How to Write a Good Tweet: 9 Rewrites From Weak to Strong

Most guides give you rules without numbers. This one shows nine rewrites. Each one shows one rule, with a number from my own posts behind it. I wrote the weak versions for this page. When the strong side is a real post of mine, I say so and link it.

Here, a “good” tweet is one that gets reach and replies. If you want leads or sales, a longer post about how you solve a problem can be the better tweet, even if it gets fewer quick replies. Also, short posts winning is true for my audience, not a rule. Across other users, longer posts did slightly better.

What makes a good tweet?

A good tweet gives the reader something to do in a few seconds. They can answer it, see themselves in it, or react to a number. I compared my top 500 posts with my bottom 200. I counted how often each trait showed up in each group. A ratio above 1 means the trait shows up more among winners. All of this comes from my 1,218 original posts from the last 180 days. The median post in my top 50 is just 10 words long.

Posts that ask for a quick, personal reply sit at the top. Posts that lecture, like expert advice and playbooks, sit at the bottom, below 0.2. Advice can still work. It just needs a bigger audience first. In a separate study of 500 outlier posts from 181 creators, most expert and playbook winners came from big accounts, while 34 of 37 question winners came from small ones.

How to write a good tweet: 9 rewrites from weak to strong

Each pair shows one rule. Most rewrites change a few things at once, like length, audience and format. So read them as examples, not tests. The numbers in each section come from all my posts, not from these pairs.

1. Cut every word that does not earn its place

On my account, short won. Posts of 1 to 8 words reached 1.29x my median. 9 to 15 words reached 1.18x, and 26 to 45 words reached 0.80x. I also checked about 54,000 posts from 669 ClimbX users in the same 180 days. There, longer posts did a little better, with 46+ words at 1.14x. No one length works for everyone. Start short and see what your own audience does.

2. Name the audience

Opening with the audience, like “Founders,”, had a win ratio of 4.0 in my posts. When a post is for everyone, nobody feels it is for them. When you name a group, they know you want their reply.

3. Add a word limit to the question

Questions with a word limit gave the biggest lift I have found on my account. They reached 7.25x my median. The full breakdown is in how to get more replies on X. Without a limit, people have to stop and think. With one, anyone can answer while scrolling.

4. Make it answerable from experience

Questions had a win ratio of 1.54 on my account. Opinions had 0.34. With a statement, people can only agree or not. With a question about their own life, they can answer.

In outlier posts from small accounts, questions got far more replies per view than playbooks. That chart and more ways to get answers are in the replies guide linked above.

5. Swap a generic claim for a specific number

I did an earlier study of 1,189 of my original posts. Milestone posts with a real number had a median of 3,767 impressions. Stories had 2,390. A number shows it happened. An adjective just asks people to believe you.

6. Drop the link from the main post

86 of my 1,218 posts had a link. They reached 0.60x my median. Posts without one reached 1.03x. If you want to send people somewhere, make the post work on its own. Then put the link in the first reply.

7. Remove “hot take:”

Posts labeled “hot take:” had a win ratio of 0.53 on my account. The label warns people before they even read the claim. If the take is good, it needs no warning.

8. Lead with the surprising part

Quick-reply hooks had a win ratio of 3.3 in my posts. These are openers that make you want to reply before you finish reading. The fastest hook is a contradiction your reader already feels. Put it first. Cut the backstory.

For more opening lines, see how to write hooks that stop the scroll.

9. End on the point, not “thoughts?”

Ending on a question made no real difference. It was 1.02x against 0.99x on my account, and 1.00x across the 669 ClimbX users. So adding “thoughts?” or “agree?” at the end does nothing. End on your sharpest line. Or make the whole post the question.

One warning. When a post does well, reuse its format. Do not reuse its words. When I repeated my winners, they did far worse than the original. The numbers are in what is considered viral on Twitter.

How do you know what works for your audience?

Count it. After 30 days, sort your posts by impressions against your own median. Check which of the nine fixes the top ones share. My length result flipped for other users, so trust your numbers over mine. For more of what worked on my account, see my 10 biggest X posts, and for which post types to write in the first place, what to post on X to get followers.

ClimbX drafts posts in your voice from what already worked for accounts slightly ahead of you. So each draft starts from a format that has worked before. You edit every draft before it goes out. When an idea needs more than one post, use a thread instead of stretching a single tweet.

My biggest post is 26 words, well past my usual length. It still follows most of the other rules. It names brands, has a contradiction people feel, and ends on a two-word question everyone in Europe could answer. Treat the rules as a starting point. You can break them.

A one-screen checklist before you post

How these numbers were measured

My numbers: 1,218 original posts on @DanielSmidstrup, last 180 days, at least 3 days old. Lift is a post’s impressions divided by my own median. The win ratios compare my top 500 and bottom 200 original posts from the same 180 days. The 1,189-post study is an earlier snapshot of the same account. Cross-user: about 54,000 original posts from 669 ClimbX users, same window and age cutoff, authors with 20 or more posts, my account excluded, lift against each author’s own median. The outlier study is a separate set of 500 outlier posts from 181 creators. Queried October 3, 2026.

Common questions