I Let an Algorithm Pick This Week’s Topic
And it accidentally proved my entire point.)
I want to be upfront about something.
I didn’t choose the subject of this newsletter.
The data did.
There’s a theory going around the creator world called the outlier method. It goes like this: don’t guess what to write about. Look at other people's posts that performed dramatically above your or their own baseline, not your most-liked posts in absolute terms. I’m talking about the ones that did 3x, 5x, 8x better than your median. Those are your outliers that the audience has already voted on. The topic is proven. Your only job is to talk about that topic from a new angle, in your own words, again and again.
So I ran the experiment on myself.
I analyzed 105 of my own LinkedIn posts against my baseline.
My median post performs at about 1.2x my average. My outliers ran between 4x and 8x. Now normally the first thing you learn when analyzing other people's outliers is that they are almost always about vague topics.
That was no different for me.
What was different for me, however, at least when it came to the posts over the last year, was that they were all about the same subject.
The post about the guy who sent me a 60-page AI-generated strategy doc he clearly hadn’t read himself got 7.3x my baseline.
The post about the 1,168-line rules file someone built to stop AI from sounding like AI, and how Claude itself refused to apply it. 6.5x.
The post about LinkedIn announcing it would fight AI slop... with more AI. 5.5x.
The handwritten note explaining why you can’t use AI to make your writing sound more human, because AI is not human. 4.3x.
Every single one of my outliers is about the same thing: why writing produced by AI sounds the way it sounds, and what it’s doing to the people who publish it.
So according to the theory, this newsletter should be about that.
So fine.
It is.
But first I want to take the theory itself apart, because it deserves better than the way it’s being used.
The outlier method is 100 years old
The outlier method isn’t new. It wasn’t invented. It was rediscovered in the same way the best ideas in marketing always get rediscovered every generation by whoever bothers to test things.
In 1923, Claude Hopkins published Scientific Advertising, and the entire book is one idea: stop arguing about which ad is best and let the results tell you. Direct-response advertisers ran coupons in newspapers and counted which headlines pulled. The ad that pulled got kept and became the control to be studied and rewritten. The ad that didn’t got killed, no matter how much the copywriter loved it.
An “outlier post” is just what a direct-response man in 1923 called “the ad that pulled.” And the analytics dashboard is the coupon. Nothing has changed except the speed of the count.
David Ogilvy built an agency on the same principle: research first, opinion second. And the reason it works hasn’t changed either: your audience is voting constantly. The problem is, and most of us ignore the ballots. And the sad part is why. Usually it’s because we’re bored with our own hits. We wrote about the thing, it worked, and now we want to write about something new, because we need novelty. But as the outliers of most content creators prove, audiences don’t need novelty. They need to know things like “how to never forget what they read,” or “never run out of ideas,” or “become the smartest person in the room.” The only novelty they want is a different opinion on how to accomplish those things from different people.
How to find your outliers
1. Find your outliers relative to your own baseline. Not your most-liked posts. A post with 90 likes is an outlier if your median is 20; a post with 400 likes is a dud if your median is 800. You’re looking for the posts where the audience’s response was disproportionate to your normal. That disproportion is information.
You can do this by hand with a spreadsheet, the way Hopkins would have. I don’t. I use Eden, which lets me pull up any creator (including myself) and filter their entire body of work by outlier score. That’s literally how I ran the analysis that opened this newsletter: 105 of my posts, ranked against my own baseline, in about a minute.
2. Extract the subject and the tension, not the format or the words. When my AI-slop posts overperform, the lesson is not “post more screenshots” or “open with a confession.” The lesson is: my audience has an unresolved anxiety about AI writing, and they show up when someone names it honestly. There are probably even deeper insights I could pull, but let's leave it alone for now. The proven thing is the nerve, not the sentence that touched it. This is important to understand because you can return to a nerve forever. But you can only repeat a sentence once.
3. Return to the topic from your current perspective. The topic is public property. Nobody owns “why AI writing sounds hollow” any more than anybody owned “how to remember what you read” or “why consistency beats talent.” What’s yours, the only thing that’s yours, is the angle you have today that you didn’t have last time. For this part, I use Magic Post. I feed it the outliers and use it to break down what’s actually underneath them: the subject, the tension, the nerve, so I’m working with the anatomy of the post instead of just eyeballing it. It’s also where I draft and schedule, so the whole loop lives in one place: outlier goes in, dissection comes out, new angle gets queued. What it doesn’t do, because nothing can, is the next step. Unfortunately, this is where everyone inserts the machine.
And here is where the whole thing breaks
The modern workflow looks like this: find outlier → paste into AI → “write a post like this about my niche” → publish. Tens of thousands of people, running the same method, on the same outliers, through the same model and publishing them on the same platform?
Do you see the problem? If not, you are certainly experiencing it.
The outlier method tells you what to say. But it is structurally incapable of giving you a way of saying it. And when everyone extracts the same proven topics and hands the treatment to the same statistical machine, the outliers stop being outliers, and everything regresses to the mean, because the mean is literally what the machine is built to produce.
Which, and I promise I didn’t plan this, is exactly why my outliers are outliers.
My audience doesn’t overrespond to posts about AI slop because the topic is magic. They overrespond because they can feel the difference between writing that comes from someone and writing that comes from a distribution.
OBERTHOUGHT: AI writing doesn’t sound bad because of the word “delve.” It sounds bad because there’s nothing to delve into.
You can ban every telltale word, and the hollowness will still survive, because the hollowness was never in the vocabulary. It was in the absence of a person with a unique perspective on a topic everyone seems to care about.
So the complete method has a step nobody sells you because it takes work and things that take work are not easy to sell:
The data finds the topic.
The human provides the perspective
And the editing supplies the human.
The first part is research.
The second part is ideation.
The third part is a craft, and it cannot be prompted into existence, because the whole point of it is that it comes from you.
Which brings me to the free webinar
Which is why I’m doing a live webinar with Tasleem Ahmad Fateh
If you don’t know Taz: he’s the OG LinkedIn guru. As I wrote in the latest Anti Guru Newsletter on LinkedIn, Taz predicted the LinkedIn selfie boom in 2022 and gained 50,000 followers in four months. He told a friend to go all-in on carousels before carousels were a thing. His one self-declared talent is spotting what’s working before the market catches on. So outlier-hunting is essentially his native language, and his new newsletter breaks those down every week.
Taz finds what works.
I make what works sound like a human being wrote it.
That’s the whole Content Hunter method that will make your writing more persuasive and more human.
Together we’re teaching the whole loop: how to write in the age of AI without sounding like every other person using AI to write.
📅 JULY 30th at 8:30 am PST:
Register here (it’s free): https://us06web.zoom.us/webinar/register/8117843441575/WN_Kj5uQvVWSOqsVJHcy_dhPw#/
If your feed feels like it was written by a server in Ohio, come find out why, and how to make sure your writing never joins it.
See you there.
— Justin
P.S If you want to get a head start before the webinar, The Copy Hunter Field Guide is the course version of my half of the stage: 52 hunts for editing anything, yours or the machine’s, until it sounds like a person with a point of view. The first 18 hunts are out. Use the code TAZ to get it for $52 while it’s being built.