Writer’s Block: Trying to use ChatGPT as My Ghost Author
I have taken an interest in Generative AI, just like about 65% of the white-collar working population (or so it seems). I’ve been exploring different tools to help me become more efficient and marketable as an employee. However, I’m not just looking for a tool to copy and paste text into a large language model (LLM) and revise it into something more concise. Instead, I’m interested in generating long-form content, and I’ve started experimenting with tools like ChatGPT and Claude to see if I can efficiently create blog content that can help me expand my expertise; my experiment so far has not played out like I thought it would.
To be clear, I knew this wasn’t about learning a magic trick. I’ve picked up enough about prompt engineering to know that you have to think through your approach to effectively coax a good response out of the proverbial LLM sea. Even with that, however, I have struggled to nail a prompt in such a way as to get a blog post that felt fully… human.
Sometimes, the content generated by the LLM feels inauthentic and lacks personality and soul. In other cases, the LLM’s attempt at humor falls short and will even make a dad joke roll its eyes. To be clear, the content might be sensical and logical. But like that coworker who knows the language but not the domain, you can sense that something is off when they are speaking.
The truly maddening part is the content is not blatantly incoherent. Rather, the content that my prompts generate is just off the mark enough to leave me unsatisfied but not off enough that I can put my finger on how to fix it without simply writing it myself. It could be the former English major in me who studied hundreds of years of prose and poetry in college. Either way, when I have finished reading a draft of LLM-generated content, I feel most people will see right through the writing — it lacks personality and soul.
So my experiment hasn’t played out how I had hoped (at least not yet). But it has not been a total failure either. What I have found that works well is using these generative AI tools to help me brainstorm blog post ideas and even generate outlines that I can begin to run with on my own. Here are some tips I’ve learned so far.
- Spend time thinking through the prompt, and be specific with the instructions you write. You can’t really shortcut this one — this is a good data in, good data out equation.
- Include details about the voice, tone, intended audience, and even reading level (8th-grade reading level) you are targeting.
- Start by asking for a brief, concise outline. You can revise the outline and ask for a topic sentence for each section. If you are aiming for a certain word count, I have found that is an easy detail to ask the LLM to provide.
You can obviously repurpose the above approach to generate ideas for a presentation, creative brief, or a brainstorming workshop. If you are ambitious, you can even create your own GPT to help you generate ideas in a consistent format — a topic for another article.
I think these tools will continue to improve in terms of long-form content. But my experience so far has been that there is a threshold in terms of the amount of content an LLM can generate before it starts to feel… inauthentic. In the meantime, I will be doubling down with these tools to brainstorm ideas and even outline my next blog post. But for the actual writing? In my humble opinion, that’s still a human’s job — at least for now.
