AI Castaway

Finding Answers with AI Not Working? 6 Tips for Getting Better Results

A huge part of making AI work for you is knowing how to engineer a good prompt. Here's a primer.

"Prompt engineering" is the official term for the black art of getting AI to deliver the answers you are looking for when asking questions. With AI getting integrated into so much software, prompt engineering is an important skill to master so you can benefit from AI and the apps that use it. You can even get a job as a prompt engineer.

While you can pay many hundreds of dollars to take courses on prompt engineering, I believe the basic skills can be summarized in a TL;DR version here.

1. Key Prompt Elements
Generative AI works on a series of rules, and if you provide what the AI is looking for, you will get better results.

When asking a question, add a lot of detail and provide a clear objective, such as, "Provide me a menu for one week, including different ways to prepare chicken, with ingredients which are readily available for purchase at U.S. grocery stores, and provide the source links for the recipes." Long and run-on sentences may be bad form for writing articles, but they are good for AI.

Be sure to provide a context that stipulates what to include in the answer. In my previous prompt, my context included the timeframe, ingredients, purchase locations and the sources for the information.

Also specify the output you want, whether that's a list, steps or, in the case of my prompt above, a menu. Specify what you want to include or exclude in your result, as well.

2. Pick Your Expert and Tone
When asking a question, pick a type of person to be the responder or the audience. For example, you could start your question by saying something like, "Act like a travel agent," or end it with, "as a nutritionist."

AI provides better answers if you give it a context, and the context can be a persona. The persona can contain a number of different elements from professions, demographic groups, tone or style of an author.

For example, I know someone who is neurodivergent and he always asks AI to rewrite his e-mails to be more compassionate based on feedback received from his manager. In their recent conversations, he was complimented on his new tone. I don't think his manager knows the source of the compassion is AI.

3. Follow Up
If you get an answer but want more information, start the next prompt with the words, "With respect to the previous question," and ask for more information. You can also ask the AI to "clarify the previous question" or to "provide summaries of all the sources."

You may find that if you ask the same question providing a persona, your results may improve. For example, "Answer the previous question as an economist" or "Answer the previous question as a Gen Z" may provide you with more insight than you previously received.

4. Get the Sources
There have been many accounts of AI providing bad information, so it is a good idea to get the sources for the details it provides. Include the words, "and provide the source links," or "from now on include sources for every request," so you don't have to ask every single time.

Also make sure that you can click on the sources and get to a Web site, as fake sites that do not resolve can be (and have been) generated just as easily as false information.

5. Ask AI How To Ask AI a Question
If you want to know something -- including how to ask a question -- ask AI how to do it.

After you ask a question, prompt the model, "Tell me a better way to ask the previous question." This way, you will learn what it wants.

One other thing you can ask AI is, "Give me some examples of how I can be more specific." The more specific the question, the better the answer.

6. Tell AI How To Think or Respond
To better understand a complex process, ask AI to "think step by step" or "provide the steps" as part of your question. This can result in more clear instructions.

Also, you might provide it with some examples and ask it for more information based on the examples you have given. Be sure to provide questions and answers, then ask the AI to answer a question in a similar fashion that does not include the answer.

Summary
Following these rules can yield a significant improvement in the results you see from the questions that you pose to AI. Accuracy and quality will definitely improve when you apply the techniques listed here.

But the most important thing is to make sure that you validate the information you receive from AI rather than taking it at face value, as generative AI does sometimes give false answers.

About the Author

Ginger Grant is a Data Platform MVP who provides consulting services in advanced analytic solutions, including machine learning, data warehousing, and Power BI. She is an author of articles, books, and at DesertIsleSQL.com and uses her MCT to provide data platform training in topics such as Azure Synapse Analytics, Python and Azure Machine Learning. You can find her on X/Twitter at @desertislesql.

Featured