How I Use AI In Writing
Prompted by a reader question on whether I'm using AI to write this newsletter, I thought to expand a bit more on how I currently use AI in my writing.
On the newsletter
The newsletter articles themselves are not written with AI, not proofread by AI (other than spellcheck) and not edited by AI. There are two reasons for this, one principled and one practical.
The principled reason is that I can't expect you to be bothered to read something that I couldn't be bothered to write. If you wanted to know what Claude, ChatGPT, or, god forbid, Grok thought about any given topic, you'd just ask those AIs directly instead of waiting for me to ask them and then pass it on to you.
The practical reason is that an AI-generated email wouldn't do me any good. A primary benefit of writing about ideas related to AI is that it forces me to think about the topic clearly enough to then write it down. Writing exposes vague thinking; prompting an AI does not.
On LinkedIn
Once I've written the newsletter, I also write a shorter version for LinkedIn. This is a translation task, so I sometimes ask Claude, with a Claude Skill I create for this purpose, to draft the first stab. It mostly helps with removing parts of the intro that don't make sense in a LinkedIn post, and generally tightens it up a bit. But it also often misunderstands the punchline of the post, or falls back to terrible cliches (I'll share the unedited output of it below, which worked quite well, actually, so you can judge for yourself.)
In Other Communication
What about emails to current, past, or prospective clients? No AI here, ever. If you are in any way in a business that's based on expertise and trust, do yourself a favour and write your own emails. You want someone to hire you for what you can do with your brain. Showing that you're too lazy to use it is not a good look then.
And what about using it to proofread and comment on it? If it's not a sensitive topic, don't overthink it. Your thoughts, expressed in your own words, will pack much more punch than the oddly polished version the AI would come up with. If it is a sensitive topic, you could use AI to interrogate what you wrote ("Am I missing something here?") or critique it like an editor would. Just don't expect good results from having the AI "polish" it for you.
Brainstorming and internal docs
Here I am more liberal with my AI use. Generating ideas and insights often follows an "expand and contract", or "diverge and converge" pattern. First, you generate lots of ideas and follow lots of threads. Then you prune out and keep the good stuff and iterate with that. In the generate phase, AI lets you quickly flesh out ideas or follow threads, so that following any one thread doesn't take too much time. I wouldn't retain or use the writing generated by AI that way, but I'd use, reuse, and combine the ideas I got from reading it.
Don't Inflate. Compress.
Maybe that's the overarching advice for how to get good uses out of AI and avoid bad uses. If your output is smaller, you're compressing, often for your own internal use, and that can be a good use. Get to the gist of something when you're pressed for time. But if the output is larger than the total sum of the input, you're inflating. That is the way of the slop-slapper. "My idea is a single bullet point but we're using AI to make it into a 10-pager." Nobody wants to read that.
So there you have it. If you're using AI much differently than that or have strong opinions about how I'm all wrong about it, hit reply and let me hear it.
Appendix - The AI-generated LinkedIn version
What the unedited output of my Claude "newsletter-to-linkedin" skill produces, plus its editorial comments. For this post, it actually worked quite well and didn't commit any annoying LinkedIn cliches:
A reader asked whether I use AI to write this newsletter.
No. Not to write it, not to edit it, not to proofread it beyond spellcheck.
Two reasons.
The principled one: I can't expect you to read something I couldn't be bothered to write. If you wanted Claude's opinion on a topic, you'd ask Claude. You don't need me as a middleman.
The practical one: writing exposes vague thinking. Prompting does not.
Where I do use it:
The LinkedIn version of each newsletter. That's a translation job, so I let Claude take the first stab with a skill I built for it. It trims the intro and tightens things up. It also regularly misses the punchline or reaches for a cliche I then have to strip back out.
Brainstorming and internal docs. Generating ideas runs on an expand-and-contract rhythm, and AI is good at the expand half. I keep the ideas and throw away the prose.
Where I never use it: client emails. If your business runs on expertise and trust, write your own emails. People hire you for what you can do with your brain. Showing you're too lazy to use it is a bad look.
The pattern underneath all of it: if the output is smaller than the input, you're compressing. Getting to the gist when you're short on time is a fine use.
If the output is bigger than the input, you're inflating. "My idea is one bullet point, let's have AI turn it into a 10-pager." That's the way of the slop-slapper. Nobody wants to read that.
Don't inflate. Compress.
Claude's editorial comments:
Closing line options
"Don't inflate. Compress." (as above, your own line from the source. Cleanest, lands the rule without explaining it.)
Callback to the hook: "So, no, AI didn't write this newsletter. It did write the first draft of this post, which is the whole distinction." Risk: on LinkedIn, out of context, some readers will read it as you endorsing AI-written LinkedIn posts and stop there.
"More words out than in means you didn't save time. You moved the work onto your reader."
What I cut and why
The proofreading section. It contains the sharpest practical advice in the piece (use AI to interrogate what you wrote, not to polish it), but it reads as a sub-case of the client email point and it was the only place I could get real length back. If you'd rather keep it, the brainstorming paragraph is the next thing I'd drop.
"Hit reply and let me hear it." Standard cut per your no-CTA rule, though the newsletter version should obviously keep it.
The "if you wanted to know what Claude, ChatGPT, or, god forbid, Grok thought" line got compressed to just Claude. The three-name list plus the aside is funnier in the newsletter and reads as filler at LinkedIn length. Say the word and I'll put it back.
Kept verbatim
"Writing exposes vague thinking. Prompting does not." Technically it's a two-beat contrast, which you usually want stripped out, but it's your sentence and it's the load-bearing one in that section.
