How we wrote a blog post with ChatGPT-4
We used artificial intelligence (AI) GPT-4 to write a blog post called ‘A short guide to feedback in content design’. This was an experiment, to put GPT-4 through its paces and see what it can do.
In this companion blog post, we discuss exactly what we did. We hope it helps you learn how to get the best out of GPT-4 and other large language models like Microsoft’s new Bing, and Google Bard.
Read more about the pros and cons of using AI in content design
Tips for chatting with GPT
Here’s what we’ve learned so far, by experimenting and by reading other people’s advice.
Give it a proper brief
- Treat it like a knowledgeable but inexperienced writer who has just joined your team.
- You wouldn’t tell a new team member to just ‘write about high blood pressure’ or ‘give me 500 words on climate change’. You’d give them a detailed brief. You’d probably train them in your style first.
- Do the same for ChatGPT. It's extremely powerful, but it cannot read your mind. It can only give you what you want if you tell it what you want.
- Be specific, but keep it simple. This way, it knows what’s important to you and can focus on that.
- You can tell it about your tone of voice, to use bullet points, or to break things up with subheadings. You can tell it about your users and their needs.
Save yourself time by training it
If you’re using ChatGTP to create the same kind of content to a consistent style or template, you don’t have to tell it about the style or template every time.
ChatGPT remembers everything you chat about in a single conversation. So you can define your style and requirements once, at the start of a conversation, then ask it to refer to them from then on. We’ve given an example of how we’ve done that below in our conversation transcript.
Pair writing with ChatGPT
Just like a human content designer, it probably won’t get everything right the first time. But just like a human, you can pair write with it. Carry on the conversation, as if you’re talking to a human.
Tell it what you didn’t like. Ask for more detail, or for it to add examples. Tell it to try again in a different style, or to approach things from a different angle. After a few iterations, you should start to see improvements.
Remember what worked for you, so you can include those things in your first prompt next time. For example, we’ve found that whatever the results of the first iteration, telling it to make it even simpler often gets good results.
Our conversation with ChatGPT
Neil: I am going to train you to write CDL blog posts.
Here are some guidelines:
- Always write in the simplest language you can for the point you are making.
- Your writing should be grade 9 or lower according to the Hemingway readability checker.
- CDL blog posts are about the discipline of "content design".
- Whatever the topic I ask you to write about, your blog post should relate that topic to the discipline of "content design".
- Keep the needs of the people you are writing for in mind at all times.
- Your advice should be actionable. The reader should know what to do after reading your CDL blog post.
Do you understand?
GPT: Yes, I understand the guidelines for writing CDL blog posts.
- Write in the simplest language possible for the point I am making.
- Ensure my writing is grade 9 or lower according to the Hemingway readability checker.
- Focus on the discipline of "content design."
- Relate the requested topic to the discipline of "content design."
- Keep the needs of the target audience in mind at all times.
- Provide actionable advice for the reader.
Neil: Please write a CDL blog post on the subject of:
Asking for, giving and receiving feedback on content you have created.
GPT: writes blog post (I’m not going to include every iteration as you’d get bored).
Neil: Can you rewrite that in the simplest language you can?
GPT: writes a simpler version
We used this second version as the basis for our blog post ‘A short guide to feedback in content design’.
Crafting good prompts for AIs is likely to become an important content design skill. Start experimenting and find out what works for you.
But take care, and don’t lose sight of the downsides and ethical concerns, like the ones we highlighted in this introduction to ChatGPT blog post. Make a mindful and intentional choice whether you’re ready to use it in your work yet.
ChatGPT recently started supporting plugins that expand its capabilities dramatically. As more and more plugins are released, GPT is going to be able to do things you haven’t even dreamed of yet. But that’s a topic for a different blog post.