Training Day: Why AI should make your writing harder
Why AI should make writing a bit harder and slower
Photo by Helen Thomas on Unsplash
I’m going to say one of the most contentious things possible here on Substack.
I write with AI.
To be clear, AI doesn’t write for me. But I do use AI at multiple junctures when I write. Claude especially has been excellent for research, pulling up articles and reference materials (no dead links or hallucinations so far).
And I always have Claude do a few final checks for me. One: does the post hit all of my structural goals? Hook, factual accuracy, no controversial references, and payoff. Two: is there any awkward grammar that can be improved, if so show me before/after so I can choose if I want to edit.
If you’ve read this newsletter, you’ve probably heard me mention a few words repeatedly: friction, the rub, texture. The final check prompts above - those are all friction points, hurdles my writing has to jump before I can call it “done.”
Maybe this will sound like an exercise in over-thinking. Or maybe it’ll sound a bit masochistic. But I’m telling you - it’s like taking a barre or pilates class - these are core muscles that benefit from strength training. They need a workout as well.
So what are some other counterintuitive ways to force better writing through resistance? Keep in mind, my recommendation is not to use all of these tactics. But picking one or two, or rotating in and out a few of these techniques, is a good way to shake up one’s writing process.
#1 - The Resistance Test
Normally I wouldn’t suggest leading questions, but in this case it’s the point. Do not ever ask a Gen AI “do you like this?” unless you’re fishing for a compliment. Instead try this.
Prompt: "What are the three strongest counterarguments to this position? Where are my logic gaps?"
This is a good exercise to go through if you write on any lightning rod subjects. Basically, ask AI to critique you, but not troll you. And if you cannot defend your position, you need to work on it until you can.
Case in point, I asked Claude to apply it to this piece. It gave me five counterarguments - and one hit particularly hard - which was “This is still AI dependence, just dressed up.”
It made me realize I needed to preface this list with a caveat. This is not a recommendation to put yourself through a five-step gauntlet, but rather to try to build occasional friction in, when that’s not the typical use of AI.
#2 - Widen The Aperture
As I mentioned, Claude is great for research. But most will ask for research on a specific POV or to help support a conclusion. See what happens when you widen the aperture.
Prompt: "What adjacent ideas should I consider? What am I not thinking about?"
I wrote a post about how AI is being used more and more often in marketing, somewhat behind-the-scenes but still throughout the seasonal communication process. I showed ChatGPT my post, and asked a question similar to this, was there anything I hadn’t considered?
It mentioned reviews - OK fair point. More interestingly it pointed out recent regulatory action I hadn’t considered. For those of us not up-to-date on the FCC’s moves and rulings, this was a useful add.
(Did you know companies are selling fake reviews, including those that are created by AI? Well now perpetrators can be fined up to $52k per violation).
