木工快进场,自己画的全屋柜子草图来了!大家给点建议

Clear communication is a fundamental skill that helps us connect, collaborate, and avoid misunderstandings. This is just as true when interacting with AI, where a well-defined framework for our instructions, known as prompts?,can significantly impact the usefulness and accuracy of responses.

But crafting great prompts isn’t always straightforward. How do we ensure AI understands and responds effectively? How can we design prompts that lead to meaningful, engaging interactions?

At Salesforce, we’ve found that applying structured communication principles – combined with creativity – can vastly improve agentic AI??-generated responses. One powerful approach is the SPEAKING model, a framework that helps ensure prompts are clear, context-aware, and effective.

The Art and Science of Prompt Design

Designing prompts is both an art and a science:

• The art lies in creativity – writing prompts that encourage AI to respond with clarity, empathy, and relevance.
• The science involves structure – leveraging linguistic principles, testing different approaches, and refining prompts through iteration.

For example, Agentforce? relies on language analysis algorithms to detect conversational?? cues (the science). However, what makes it truly effective is how we design its prompts – the wording, tone, and structure that shape a thoughtful, human-like experience (the art).

By blending these two elements, we create prompts that go beyond just being functional; they become seamless, engaging, and intuitive.

Design Better Prompts with the SPEAKING Model

Inspired by linguistic research, the SPEAKING model (Hymes, 1974) helps us craft prompts that feel natural and purposeful. 

Let’s explore each element, illustrating its practical application with the Agentforce example: designing prompts for an AI-powered tool that assists Salesforce users with troubleshooting.

Setting 

Where and when is the prompt used? (e.g., customer support chat, voice assistant, internal team tool)

• Agentforce Example: Users ask for help on Salesforce Help?.

Participants

Who is interacting? (e.g., new users, experienced developers, customer service agents)

• Agentforce Example: A mix of experienced admins and first-time users.

Ends

What’s the desired outcome? (e.g., resolving a technical issue, retrieving account details)

• Agentforce Example: Users should receive clear, accurate guidance to resolve their issue.

Act Sequence

What’s the order of the conversation? (e.g., greeting → clarifying the request → response)

• Agentforce Example: The AI first confirms the issue, asks for clarification if needed, and provides a step-by-step solution.

Key

What’s the tone and style? (e.g., formal, friendly, concise, empathetic)

• Agentforce Example: The tone is supportive, professional, and easy to understand.

Instrumentalities

What’s the medium? (e.g., text, voice, multimodal interface)

• Agentforce Example: Users interact via a chatbot or voice assistant.

Norms

What are the communication expectations? (e.g., quick responses, detailed explanations)

• Agentforce Example: The AI should provide clear, concise responses—no unnecessary jargon.

Genre

What type of interaction is it? (e.g., multi-turn chat, command-based prompt, FAQ-style response)

• Agentforce Example: This is a multi-turn troubleshooting conversation, where the AI guides users through a solution step by step.

By keeping these factors in mind, we ensure that AI interactions feel natural, effective, and aligned with user needs.

Test, Evaluate, and Refine Your Prompts

Creating great prompts isn’t a one-time task – it’s an ongoing process of testing and refinement.

Here’s how you can improve prompt effectiveness:

1. Start simple: Write clear, user-friendly prompts and test them with real users.
2. Analyze feedback: Are responses accurate? Do users find them helpful? Look for patterns in interactions.
3. Refine wording: Avoid vague or overloaded prompts. Clear, concise phrasing enhances comprehension.
4. Iterate constantly: AI evolves, and so should your prompts. Regular updates ensure relevance.

To give you an idea of how prompt refinement works, let’s look at this example:

Before: “Tell me more about your issue so I can better understand and provide relevant guidance.”

• Why it’s unclear: Users may not know what details to provide.

After: “What error message do you see? Or briefly describe what’s not working.”

• Why it’s better: It gives specific, actionable guidance, leading to faster resolution.

By continuously refining prompts, we improve AI responsiveness, reduce misunderstandings, and create more seamless interactions.

Ready, Set, Design

AI is only as good as the instructions we give it. By applying the SPEAKING model and following a structured approach to prompt design, you can create AI interactions that are clear, effective, and user-friendly.

Want to dive deeper? Learn more with Prompt Fundamentals on Trailhead and start building better AI experiences today.