How to Write AI Prompts for Quality Outputs

Artificial intelligence is, whether we like it or not, making itself at home in our society. There’s the viral “AI yearbook” trend, in which your head-to-toe-denim-wearing likeness eerily resembles your parents’ or – forgive me – your real headshot from 1995. There’s ChatGPT, where you’ve undoubtedly engaged in a full-on conversation a la AIM’s SmarterChild… am I aging myself? And then there’s those creepily realistic covers of AI-resurrected Freddie Mercury singing “I Will Always Love You” on YouTube – you get my point. AI is working its way into everything. It’s text. It’s visual. It’s sound. And it’s probably not going anywhere.

While this can be fun on a personal level, it’s incredibly daunting in the professional world. With little “official” guidance and next to no regulation, things can get messy (looking at you, Sports Illustrated). Like other marketing professionals, I’ve been tasked with generating blog content, code, and the like. As an SEO Content Editor at Wayfair, I’ve tested seven different generative AI tools in 2023 – but how do I ensure my outputs are quality? The answer, my friends, lies in the prompts I use. While I’m still a firm believer that the best way to make people-first content is with people, I can’t deny that generative AI does have a place in some of my work. I’d love to expand on that and share some of my learnings if you’ll humor me.

What’s in a Prompt?

Fact: large language models allow us to automate the process of generating copy. For many, this sounds shiny, new, and exciting. For writers and editors, this sounds like a nightmare. Will the robots steal our jobs?! And honestly, after months of using these tools, I’ll say it: if employers want to use AI most effectively, the answer is no. The success of these models relies heavily on the quality of the prompts that we – real live humans – use to guide them, and one sentence prompts aren’t going to cut it. The output will not only sound robotic, but it will also likely be full of hallucinations and fluff text. Writing descriptive prompts for AI can result in highly relevant and engaging content that resonates with your target audience. Here’s how to do that.

1. Identify your goal output.

Typically, this is one sentence that states exactly what you need from the tool. While you should be concise, you need to be as specific as possible about what you want and where it’ll be used. If you expect a certain word count, say that. If you need copy for an email, a blog article, or a bulleted list, say so. I’ll use my own example prompt below, and build upon it in each subsequent step:

2. State your target audience.

If you received a writing assignment, you’d want to know who was going to be reading it, right? The same goes for your AI tool, which (thankfully) doesn’t have common sense comparable to a human. So, while you might infer from the prompt above that my audience is obviously couch shoppers, AI doesn’t “think” like that. In order to receive a quality output that’s tailored to your audience, you have to be clear about who that audience is:

3. Get really, really specific.

Seriously, talk to this thing like it’s an alien from deep space with absolutely no contextual knowledge on what you want. What are some characteristics the copy must have? Think about the ideal tone of voice, must-have topics and keywords, and formatting needs, like bullet points, numbered lists, or headings. Avoid using complicated language or asking open-ended questions that could result in irrelevant (or just flat-out nonfactual) information. For the couch prompt, I know I want the output to 1) be in an educational yet casual tone, 2) not overuse exclamations, and 3) include related keywords:

You might notice that I repeated the primary keyword “how to choose a couch” here. This helps remind the tool of the overall context of my request – without that reminder, I’d be willing to bet the output would go off on a tangent on my related keywords and lose the larger point of the article description (see: space alien).

4. Say when and why.

Again, if you were given a writing assignment, you’d want to know if your client expected evergreen or seasonal content. Tell the AI if you need your copy to be continuously relevant or if it should be more tailored to a specific time period. On the same hand, you should tell the AI your ultimate goal – why are you trying to create this content? To inspire a purchase? To provide information? To rank well in search engines?

5. Revise, revise, revise.

If the AI is not outputting exactly what you want, revise your prompt until it gets close. Certain use cases may require more specifics, while others may require less. If the AI does give you bits and pieces you find useful, tell it what you liked and why in your revised prompt to help it “learn” what you’re looking for. And please, for the love of God, fact check and copyedit. I’ll iterate that the best people-first content is made by people. No one wants to read an article with no personality, nonsensical information, and grammar mistakes. Think of your AI output as the bones of your final product – a starting point to make it best-in-class. Though these tools are getting more sophisticated, they are far from perfect.

Finally, if you take nothing else from this, remember that AI tools are only as good as your inputs. If you take the time to be highly descriptive and thoughtful, your output will be higher quality and closer to a usable final product. With that, happy prompting!

Oh, P.S. I’d love to shout out Jasper.ai as an excellent AI tool for content marketers. Their prompt engineering is highly customizable and their content templates are top-notch.

Please note: The above prompt was an example and not an actual prompt used in my job at Wayfair, and all opinions shared here belong to me alone.