There are many user guides to artificial intelligence (AI) prompting on the internet, most of which are free but some charge a fee for a course with certification.
This guide is different. Most of the prompting techniques here come from the tech companies that created these AI assistants in the first place: OpenAI, Google, Anthropic and Meta. If anyone knows how to prompt, it’s them.
What Is a Prompt?
A prompt is a request to do a task, written in natural language — as opposed to computer language — given to an AI chatbot or assistant such as ChatGPT, Gemini, Claude, Microsoft Copilot, Meta AI and others.
Think of an AI assistant as a very smart and fast new employee. It can help one write emails, reports and essays; it can analyze business data and create reports; it can do market research; it summarizes meetings and automatically sends follow-up emails; it can even brainstorm and think. Some AI assistants can also generate images, code, video and audio.
How good they perform depends on the quality of one’s instructions to them. Some also don’t real-time information. Ask the chatbot to make sure.
Why Prompts Matter
The better the prompt, the better the response. Good prompts generate more relevant, accurate, insightful, efficient and helpful responses.
For businesses, a well-crafted email can mean the difference between a good and a great customer experience; a detailed prompt on market research for a new product can help the CMO make more insightful decisions; and a rich summary of a meeting plus automatically generated follow-up emails can help ensure employees don’t miss the meeting’s key themes and actions.
Examples of basic business prompts:
- Summarize today’s market trends.
- Write a customer service email apologizing for a delayed order.
- Generate a five-step plan for launching a new laundry detergent.
Insider Tips
Before jumping into the best practices for prompting, here are insider techniques that may not be on everyone’s radar:
- Make your prompts longer: The most fruitful prompts average around 21 words with relevant context, yet the prompts people try are usually fewer than nine words, according to Google’s prompting guide.
- Use action verbs, define the length and format (for example, essay), and specify a target audience, according to Google.
- Give the AI context and constraints: To avoid getting a response that sounds like a generic AI-generated output, be more specific and add restrictions, according to Wharton professor Ethan Mollick.
- Rephrase the prompt to see if results improve. Also, ask the AI assistant itself to make the prompt better, according to Google.
Guidelines for Prompting
1. Provide clear, direct and detailed instructions.
For example:
- Basic: “Summarize the meeting.”
- Better: “Summarize the meeting notes in a single paragraph. Then write a markdown list of the speakers and each of their key points. List the next steps or action items suggested by the speakers, if any.”
Even better, add the goal of the task, its audience and, if applicable, how it fits into the overall workflow.
For example:
- Basic: “Craft an email marketing campaign for the new features of the AC Security software.”
- Targeted: “Your task is to craft a 200-word marketing email for our Q2 AC Security feature release. Write for a target audience of mid-size tech companies with 100 to 500 employees. Highlight three key new features: advanced data encryption, cross-platform sync and real-time monitoring and detection. Use a professional yet approachable tone. Include a CTA for a free 30-day trial. The subject line should be under 50 characters and mention ‘new security features.’
2. Add context, examples and adopt a persona
AI works best when it understands the background of the task. Instead of a generic prompt, provide a scenario.
For example:
- “Our CS team is overwhelmed with unstructured feedback. Your task is to analyze feedback and categorize issues for our product and engineering teams. Use these categories: UI/UX, Performance, Feature Request, Integration, Pricing and Other. Also rate the sentiment (Positive/Neutral/Negative) and priority (High/Medium/Low).
- Here is an example:
- Input: The new dashboard is a mess! It takes forever to load, and I can’t find the export button. Fix this ASAP!
- Category: UI/UX, Performance
- Sentiment: Negative
- Priority: High
- Now, analyze this feedback (attach actual feedback data)”
Another example:
- “You are the general counsel of a Fortune 500 tech company. We’re considering this software licensing agreement for our core data infrastructure: (Attach contract)
- Analyze it for potential risks, focusing on indemnification, liability and IP ownership. Give your professional opinion.”
Why can role-playing improve results? A data scientist might see different insights than a marketing strategist looking at the same data.
3. Use reference texts to reduce errors
If the AI assistant doesn’t have enough information on a topic, it could hallucinate. Give it the reference material to improve its accuracy.
For example:
- “Use only the following data or document to answer. If the answer is not found, state ‘I can’t find the answer.” (This is actually the core function of a technique called Retrieval Augmented Generation, RAG.)
One tip is to re-run the prompt several times to see if it confirms the answer. This is especially important for math because many LLMs still aren’t good at it, according to Meta, which calls this technique ‘self-consistency.’
4. Break down complex tasks into simpler steps
For example:
- Basic: “Analyze customer reviews and suggest improvements.”
- Better: “Step 1: Categorize reviews as positive, neutral or negative. Step 2: Identify common themes in negative reviews. Step 3: Recommend improvements based on findings.”
5. Tell the AI to think through responses
Although new reasoning models today can do it, this is still a best practice since not all AI assistants have reasoning skills.
For example:
- “Draft personalized emails to donors asking for contributions to this year’s Care for Kids program.
- Program information: (add program details)
- Donor information: (add donor details)
- Think before you write the email. First, think through what messaging might appeal to this donor given their donation history and campaigns they’ve supported in the past. Then, think through what aspects of the Care for Kids program would appeal to them, given their history. Finally, write the personalized donor email using your analysis.”
6. Experiment and iterate
Vary the prompts: Using different prompts can help the model learn more about the task and produce more diverse and creative output, according to Meta. Try using different styles, tones and formats.
Test and refine: After creating the prompt, test it on the model to see how it performs. If the results are not as expected, try refining the prompt by adding more detail or adjusting the tone and style.
7. Don’t forget to review the results
Even the best AI assistant can slip up. Make sure to check the results for accuracy before using it for business purposes.
See More In: AI, AI prompts, Anthropic, artificial intelligence, ChatGPT, Claude, Gemini, GenAI, generative AI, Google, Meta, Meta AI, Microsoft, Microsoft Copilot, News, OpenAI, prompt engineering, PYMNTS News
