Two Kinds of Writing
Writing can be a means for thinking. AI can help — but make sure you’re driving.
There are two reasons to write: to perform or to think. AI can help with both, but can also undermine them. Understanding which you’re doing will help you use AI more effectively.
The distinction isn’t new. Aristotle explored rhetoric as a distinct mode of communication, and “notes-to-self” also have a long history. More recently, neuroscientists have done influential research on how things in our environments extend our minds.
I won’t go into any of that here. Instead, this post aims to give you practical suggestions. Before we get into it, note I’m using writing as a verb. You’ll have “a writing” when done, but here we’re focused on the process of writing.
Performative writing
The first kind of writing we’ll examine is what I call performative. This is when you aim for words to perform a specific function involving other people. Often, that entails persuading them, but it can also be helping them understand something.
Examples include:
An paper meant to present research results
A syllabus meant to tell students what they can expect to learn in a course
A press release meant to promote a new product
A blog post meant to build your new consultancy’s pipeline
A business proposal meant to close a sale
An email meant to convince someone to attend your workshop
Social media posts meant to increase your status
Because the goal is moving or informing others, this writing must be carefully crafted: the final outcome matters. The text tends to be “fixed”: once published, it’s not often changed much. For example, news media and journals publish retractions and corrections when new facts arise.
Cognitive writing
Cognitive writing is in some ways the opposite. Rather than writing for others, you’re writing for yourself. Your focus isn’t on producing a “finished” text but on thinking through writing.
There are three primary reasons to use cognitive writing:
To hash out ideas
To remember things
To pay attention
Grokking this takes a bit of unpacking. You don’t think exclusively with your brain. Instead, thinking happens in a system that includes your brain, your body, and the world around you.
The word processor or notebook become parts of your “cognitive apparatus”: a means for thinking. The text lets you externalize and develop ideas; the “final” outcome is merely a record that thinking happened.
(Look up the extended mind thesis or read my book Duly Noted for more.)
Cognitive writing can take many forms:
Journal entries
Research notes
Outlines and mind maps
Early post drafts (such as I’m writing now)
Meeting notes
Creative writing exercises
As letters, words, and sentences pop out of your fingers and into the world, a feedback loop develops: the words and doodles suggest new ideas; you write those down, and they, too suggest new ideas.
So while performative writing is often “fixed,” cognitive writing is meant to evolve. It’s only done when you walk away — which you must do once you’ve sorted out your ideas or are ready to share them.
Which brings us to the fact that…
Much actual writing includes both kinds
All performative writing projects entails some cognitive writing. For example, consider this post. I shared it to perform several functions, such as teaching an important concept and bringing attention to my book. But first, I had to think about what I’d write.
To do this, I spent time noodling in Obsidian. I started my notes with a vague idea: that there’s a distinction between the kind of writing one does for oneself vs. others. I revisited my notes on extended mind theory and did some additional research, taking new notes along the way.
I hashed out the story through several drafts. I tried different phrases and experimented with moving stuff around. None of it was fit for sharing — it was messy and only made sense to me at the moment. (It may not if I returned to it in a few months!)
When I decided to share the post, I had to clean it up all this stuff. That required shifting into performative mode. I still did it in Obsidian, but changed how I went about it and what I focused on — and what I asked my AI assistants to focus on.
When to use AI
Which brings me to the main point of this post: AIs can help with both kinds of writing, but how you use it will differ depending on the mode you’re in. Let’s uncover how AIs can help in either case.
For performative writing, AI shines at:
Editing texts
Suggesting improvements
Allowing you to see your ideas from others’ perspectives
Improving structure and flow
Think of it like working with an editor. For two of my books, I worked with a fabulous editor named Marta. ChatGPT won’t replace her, but it can offer other kinds of valuable editorial suggestions.
For example, an AI can catch embarrassing mistakes and make texts more compelling and readable. But beware: given their statistical underpinnings, LLMs tend to produce middlebrow results. Accepting their changes uncritically will turn your writing bland and generic.
Yes, you want your texts to be clear. But you also want them to be memorable, and for the right reasons. Readers can smell AI-generated text as easily as they spot AI-generated images. If you use AI as an editor, do the final pass yourself to preserve your intent, style, and tone of voice.
Using LLMs for cognitive writing is trickier, because you won’t get the main benefits of thinking — the pleasure of exploring, developing, and internalizing an idea — if an AI does it for you. That said, LLMs can help in this stage as well.
For cognitive writing, AI excels at:
Translating texts from a different language
Suggesting related concepts you might have missed
Exploring different directions
Researching ideas more deeply
Looking for prior art that substantiates or negates your theses
Steelmaning your ideas (this is especially useful!)
Many people use AIs to transcribe meetings. I prefer to take notes by hand. AIs can accurately capture what was said, but they won’t internalize the conversation for you. And scribbling notes isn’t just a way to get a transcript: it’s also a way to be more present.
AI doesn’t obviate the need for writing
We write for different reasons. AI can help with some and hurt with others. If you’re looking to share your work, AI can help you produce more effective texts — as long as you’re still the one driving. And you must be even more careful with AIs when writing to think.
Some things you just can’t delegate. You can’t have others think or learn on your behalf any more than you can ask them to use the toilet for you. If you’re writing to think, ensure you’re actually thinking. AI can augment your mind, but shouldn’t replace it.
Both performative and cognitive writing require thinking, but thinking is the explicit purpose of the latter. Knowing why you’re writing will help you use it appropriately — and it starts with understanding the difference between writing for sharing and writing for thinking.
A few prompt suggestions
For performative writing:
Ask for general improvements: e.g., grammar, spelling, flow
Ask for audience-specific feedback
Ask LLMs to preserve your style and tone of voice
Include examples of your writing
Example:
Read this from the perspective of a chief marketing officer. What stands out? What’s missing?
For cognitive writing:
Ask for what might be missing
Ask for factual errors or inconsistencies
Ask for followup research on things it might uncover
Tell the LLM explicitly to not to correct your copy
Example:
Give me feedback on the ideas in this post. What am I missing? What is flowing well and what isn’t? Don’t offer to rewrite my copy — only suggest ideas for improvement. Be succinct.
This post first appeared on my blog.