skill, it takes practice to master.
Write a great AI prompt
Tailor the level of detail in your prompt to the task's complexity. For simple questions, use plain natural language as you would with a colleague. For complex tasks, add context and instructions so the AI can better support the workflow.
A well-crafted generative AI prompt follows TCREI framework, which stands for:
- Task: Define the outcome you are expecting from Rovo.
- Context: Provide helpful background information that can help Rovo understand the scenario.
- References: Include any Confluence pages, Jira work items, or links that are relevant to guide Rovo in generating its response.
- Evaluate: Assess Rovo’s response for accuracy and relevance. Identify if something is missing or is unclear.
- Iterate: Use Rovo’s response and your evaluation to refine your prompt. This is where you add more detail, clarify your request, or break it down into simpler requests.
👇 Here's an example of a great prompt that follows the TCREI framework.
In addition to the required components, several optional elements can enhance the quality of your AI prompts and improve the AI’s response.
Depending on the type and complexity of the task you’re asking the AI to perform, a good prompt may include the following additional elements:
- Format: Specify the desired format of the output.
- Tone: Define the writing style or mood of the AI’s response.
- Examples: Provide good examples to guide the AI.
👇 Click the boxes below to explore each optional element in more detail and see examples.
Prompting best practices
When crafting prompts for AI, following best practices can significantly improve the quality and relevance of the responses.
DO | DON’T |
|---|---|
✔️ Be specific and clear ✔️ Be concise and include relevant context ✔️ Use simple, natural language ✔️ Break highly complex tasks into a series of smaller prompts ✔️ Verify or test the AI results to ensure validity ✔️ Use follow-up prompts to iterate and refine the results ✔️ Specify what not to do if needed | ❌ Use vague or ambiguous instructions ❌ Overload your prompt with irrelevant information or details that can confuse the AI ❌ Include confidential or proprietary information in your prompt ❌ Use jargon, acronyms, and complex language ❌ Ask the AI to perform multiple jobs or a very complex job in a single prompt ❌ Blindly assume the output is error-free ❌ Start over with a new prompt if the results aren’t quite right |
Continue the conversation with follow-up prompts
If you’re not happy with the AI’s response, refine it with follow-up prompts instead of starting over. Prompting is an iterative process.
After the initial response, provide additional prompts to clarify, expand, or adjust the output until it fits your needs. Ask questions or add instructions based on the reply to get more accurate, relevant results.
Some tips for effective follow-up prompting:
- Be specific: Clearly state what needs to be changed or improved, such as the tone, format, or level of detail.
- Provide positive and negative critique: In addition to specifying what you’d like to see changed, mention what aspects of the initial output were good to ensure those elements are retained in the next iteration.
- Provide additional context or instruction: Offer more background information, context, or instructions to guide the AI in generating a more accurate response.
- Iterate as needed: Continue the process of reviewing and prompting until you achieve the desired result.
While AI can boost productivity, it can have limitations. AI can sometimes have hallucinations, or produce information that appears correct but is actually inaccurate or misleading.
Always manually validate AI-generated results.
👇 Click on the tabs below to see examples of how follow-up prompts can be used to refine AI outputs.
Resolve ambiguities or unclear aspects of the initial response.
👉 For example: "Can you provide more details on the current risks and how they are being mitigated?"