What is a prompt?
A prompt is the instruction you provide Clarity or any commercial chatbot to obtain the appropriate response or output you desire. You may want a draft of an email, a summary of a document, or suggestions of study questions to help students prepare for a quiz. The prompt is how you relay your requirements to an agent or chatbot.
You can think of the prompt as a recipe - the details you provide to the agent are the list of ingredients and cooking instructions.
However, the ingredients alone don’t determine the merit of the meal – the quality of those ingredients, and how they are used, directly impact the excellence of the result. Prompts work the same way – quality prompts produce quality responses. A quality prompt requires clear, specific instructions to get the best results. By following a few core principles, you can significantly improve the quality, relevance, and usefulness of the agent’s responses.
How would you like Clarity to respond?
One way to write an effective prompt is to break the prompt down into the following components:
Goal: What response do you want from the agent?
Context: Why do you need it and who is involved?
Source: Are there sources of information that the agent should include, or an example it should follow?
Expectations: How should the agent respond to best meet your expectations?
Sample prompt:
Goal: Help drafting an email.
Context: You had a lot of people apply for your fellowship program and you cannot accept them all.
Source: If the agent supports file uploads, you might upload a sample email for the agent to follow, or paste those details into your prompt.
Expectation: The email should express gratitude for the applicant’s interest and should have a kind tone.
PROMPT: “Our department recently received an overwhelming number of applications for our upcoming fellowship program. However, we cannot accept everyone. Please help me draft an email to send to those people who were not accepted. The email should express gratitude for their interest and should have a kind tone. Please use the email I have uploaded to this chat as a reference for your response.”
Other things to consider:
State your specific role when that information is relevant to the response you want to receive.
EXAMPLE: “I am a professor at Yale University and am currently brainstorming ideas for a lesson plan on [TOPIC]. Will you please assist me in generating class exercises for first year students?”
Provide specifics about how you would like the response formatted:
- “Provide 5 bullet points”
- “Write a 200-word summary”
- “Create a step-by-step process”
Provide enough context:
- Explain the background situation
- Mention your audience or purpose
- Include any constraints or requirements
Iterate and revise: If the first response from the chatbot isn’t perfect, you can pose follow-up questions or ask for specific changes like “Can you make the tone more formal?” or “That’s a good start, but can you focus more on [Specific Aspect]?”
Pay attention to the kinds of revisions you routinely ask of the agent, because this will also help you improve the initial prompts you enter.
Best Practice: Start a New Conversation
We recommend starting a new conversation when you shift to a different topic, task, or project. This helps ensure optimal performance and relevant responses from the agent.
Why this matters:
- Improved accuracy: Starting a new conversation allows the agent to focus on your current topic without being influenced by unrelated previous exchanges.
- Optimal performance: Shorter, focused conversations help the agent provide more precise and helpful responses. Longer conversations may result in decreased accuracy and unexpected issues.
When to start a new conversation:
- Changing subjects (e.g., switching from analyzing literature to writing code)
- Beginning a new assignment or project
- Addressing a different course or research area
- When the conversation becomes lengthy or unwieldy
- If responses seem less relevant or the AI appears to misunderstand your current context
When continuing is fine:
- You’re building on the same topic or task
- Following up on previous responses with clarifications or refinements
- Working through multiple steps of the same project