Discover more from IN TRANSIT
How to write with AI: A practical example
Using AI to write more and faster while keeping your unique perspective and original ideas.
Hi everyone. Let’s take a small detour from the typical content, as it's more of a "behind the scenes." Hopefully, you'll find it useful!
ChatGPT can help you pump out tons of generic content without perspective and character. But, it can also be a helpful tool to augment your writing instead of replacing it.
AI can be useful for suggesting structure when your ideas are not.
AI is most valuable (to me) to get through the trough between an idea and the first pass.
Thanks for reading IN/TRANSIT! Subscribe for fresh ideas on the business of entertainment and culture (and the occasional tips and tricks)
Setting the Scene
I wanted to break down how ChatGPT fits into my current writing workflow, but I wanted to write something other than a generic list of tips. Instead, I wanted to show how I apply these principles in practice. So, I waited for an apt use case.
Last week I had a chance to write a guest column for Variety about Elon Musk rebranding Twitter to X.
The opportunity came through on a tight deadline. The time could have been better. This was our travel day back home from family summer break. Multiple airports. Three kids. There are better places to sit down and write.
But, that wasn’t stopping me. This is how I expanded from an idea in a LinkedIn post to a guest column. On the go, with the initial work done on my phone and leveraging ChatGPT for speed.
Not to create some generic, robotic, platitude content. But to create an opinionated piece of writing with my perspective in my style. Only faster than I could before.
Most of my writing happens this way; I start with core ideas and concepts on LinkedIn, and then I expand on the ideas I (and others) find the most interesting, turning them into longer-form posts over here.
My current inclination towards using generative AI in writing is that it can help with speed, quality improvement, and research. It can't do all the writing for you.
(Well, it can, but that will not produce your best possible writing.)
I currently use three AI tools in my writing:
ChatGPT for augmented outlines, drafts, feedback
Perplexity.ai for quick research. It's a great way to go from overview to deep dives. Fast.
Oasis AI for voice recording with transcriptions. I didn't use Oasis for this specific essay (voice recordings with three kids in an airport didn't seem like the right tool for the job.) Maybe I'll cover my voice workflow in a later post.
My workflow looks like this (AI involvement in italic)
First pass (AI Writing)
Second pass (Actual writing and editing)
Here's how it works in practice:
Most of my writing starts inside Apple Notes on my phone. Not because it's the best way to write but because it's convenient.
For the guest column about Elon Musk, I started writing down essential concepts, key takeaways, and analogies (this is a great way to direct GPT away from "boring" and add more style to it).
This is what my note looked like after a few hours of thinking and adding to it in between doing other things:
In the note, I also mark research tasks with a new line beginning with [re]. You'll see how that works in the next step.
I've found the key to setting ChatGPT up to be a great writing assistant is being detailed in the initial prompting. I think of it as "programming with natural language."
Let's take a closer look at each step, but first, here's the full prompt to kick off the AI write fest:
After pasting in my note from Apple Notes, I get the following summary:
I tell it to proceed, but most times the research it does is not enough. This is because of the AI knowledge cut-off date. Even a robot can’t research things it doesn’t know has happened.
When OpenAI re-enables browsing, it might be possible to do all the research inline. For now, I bring each of these tasks to Perplexity.
Perplexity is great because it lets you ask questions and receive bespoke summaries and sources. It's easy to ask follow-up questions, retaining the context of the previous question(s).
With my research done, I'll return to my notes and replace the research task lines with relevant information.
On to the writing!
First Pass (AI Writing)
I've found this to be essential to move the outputs of GPT closer to my own style of writing. I grab two previous articles, give them to ChatGPT, and ask it to emulate the tone of voice, paragraph length, and style.
Bonus tip: You can also use this to improve your writing by
Give it two pieces of writing from someone you look up to
Give it some of your own content
Ask it to be rewritten in the style of No. 1
The next step is to feed in the updated note to get a draft outline. I then go a few rounds with ChatGPT to re-order and make adjustments.
Next, I instruct ChatGPT to write out a draft based on the outline. You might have to remind it about things like length, the style calibration you did earlier, as it sometimes “forgets” parts of the previous instructions (which makes me wonder how sentient it is…
Sometimes I'll do this a few times to get different passes and find the one that makes for the best foundation.
I've gone from a note with fragmented ideas around a common theme to a body of text I can continue working with.
For the next step, I paste the draft into Google Docs and start working. I'll keep the good parts. Others I delete. I rewrite a lot and add depth where it makes sense. And go on until it feels complete.
There's not much more to write about this part, but it is part of the process where I spend the most time. Naturally, this is where the writing happens.
When I'm satisfied, I return to ChatGPT with the updated draft. The prompt is simple:
"Give me 7 tactical ways to improve this for clarity and readability".
I might repeat this step a few times more to see what other things it brings. I'll address some of the suggestions it makes.
And that’s it! The final piece of content is a healthy balance between AI and my own words.
The initial brief plays a crucial role in shaping the final piece. It's possible to have an AI write entire articles when it's about something fairly general – like "10 productivity hacks" (but do we really need another article about that?).
Using ChatGPT in the way I’ve described feels like an augmentation, not a replacement.
Lately, I've started exploring training GPT on some of my frameworks and mental models, like The Entertainment Participation Spectrum. And then feeding it ideas and cases to run through the framework for me. It becomes an interesting idea-making machine. Maybe I'll write about that next.