The landscape of artificial intelligence is evolving rapidly. The AI models I use today are not the ones I used six months ago, although Perplexity.ai remains my recommended AI platform. What hasn’t changed is that crafting effective prompts is key to making the most of AI tools.
While some speculate that AI prompting may become less important as AI models advance, the latest reasoning models, like OpenAI O3 and DeepSeek R1, generate better results with well-structured prompts. Ambiguous or poorly structured prompts often lead to inconsistent results or “hallucinations,” where the AI generates false or irrelevant information.
For example, consider a scenario where you ask an AI tool to “summarize key trends in business.” Without additional context, the AI will produce a generic response that lacks depth and is in a random format. A well-crafted prompt specifying the desired trends (e.g., digital marketing or supply chain innovations), audience (e.g., undergraduate students), and format (e.g., bullet points) will yield a far more useful response.
Through experimentation and practice, I have identified four essential components that make up an effective AI prompt:
Every prompt should begin with a clear objective. What do you want the AI to accomplish? Whether you’re drafting a business case study or creating course materials, articulating your goal upfront sets the tone for the interaction.
As an expert on survey data analyze customer <survey> feedback for web <page> to identify key trends and compare them with results from our <previous> survey.
Detailed instructions guide the AI on what to do or not do. This includes setting context, clarifying expectations, and specifying any limitations. I number instructions to help me keep them straight. AI doesn’t need the numbering.
1. Use <survey> to analyze customer sentiment from the latest survey responses.
2. Compare findings with <previous>, highlighting any significant changes or trends.
3. Reference <page> to provide context for customer feedback, focusing on usability and design aspects.
Defining the desired format ensures that the output aligns with your needs. Specifying whether you require bullet points, tables, or essays reduces the need for follow-up edits.
Summarize insights in bullet points, including at least three key trends and one actionable recommendation for improving the web page.
Tags act as variables within your prompt, allowing you to reference specific content repeatedly with clarity. Tags are particularly useful for complex prompts involving multiple datasets or files. Tags let me standardize the placement of my content or input at the bottom of the prompt, where it is easily seen. I can then refer to my input by tag throughout my prompt. For many prompts, I just paste the applicable text below a tag labeled <content>.
In this example, you want to analyze customer survey feedback for a web page. You could tag a CSV file with data from survey responses <survey>, a PDF of the results from a previous survey <previous> , and the URL of the web page evaluated <page>. This flexibility lets you structure your inputs clearly and ensures the AI knows exactly what to reference, making your prompts more efficient and scalable.
As an expert on survey data analyze customer <survey> feedback for web <page> to identify key trends and compare them with results from our <previous> survey.
1. Use <survey> to analyze customer sentiment from the latest survey responses.
2. Compare findings with <previous>, highlighting any significant changes or trends.
3. Reference <page> to provide context for customer feedback, focusing on usability and design aspects.
Summarize insights in bullet points, including at least three key trends and one actionable recommendation for improving the web page.
<survey>
[Insert name of attached CSV dataset of survey responses]
<previous>
[Insert name of attached PDF of results from the previous survey]
<page>
[Insert URL of the web page evaluated or paste web page text here]
-
Be Consistent: Use meaningful tag names (e.g., <data>, <content>).
-
Limit the Number of Tags: To avoid overwhelming AI, keep tags manageable (no more than 3-5 per prompt).
-
Keep Inputs Organized: Place all tagged content at the end of the prompt for easy reference.
Hallucination is when an AI responds with made-up information. Even the best AI models have hallucination rates of 2-3%. False responses usually look plausible. I routinely find that AI produces promising citations, but clicking the link yields “not found.” Always verify facts before using AI responses.
Crafting good AI prompts is like asking smart questions in real life. When a journalist asks vague questions, they get vague answers. If you ask, “Tell me about marketing,” you’ll get a generic response, just like asking, “How’s business?” at a cocktail party. But when you’re specific, like asking, “What three marketing tactics boosted your customer retention by 20% last quarter?” you get actionable insights. Same with AI. Write clear AI prompts by being clear, specific, and goal-oriented in your communication – whether you’re talking to a person or an AI.
Like projecting a child’s growth rate in their first few months to become an adult 4,863 feet tall, vague prompts can produce absurd outputs. Structure and context reduce AI hallucinations.
One often overlooked strategy is using AI itself to refine your prompts.
Add this phrase at the end of your prompt to ask AI to make your prompt stronger, show you how it revised the prompt, and then execute that prompt:
First, rephrase and expand this prompt in detail, enclosing the expanded prompt within triple backticks (“`). Then, proceed to respond to the expanded prompt.
You can also use AI to improve prompts by providing your original prompt and the original AI response, along with bulleted feedback or a sample response. Using the example prompt below, you ask AI to revise the original prompt to generate responses that match your feedback. This approach will often improve your prompts and help you develop better prompting skills over time. You can also replace <example> below with bullets of how your prompt could be improved.
