Professional writers aren’t haphazardly punching in a few prompts and then publishing whatever comes out. Those who use it are co-creating with ChatGPT in a far more structured way. Whether for ideation, execution, editing or repurposing, they have mastered the art of prompting the AI tool, using ChatGPT to supercharge parts of their process. You can copy their methods by following these 5 rules, to adapt your prompts for better results. While you might not be a professional writer, your audience expects greatness from your articles, tweets and emails.
If you want your written work to stand up to questioning and give the best impression of your company, improve your prompts with this guidance.
One of the most common mistakes when prompting ChatGPT is to ask for too many things at once. This risks the LLM unevenly focusing on some elements of the tasks over others. Counteract this by breaking tasks down into simple, singular prompts. ChatGPT is your intern who requires focused guidance, and don’t forget this when prompting it to perform. Even when you ask ChatGPT to rework something it has produced, change one thing first. After that, change another thing. Give incremental instructions to see the working and adapt accordingly.
In an ideal world, your prompt is 80% context and 20% instruction. You’re trying to give more than enough information to guarantee the perfect output. It definitely takes a little longer, but the results will be worth it. If in doubt, ask ChatGPT, “Where shall we start?” and get its suggestions for how to best begin.
Craft precise instructions
If you give vague prompts you will get vague results. When prompting ChatGPT, you should not only follow the 9 rules for an effective prompt, but also ask ChatGPT if the prompt you’re about to give is suitable. As you might expect, ChatGPT composes very good prompts for ChatGPT, so ask it to critique what you’re about to say and get its ideas for amendments.
It’s safe to assume that people who say ChatGPT produces mediocre writing are simply providing mediocre prompts. With careful consideration of what you ask for, and how you describe it, you can co-create complex masterpieces that you wouldn’t know a human didn’t write. Go into detail about the outcome. Do you want an article, a tweet or a haiku? Two sentences, one punchy paragraph or a list of 7 ideas? Be precise in every instruction to get exactly what you need.
Be clear in your communication
Imagine that a new intern called Chad came to work at your company. You wouldn’t sit them in the corner without giving them any context or instructions and expect them to conjure up miracles out of nowhere. It’s exactly the same with ChatGPT.
To set up for success, explain what you want to write. Create a checklist, explain the rationale behind your prompt, and check ChatGPT has everything it needs before it begins. Don’t let it go off on its own steam without being totally clear it knows the goal. That’s a waste of tokens, energy, and everyone’s time. Ask it if your communication is clear. Consider asking it to repeat the task back to you, to make absolutely sure.
Divide the labour accordingly
Some people use ChatGPT for ideation and some use it for execution. Either way, you as the human have to play a part somewhere along the journey. If you have the idea, explain what you want ChatGPT to do with it. If you’re looking for ideas, explain the parameters within which they should sit.
Division of labour means success with AI, but you should remain the master of the output. Following the rules for effective prompting, explain why you want something, specify what you want, and the format in which you want it delivered. If you’re happy with your subheading but you want ChatGPT to work on the opening hook, state that explicitly. Clearly explain which tasks ChatGPT is responsible for, leaving no doubt as to which it should tackle.
Master feedback loops
If the output you had in mind doesn’t come out on the first go, take a moment to pause and reflect. Re-read your original prompt and figure out what you said that caused ChatGPT to go in the wrong direction. It could be a misleading word or the phrasing of a question. Different LLMs will approach the same prompt in different ways. What matters now is the feedback and the reprompt.
For version two to be closer to the mark, master feedback loops. Don’t get upset, don’t get frustrated, just apply your logical brain to the conundrum of what needs to change. Explain what was correct about the output and explain the edit. As with the first rule, giving one piece of feedback at a time will mean you can make incremental changes until you arrive at something you’re proud to say is yours.
Prompting ChatGPT like a professional writer
Professional writers take pride in their work. They’re not creating mediocre content from ChatGPT. If they’re using it at all, it’s for ideation, specific elements of their art, or to improve what they’re already very good at. They’re crafting precise instructions so it’s clear what’s required, getting intentional about how the output should be delivered, and splitting the tasks by ideation and execution. They’re playing the long game, giving one prompt at a time to hone the words, then perfecting how they give feedback to iterate towards perfection.
Put more thought into how you prompt ChatGPT for better results than you’ve been getting so far. Copy the methods of the great writers who have mastered it.