If you’ve played with AI for any purpose, but particularly for your business’ benefit, you may have found your results less than impressive. One of the biggest challenges of working with artificial intelligence is our assumption that our instructions are sufficient, even when they are vague or non-specific.

A prompt of this caliber will not produce the quality responses that you need. Instead, you need to use “prompt engineering” to come up with a request that will. Fortunately, you can improve your prompts considerably by simply remembering three things.

The Recipe for Great AI Prompts

The three things you need to remember as you construct a prompt are as follows:

Context
Specificity
Iteration

Let’s go over what each of these points involves.

Context

Let’s face it: there’s a reason you’re prompting the AI to generate its response to your query. Therefore, it simply doesn’t make sense to expect the AI to operate in a vacuum. By providing background information, you’re helping the AI “understand” the underlying purpose behind the prompt, which gives the AI the ability to operate more effectively and give you a better result.

For example, let’s assume you're an auto mechanic and need social media content to promote a special offer you're running… maybe a buy three, get one free tire special. Instead of simply telling the AI to “write social media posts,” you would have much more success giving it something like the following:

“You are a marketing professional for [Garage Name], a mechanic shop serving Silicon Valley car owners. Please provide three Facebook social media posts to inform our followers about a buy three, get one free special on automotive tires. These posts should be conversational and engaging, helping us catch the attention of local drivers who care about their vehicles and their performance.”

Pretty different, huh? All the added details give the AI more to reference and use, resulting in a better outcome.

Specificity

As you add these details, make sure you’re being sufficiently specific. The more details and qualifiers you can give, the better your results are likely to be. We recommend:

Using action-oriented language, primarily using verbs that clearly describe what you want—research, summarize, or investigate, for example.
Including a preferred format in your instructions to the AI. If you want a list, ask for a list. If you want a complete blog, ask for one.
Providing the AI with all limitations and restrictions that may apply, whether that’s a word count, a tone of voice, or specific words to include or omit.

Iteration

Don’t stop after one attempt if you aren’t completely satisfied with the result. Artificial intelligence is, after all, intelligent—it can refer back to what has been said and use it to improve. Don’t be afraid to repeat your prompts, reiterating what you’re trying to accomplish, refining them until the tool gives you the outcome you’re looking for.

It can be helpful to treat these prompts like a conversation, adding context as you go to help the AI hone in on its task. Respond to it with feedback! Tell it that something is on the right track, but needs to be written for a different audience, or if you want more detail on a particular point. Let the AI do the heavy lifting as you generate multiple versions and select the best parts of each.

Remember, You Can’t Trust AI Entirely

AI is far from perfect. After all, it was created by us human beings, who are famously imperfect. As such, AI has inherited a few issues that you need to be aware of as you use it.

In short, AI isn’t always right, as confident as it may seem.

This is because all the AI is doing is processing a ton of data and extrapolating a response. Any biases present in the data it uses will ultimately come out in its answers, skewing them and providing you with bad information. It also doesn’t help that AI is designed to give you an answer—even if that answer is incorrect—as a means of keeping you happy. This is called a hallucination, and it always needs to be on your radar so you can check its work.

AI Can Be a Powerful Tool…

…provided you give it the context it needs, specifying what you are looking for across multiple iterations. You get what you give, in a way.