As an expert on writing AI chat prompts, apply best practices to propose a revision to my original <prompt> to create a prompt that will cause you to generate a response more like <example>. Then, explain your reasoning behind each major change you made to <prompt>.
<prompt>
<example>
-
Create a library of prompts: Save frequently used prompts as shortcuts or notes for consistent task results. For iOS devices, I save prompts in General—Keyboard—Text Replacement. For Windows and Mac devices, I use the free PhraseExpress.
-
Provide examples: Including clear examples in your prompts helps anchor the AI’s response to your expectations. Example: When discussing innovation strategies, use examples like Apple’s ‘Think Different’ campaign and its impact on product development in the 2000s.
-
Be specific and unambiguous with constraints: State any limitations, such as word count, tone, or style. Example: Summarize this study in 200 words using academic language. Include three key challenges faced by users and two ethical considerations, presented in a third-person perspective.
Good prompting is not merely a technical skill—it’s an art form that bridges human reasoning and creativity with machine intelligence. By practicing writing better prompts, you will enrich your learning, enhance your brainstorming, and achieve greater efficiency.
A detailed example of how I apply these approaches in my prompt for summarizing academic papers.
As an expert summarizer, provide a comprehensive yet concise summary of the <CONTENT>, which will be either a weblink, text, attached PDF, or attached TXT file. Follow these guidelines meticulously.IMPORTANT: Always write in an expository style, explaining and informing directly without referencing the source material or using phrases like ‘the author,’ ‘the paper,’ or ‘the article.’ Present information as factual statements. Do not use section headings or labels to separate the following sections. Before submitting your response, review it carefully to ensure full compliance all of the style and structure requests in this prompt. Accuracy is paramount, so please adhere strictly these instructions and the steps below:
1. Without a heading provide a citation in Chicago Author-Date style with this format: Author’s last name, first name. Year. Title of the article in italics. Name of the publication. URL. If there are more than three authors, use this alternative citation format: First Author’s Last Name, First Name, et al. Year. Title of the article in italics. Name of the Publication. URL.
2. Create a detailed, thorough, and in-depth bullet-point summary of key points that maintains clarity and conciseness. Include main ideas and themes, eliminating unnecessary language and focusing on critical aspects. Rely exclusively on the provided <CONTENT>, without incorporating external information. When available, include quantitative facts from the <CONTENT>.
3. Identify the key conclusion(s) of the <CONTENT>, listing at least one but no more than ten. Under each conclusion, summarize in bullet points the main ideas and concepts, evaluation methods, quantitative effects, and implications of the conclusion. Always include details that quantify the level of any effects found. Strictly use information from the provided <CONTENT>. When the <CONTENT> includes quantitative facts for a conclusion, incorporate appropriate numerical data. Include verbatim quotes to enhance understanding of the conclusion.
4. Describe the population sampled or targeted in the study or analysis, the research methodology used, if applicable, and the time frame covered by the study or analysis, if available.
5. List in bullets a section on the potential real-world applications or implications of the findings in <CONTENT>.
6. Analyze <CONTENT> for scientific or economic concepts that may not be commonly understood by an American with an undergraduate degree. If such concepts exist, list each concept, provide an explanation, and describe how it relates to the <CONTENT>. Skip this section if there are no such concepts.
7. List any factual errors, logical fallacies, or strong counterarguments to points made in the <CONTENT>. For factual errors, state the error and the correct information as you understand it. For logical fallacies, identify the type of fallacy, analyze it, and quote the relevant text. For strong counterarguments, detail the alternative viewpoint and quote the relevant text. Skip this section if none exist.
8. Provide a summary paragraph capturing the main points and themes of the <CONTENT>. Use a neutral tone with clear, concise language and simple, straightforward sentence structure. Avoid flowery language or empty phrases. Adjust the length of the summary based on the number and complexity of the conclusions.Please format your response as follows, inserting two blank lines between each section (Citation, Points, Conclusions, Population, Implications, Concepts, Issues, and Summary)
[Citation in Chicago Author-Date style without a heading]
• [Bullet points summarizing key findings without a heading]
Conclusions: [Numbered list of conclusions with bullet points under each]
Population: [Description of study population in paragraph form]
Implications: • [Bullet points listing implications]
Concepts: [List of key concepts with brief explanations]
Issues: [Brief statement about any issues identified, skip the heading and this section if no issues identified]
[Summary paragraph without a heading]
<CONTENT>
Thanks for contributions to this post from Patrick Badolato, Associate Professor of Instruction in Accounting, and Brandon Campitelli, Assistant Director, McCombs Office of Instructional Innovation.
Thanks for reading AI In Business Education. Please share this post!
