Make AI Tweets Sound Less Generic: What to Cut and Add
To make AI tweets sound less generic, give the AI your own facts and examples of your voice before it writes. Then edit the drafts for the patterns that give AI writing away: stock openers, questions it answers itself, lists of three, abstract nouns and invented numbers. Replace each of those with something only you could say: a real number, a named situation, a mistake you made. Generic posts come from generic input; most of the fix happens before the draft exists.
How can you tell a post was written by AI?
Readers on X have seen a lot of AI writing, and the patterns are easy to spot:
- “Here’s the thing:” or “Let that sink in.”
- “Most people think X. But the truth is Y.”
- A question answered in the next line: “The secret? Consistency.”
- Three adjectives or three short phrases in a row.
- Big abstract nouns: “journey”, “landscape”, “game-changer”.
- A confident number with no source.
- Every post the same shape: hook, colon, list.
None of these is wrong once. The problem is when every post has two of them.
What does a generic post cost you?
A generic post doesn’t just underperform. It teaches readers that your posts can be skipped. Over a few weeks, people who once read you start scrolling past, and the posts with real substance get skipped with the rest.
What should you give the AI first?
Most of the fix is input. Before asking for drafts, give it:
| Input | What it replaces |
|---|---|
| Your facts: what you built, numbers you measured, what went wrong | Invented results and vague claims |
| 3 to 5 of your best posts | The AI’s default voice |
| A short tone note: “short sentences, dry, no exclamation marks” | Its upbeat, polished default |
| A banned list: phrases you never use | Stock openers |
| Top posts in your niche this week | Formats and lengths from nowhere |
Write the tone note from your own best posts. If they’re short and blunt, say so. Writing X posts with ChatGPT and Claude has a full prompt that uses all of this.
What should you cut from the drafts?
Go through each draft and cut:
- The first two or three words if they’re a warm-up. “Here’s the thing: pricing is hard” becomes “Pricing is hard”, and then you make it specific.
- Self-answered questions. “The fix? Ship smaller.” becomes “Ship smaller.”
- One item from every list of three, unless all three are real.
- Abstract nouns. “Our growth journey” becomes what actually happened.
- Any number you can’t source. Replace it with your own or drop it.
- Closing lines that restate the post. “That’s the real lesson.” can go.
What should you add?
Specifics only you have:
- A real number: “Took us 3 weeks”, “our second price change”.
- A named situation: “the Friday we shipped without a changelog”.
- A mistake: “I wrote the onboarding emails before talking to a single user.”
- An opinion with a reason: “Annual-first pricing is right for tools people use weekly, because the habit is already there.”
Compare:
Consistency is the key to growth on X. Show up every day and the results will follow.
I posted every weekday for two months. The posts that got replies were all about mistakes. The tips got likes and nothing else.
The second can’t be written by anyone else. (Write yours from your own record, not this example.)
How do you keep your voice across many posts?
Save your tone note and best posts somewhere the AI always sees them: a Claude Project, ChatGPT custom instructions, or a tool that feeds them in. Vary formats across the week so the shape doesn’t repeat; Twitter hooks examples has twelve first-line patterns to rotate.
Then get a second opinion before posting. Reading your own drafts, everything sounds fine.
How does Kalen check for generic writing?
Kalen scores every draft with Jev, a judgment model from TypeSafe. One of the questions he asks of each draft is whether it reads like generic AI writing; another, when you’ve saved a tone, is whether it matches it. The 0 to 100 score combines those with the hook, how concrete the post is, and how close it is to what earns engagement in your niche. When a draft reads generic, that’s the reason he gives, and your AI rewrites it, up to three rounds. Kalen also hands your AI your own best posts in every briefing, so it has your voice before it writes. More in what an X MCP server does.
If you write every post yourself, you don’t need any of this: your voice is already yours. Kalen is for people who draft with an AI and want a check for generic writing before anything goes out.
FAQ
Can people tell if I use AI for posts?
They can tell when the posts are generic. Posts built on your own facts and edited in your voice read as yours.
Should I tell the AI to “sound human”?
It rarely helps. Specific instructions work better: your tone note, a banned list and examples.
Do emoji or hashtags make AI posts feel more natural?
Not usually. Use them only if your niche and your own past posts do.
How long should editing take?
A minute or two per post once you know the patterns. Cutting is faster than writing.
Edit one post now
Try it now: take your last AI draft and run the cut list above on it. To have every draft checked for generic writing before it’s scheduled, connect Kalen to Claude, ChatGPT and Cursor and see the plans.