Competent is the problem, not the goal
Ask a model for a post about hiring, or churn, or your first launch, and you will get something clean, balanced, and correct. That is the trap. A competent post about a topic is the single most common thing on X, and it is the thing people scroll past without a thought. Nobody stops for correct. They stop for specific.
What travels is narrow and strange: the exact number you saw on a Tuesday, the customer who left and told you why, the thing you believed for two years and were wrong about. A model does not have those. It was not in the room. No prompt hands it a memory it never had, which means the part of the post that actually earns the follow is the part you have to bring yourself.
The honest division of labor
This is not AI versus you. It is a split, and the split is not close. A model is good at the work around the idea. You are the only source of the idea.
| A model does this well | Only you can do this |
|---|---|
| Get you past a blank page with a first draft | Supply the experience the post is about |
| Give you five openings so you can pick one | Take the risk of an actual opinion |
| Tighten a draft you already wrote | Know which of your failures is worth admitting |
| Hold your cadence on a week you are tired | Judge whether the post is even true |
| Adapt one idea into a thread or a reply | Decide it is worth publishing at all |
| Flag a weak first line | Remember what it felt like when it broke |
Read down the right column. Every item needs a person who lived something and has a view about it. That is not a gap the next model closes. It is the shape of the tool. A model averages what has been written. Your best post is the thing that had not been written yet, because it only happened to you.
The specificity pass
Here is the practical move. Take the AI draft and go through it once, replacing every generic noun with something only you could have written. You are not polishing. You are trading averages for facts.
- "A lot of customers" becomes "eleven of our first thirty".
- "We learned a lot" becomes the sentence you actually said out loud when it broke.
- "Engagement improved" becomes the number, the week, and what you changed right before it moved.
- "Many founders struggle with this" becomes the specific thing you got wrong, and for how long.
- "It was a great result" becomes what the result let you do that you could not do the week before.
Every swap does the same job. It replaces a claim any account could make with one only your account could. When you run out of generic nouns to replace, the post is done, and it no longer sounds like anyone. It sounds like you.
Churn taught us a lot about our product. A lot of customers were leaving for reasons we did not fully understand, so we did some research and made improvements. Retention is now much healthier.
Eleven of our first thirty customers churned in one quarter. I called all eleven. Nine said the same thing: they never got past setup. We had spent the year on features nobody had reached yet. We deleted half the onboarding and the next cohort stuck.
The first is a plausible, competent draft. It is also empty: no number, no risk, nothing that could only be true of one company. The second is the same story with the facts put back. The number, the phone calls, and the admission are the whole post.
Consistency is what separates accounts that grow from accounts that stall. Show up, provide value, and over time you will build a loyal audience that trusts you.
I posted daily for two months and gained 40 followers. I almost quit. Then I stopped posting and spent that time replying to three bigger accounts a day instead. That did more in two weeks than the two months did.
The generic version is advice a model can generate about anyone. The edited one has a timeline, a real number, a moment of nearly giving up, and a move that contradicts the usual advice. Only the person who lived it can write it.
How to spot a post nobody edited
Unedited AI posts share a look. Once you can see it, you cannot unsee it, and neither can your reader. These are the signs that a draft went out with nobody home.
- Perfectly balanced clauses, where both halves of every sentence are the same weight and length.
- A summary sentence at the end that repeats the post without adding anything.
- A rhetorical question dropped in the middle to fake momentum.
- Claims that hedge in both directions so the post never actually takes a side.
- The closing "what about you?" bolted on to invite replies it did not earn.
- Not one number, name, date, or detail that could only belong to the person posting.
Using the tool without hiding behind it
None of this is an argument against drafting with AI. It is an argument for keeping the right job. Let the model carry the draft to the doorstep. You walk it the last few feet, where the specifics live, because that is the only part a reader remembers.
The model can write a sentence about your worst week. It cannot write the week. That part was always going to be your job.
So the rule is simple. Never publish a draft you have not put yourself into. If you cannot add one thing to a post that only you could have written, the post has nothing to say yet, and the honest move is to wait until you do.