How to Steer ChatGPT Effectively for Creative Content
Generate engaging content with fewer rewrites
Creative inputs matter.
Using OpenAI’s ChatGPT to create engaging and informative content can be a real game-changer, but you need to know how to talk to it to get the best results. In this post, we’ll look at the top tips for crafting prompts that help ChatGPT generate high-quality and focused content. From understanding your audience to using clear and specific prompts, these tips will help you get the most out of ChatGPT — in fewer prompts.
1. Be more specific.
ChatGPT’s ability to generate accurate and relevant output depends largely on the quality of its input prompts. To help guide the model in the desired direction, it’s important to provide specific and detailed prompts. However, before ChatGPT can process a prompt, it must first tokenize the input. Tokenization is a process that breaks down the prompt into smaller units of meaning (tokens) that the model can better understand.
When the input prompt is tokenized, each token is assigned a numerical value that reflects its meaning. These values are used by the model’s transformer architecture to analyze the input at a granular level. This allows ChatGPT to better understand the relationships between different tokens in the prompt, and to generate more accurate and relevant output as a result. For example, if the prompt includes the token “dog” and the token “walk,” the model can use its understanding of these tokens to generate output about walking a dog.
By using meaningful tokens in your prompts, you can provide the model with the context it needs to generate more useful and targeted content.
Keep in mind that your input can be hundreds of words long if needed. Put it to use! Provide detailed instructions about the audience, tone, and content of the requested story.
Let’s look closely at a few of those aspects.
2. Specify the Audience.
When generating content with ChatGPT, describing the target audience in detail can drastically shape the output. Providing clear information about the intended audience helps the model understand the context and adapt the content accordingly. Consider:
- How familiar is the audience with the topic?
- Is there important demographic information that will shape the content? (e.g. age, gender, location)
- Why are they reading it? Do they expect a high-level overview or detailed analysis? Or are they just looking for something funny to read in the elevator?
- How many minutes are they expecting to spend reading it?
- What makes content feel trustworthy to this audience? (If you don’t know, try asking ChatGPT.)
3. Describe the writing style.
It’s instinctive to tell ChatGPT what you want, but you probably have a shape in mind for how it should be. Pull those details out of your head and into the prompt window! Be explicit about the writing characteristics you want ChatGPT to adopt.
Here are some aspects to consider:
Tip: if you don’t know how to describe the writing style you have in mind, try copying text from one of your existings works into the ChatGPT prompt and asking it to describe your tone and style.
4. Orchestrate the discussion around specific sections.
Generating something long? Sometimes ChatGPT output gets cut off mid-narrative.
When it does you cut you off, can simply submit the word “continue” to see the next chunk.
But if you’re like me and you’re working around the GPT-4 message cap, you might not want to spend one of the precious messages on “continue.” You can start being specific about what edits you want to see.
For example, say ChatGPT drafts a blog post for you that requires two (or more) output cells. If you request an edit in three of the paragraphs, you’re likely to get the revised full post, again requiring the two output cells to view it.
Instead, you could tell ChatGPT that you want to see revisions on only the specified sections. This output manipulation is also a nice way to pack value densely in each output cell. You could even ask for a couple variations of the selected sections.
Technical side note: Wondering why I’m not naming an actual word count limit? Because it’s fuzzier than that. We come back to the concept of tokens. Token count roughly measures the amount of content in a text. There is a set token count shared across the input cell and the output cell of a single exchange: N maximum tokens for “prompt” text + “completion” text, based on which model you’re using. In practice, it means there’s no hard word count limit. And if you feed in a massive example text as the input, you’ll likely get a short output before it’s cut off.
For more detail about tokens, see OpenAI’s blog post “What are tokens and how to count them?”
5. Explore your piece’s readability
ChatGPT often seems to create content that doesn’t necessarily focus on brevity and clarity by default. Even if you do like the current version of your creative content, consider testing what the “simpler” version is.
Example prompts to improve readability:
- “Rewrite the previous output using shorter sentences and simpler vocabulary. Include absolutely zero fluff. Every sentence should be important.”
- “Rewrite that to be 50% shorter, focusing on readability. Do not use any run-on scentences.”
- “Make it more straightforward. Focus on providing clear information in an engaging, accessible way.”
- “Make it sound like a human wrote it.”
TL;DR: Get comfortable describing your creative work.
Tailoring your prompts is the key to unlocking ChatGPT’s full potential and creating truly engaging content. As you fine-tune your approach, you’ll discover the power of specificity and context in shaping the output. Hopefully these tips will help you consistently generate content that stands out and connects with your target audience. Creative in, creative out!
