Tech expert Tim Stackpool shares guidelines for creating the perfect AI prompt. For Executive Assistants, AI is here to stay. Good results start with instructions.
Remember the first time you had to train a new employee? How carefully you chose your words, provided context, and set clear expectations? Well, welcome to AI prompting – it’s surprisingly similar, only this time, the “trainee” is remarkably capable but takes everything literally. Very, very literally.
When PricewaterhouseCoopers (PwC) started hosting “prompting parties” for their staff, they weren’t celebrating punctuation. They were acknowledging a crucial truth: the way we ask AI for help directly impacts the quality of help we receive. So, it’s time to share the secrets of crafting the perfect prompt, along with some real-world examples.
| The CRAFT Formula: Your new best friend
After testing countless approaches, take a look at what’s called the CRAFT formula for foolproof prompting: Context: Set the scene. Role: Define the AI’s perspective. Audience: Specify who it’s for. Format: State how you want it presented. Tone: Indicate the desired style. |
Let’s see the CRAFT Formula in action with a typical real-world example. Imagine you need to draft a difficult email to a client about a missed deadline.
Bad prompt: “Write an email about a late project.”
Better prompt using CRAFT:
“Context: We’re two weeks behind on delivering the Q1 report due to data verification issues. Role: Act as a senior account manager with 15 years of experience. Audience: Writing to our biggest client’s CFO, who is generally understanding but needs detailed updates. Format: Professional email with clear next steps. Tone: Confident and solution-focused while showing appropriate concern.”
See the difference? The first prompt might get you anything from a stern telling-off to a casual “sorry we’re late!” The second prompt consistently delivers a professional, contextually appropriate response.
The psychology behind perfect prompting
Dr Sarah Martinez’s 2023 study at Stanford University revealed that AI models perform 40% better when given clear context and role definitions. “It’s like giving someone a map instead of just a destination,” she explains. “The more context you provide, the more likely you are to reach your desired outcome.”
Real-world examples: The good, and the bad.
Example 1: Planning an office party
Vague prompt: “Help me plan an office party”
Result: Generic party planning checklist that could be found anywhere online.
Refined prompt:
“I need to plan a retirement party for our company’s beloved office manager of 25 years. Budget is $500, attendance expected around 40 people, in our office space next month. Please suggest unique ways to celebrate her love of gardening and mystery novels, while keeping it professional. Include decoration ideas, catering suggestions, and a rough timeline.”
Result: Detailed, personalised plan with creative touches like mystery-themed decorations and succulent party catering.
Advanced prompting techniques
Try the ‘Layer Cake Method’. Think of prompting like building a cake – layer by layer: Start with your base request, add specifications, request a first draft, then review and refine with follow-up prompts.
For example:
- Layer 1: “Help me create a monthly newsletter template”
- Layer 2: “Make it suitable for a law firm’s internal communications”
- Layer 3: “Include sections for case wins, new hires, and industry updates”
- Layer 4: “Now, add specific formatting for Outlook email distribution”
Best practices beyond prompting
Using AI in this regard is a game of education. Common users talk of “training” the AI. The same is true for anyone new to using AI. As such, here’s a few tips for early adopters:
Keep a prompt library. Create a document of your most successful prompts. Think of it as your personal AI phrasebook.
Use temperature control. When available, adjust the AI’s “temperature” setting. You can actually instruct the AI to apply different temperatures in the prompt.
Fact-check important information. Always remember that AI can make mistakes, and ‘hallucinations’ are a real thing, where the AI can actually manufacture information that has no factual basis. Always verify crucial details, especially numbers and dates.
Be aware of the context window. AI has limits on how much it can “remember” in one conversation. For complex tasks, break them into manageable chunks.
Common pitfalls to avoid
The Kitchen Sink Approach. Throwing every possible detail into a prompt isn’t helpful. Stay focused on what’s relevant.
The One-Word Wonder. Single-word prompts rarely yield useful results. “Newsletter?” isn’t going to get you what you want. “Create a one-page monthly newsletter template focusing on employee achievements and company updates” more likely will.
The Assumption Avalanche. Don’t assume the AI knows your industry jargon or company-specific terms. Define acronyms and specialised terms.
The Human Touch. While mastering AI prompting can revolutionise your workflow, remember that it’s a tool, not a replacement for human judgment. AI is like having a brilliant intern who needs clear direction but can produce amazing work when guided properly.
As AI technology evolves, prompting techniques will too. The key is to stay curious and experimental while maintaining professional standards. Remember, the goal isn’t to create more work for yourself but to make your existing work more efficient and effective.
The multi-model approach: knowing which AI to use when
Finally, one of the most overlooked aspects of working with AI is choosing the right tool for the job. Just as you wouldn’t use Excel to write a novel or PowerPoint to manage your budget, different AI models excel at different tasks.
Large language models like GPT are excellent for writing and editing, research summaries, creative brainstorming, document analysis and email composition but are not recommended for image creation or editing, real-time data analysis, or handling sensitive personal information
Barbara Wong, Operations Director at TechFlow Solutions, shares her workflow: “I use different AI tools like I use different applications. One for creative writing, another for data analysis, and a third for image generation. Understanding these distinctions has transformed how I approach each task.”
Remember, the goal isn’t to become an AI expert – it’s to be an expert at using AI to enhance your existing expertise. As Rachel Goldman, a veteran Executive Assistant, puts it: “I’m not in the AI business, I’m in the getting-things-done business. AI is just one of the most powerful tools in my arsenal.”
Also read: Should you mind your AI manners? | Executive PA Media
