I use AI to help write this blog. Oooooo. That’s a huge no no these days. I can already hear the Luddite trolls screaming, “That’s not writing!” “That’s AI slop!” Cool your jets folks and let’s look at what “help” means. Just the other day I worked with Claude to turn a rambling stream of consciousness about my Google Pixel trade-in deal into a post. It’s a great example of how I like to build a post because you can see every stage of the process.
The raw material
Here’s what I typed into the chat before any AI touched it:
Google just gave me a personal buy one get one free offer on Pixel phones. Well, sort of. My current phones are the Pixel 9 Pro and the Pixel 7a. I keep two phones because I like to have a spare, particularly when I travel abroad. I don’t want my main phone to break and then I’m suddenly wasting time scrambling to buy a new phone in a foreign country. So, I travel with two. I like the form factor of the 7a. It’s a little smaller with a thinner camera bar and the battery life was way better than my Pixel 3. Phones being phones and the constant upgrade of software means they get sluggish with age just like an old PC. So I traded in my Pixel 3 to get the Pixel 9 Pro. It was the first Pro phone I ever purchased. I tended to stay away from them because they’re so much larger than the base models. This time the Pro model was about the same size as the 7a so I pulled the trigger. I didn’t bother with the Pixel 10 series at all because it was not an upgrade from the 9 series and I didn’t replace my 7a with the 10a for pretty much the same reason. Wait long enough and the difference between the hardware on your old phone and the new ones finally make a difference. Especially in the AI age when hardware matters to run the models. Along comes Google’s Pixel 11 series and of course as is well known with Google all the rumors of the phones’ hardware is true and it was a good enough upgrade I felt for my 9 Pro. So I looked into the offers for buying one with a trade in credit. They offered $500 for my 9 Pro! Good deal. But, on top of that they gave me $321.81 in Google Store credit to be granted after the new phone ships. That got me to thinking about how to spend that credit… I got my calculator out to start to do some figurin’.
That paragraph goes on for another 500 words covering the trade-in math, ending with a statement that Google had handed me a buy one, get one free deal. Every idea from my rambling draft is in there. None of it came from AI. I told it to turn this into a blog post, edit for structure, grammar, and clarity, but not change my voice.
The structural edit
The first thing I need an editor to do with a wall of text like that is to reorganize it into a coherent narrative arc: why I carry two phones, the upgrade history that got me to a 9 Pro and a 7a, the trade-in offer itself, the math, then my opinion about what it means for Google’s margins. AI didn’t add any ideas to my post. It took ideas I gave it that were out of order and put them in order.
The clearest example is the opening. My raw version buried the “why two phones” explanation in the middle of a sentence about foreign travel. The edited version pulled it forward as the first thing the reader sees, because it’s the premise the rest of the post depends on.
Applying the rules
Before I let a draft anywhere near my own voice, I run it against a list of rules I’ve built up over time: no “not X, it’s Y” contrast constructions, minimal adverbs, active voice, paragraphs of at least three or four sentences, no em dashes, numerals instead of spelled-out numbers, no sentence fragments. I created these rules because AI has a writing pattern that I don’t like. It’s a pattern that doesn’t sound at all like me. So, over time, after using AI as an editor I created a set of rules to remove the things I don’t like. I revise the rules from time to time if I notice a repeated phrasing or construction that sounds nothing like me or that I just plain don’t like.
Where AI overstepped, and I caught it
AI’s first draft of the Pixel post included this sentence: “My main phone can die, get stolen, or just glitch out on me.” I never wrote that. I never said it, implied it, or thought it in the original text. It reads like something that someone might say, but it isn’t what I said.
I told the AI: “You added too much to my original text. Don’t add in details like ‘get stolen, or just glitch out.’ I didn’t say that. Take out your embellishments.” The same draft had also rewritten “got to figurin’” into “started running the numbers,” swapping my phrasing for something generic. I told it never to replace colloquial phrasing with neutral language again. Those examples in some ways matter more than the rules-based editing does, because that’s how I keep my voice, and when AI fabricates details it’s a content problem, which is the difference between “AI helped me write this” and “AI wrote this.”
The final version
Here’s the opening of the published post:
I keep two phones, particularly when I travel abroad. I don’t want my main phone to break and then find myself scrambling to buy a new one in a foreign country. So I travel with two.
Compare that to the raw version’s opening, buried mid-paragraph: “Google just gave me a personal buy one get one free offer on Pixel phones. Well, sort of. My current phones are the Pixel 9 Pro and the Pixel 7a. I keep two phones because I like to have a spare, particularly when I travel abroad. I don’t want my main phone to break and then I’m suddenly wasting time scrambling to buy a new phone in a foreign country. So, I travel with two.” What changed between raw and final is order, grammar, punctuation, and trimming. That’s all.
How this post got made, compared to that one
This post came together differently, and I’m pointing this out because sometimes I know the angle I want to take but I’m not sure exactly how to put it into words. These are the two basic ways I compose a blog post. With the Pixel post, I wrote the entire raw stream of consciousness blob, in my own words, before I handed it off to AI. The tool’s job was pure organization, editing, and clean up. I supplied 100% of the content and AI supplied structure and grammar checking.
With this post, I did something closer to directing its creation. I gave AI the premise, my angle, the specific argument I wanted to make about AI as editor, and the source material to pull from. Then I let AI produce the sentences of a first draft, which I’m now rewriting in my voice before I post this. The ideas and the evidence are mine. The first draft is not.
Neither makes AI the author. In both cases, every claim in the final post is a decision I made. The Pixel post is closer to hiring a copy editor. This post is closer to hiring a first-draft writer, which I then rewrite as the person making the argument. I think both are perfectly legitimate ways to use the tool.
So did I write this or did AI? I gave the argument and the structure, and had the final judgment on every sentence that gets posted. AI supplied a rough draft I’m rebuilding in my own words. The tool did different amounts of work in each case, but it never wrote the post.
Leave a Reply