Every growth guru has a rule about post length. Keep it short. Write long. Never go past one line. I got tired of being told, so I went back to the complete stretch of my own X history, 20,306 original posts from March 2022 to September 2023, and measured whether length actually moves likes. It mostly does not. What does move the middle is one character, the question mark: posts that asked something earned a median of 11 likes against 8 for posts that did not, and they were more than twice as likely to clear 100. Here is the data, the method, the limits, and what I am changing because of it.
Why I Went Looking
A while ago I counted every post I have ever made and wrote up what the whole pile looks like. The headline from that piece was that attention stacks instead of spreading: a tiny slice of posts earns most of the likes, and the typical post lives in single digits. The most common reply I got was not about that at all. It was some version of "okay, but are your captions too long?"
Fair question, and I had opinions about it, which is exactly the problem. I have a strong feeling that short and mean beats long and sweet. Feelings are not a method. So this time I picked one variable everybody argues about, length, plus one I had a hunch about, questions, and let my own timeline settle it.
Data and Method
The source is my own X archive export, the same file behind the earlier count. I only used the window where the export is complete, March 1, 2022 through September 23, 2023. Rows after that date come from partial scrapes, and a partial sample would make long gaps look like quiet months, so they stay out.
Inside that window I have 20,797 of my own posts. I removed 491 that were replies to other people, because replies are conversation, not broadcast: their median was 1 like and almost a third of them, 32.8 percent, got zero. That leaves 20,306 original posts. Of those, 20,137 were text only and 169 carried a photo or video. I analyzed those two groups separately because they live on different planets, and mixing them would let a handful of pictures drown out twenty thousand text posts.
Length means words, after stripping links and @handles. An emoji standing alone counts as a word. Likes are the only engagement number I used, because views were not recorded for most of the export. For each group I report the median, the 90th percentile, the share of posts that reached 100 likes, and the share that got zero. I lean on the median because the mean is dragged around by a few monsters; my overall mean across originals was 22.3 while the median was 9.
Finding 1: Length Barely Moves the Median
Here is every text-only post, grouped by word count.
| Words | Posts | Median likes | 90th percentile | Reached 100+ | Got zero | |---|---|---|---|---|---| | 0 (link only) | 217 | 3 | 8 | 0.5% | 15.2% | | 1 to 3 | 682 | 6 | 28 | 1.0% | 10.1% | | 4 to 7 | 4,090 | 9 | 36 | 1.5% | 4.4% | | 8 to 12 | 5,772 | 9 | 38 | 2.1% | 4.0% | | 13 to 20 | 4,704 | 9 | 34 | 1.2% | 2.8% | | 21 to 35 | 3,144 | 8 | 27 | 0.3% | 2.2% | | 36 or more | 1,528 | 10 | 34 | 0.6% | 0.9% |
Once a post has at least four real words, the median sits between 8 and 10 no matter how long it gets. A 5-word post and a 40-word post land in the same place on an ordinary day. The only length that clearly hurts is almost no length at all: a bare link earned a median of 3, and three-word posts earned 6.
I also checked the old 140-character line, since a lot of people still write as if it exists. Text posts at or under 140 characters had a median of 9. Text posts over 140 characters also had a median of 9. My typical original post is 12 words and 58 characters, which is short by habit, not by rule, and the data says the habit is neither helping nor hurting the middle.
Finding 2: Length Changes the Shape, Not the Middle
The median hides something more interesting. Length changes how a post fails and how it wins.
Long posts almost never flop. Only 0.9 percent of my 36-plus-word posts got zero likes, against 15.2 percent of link-only posts. If you write enough to say something, somebody shows up. But long posts also rarely break out. Posts of 21 to 35 words reached 100 likes just 0.3 percent of the time, while posts of 8 to 12 words did it 2.1 percent of the time, seven times as often. Over the 140-character line, the breakout rate was 0.4 percent; under it, 1.5 percent.
So length works like a dial between safety and upside. Short-but-real posts, roughly 8 to 12 words, have the best odds of a big night. Longer posts have a higher floor and a lower ceiling. Neither is wrong. They are different bets, and I had been treating them as the same bet.
Finding 3: The Question Mark Effect
This is the part I did not expect to be so clean.
| Text posts | Posts | Median likes | Mean | 90th percentile | Reached 100+ | |---|---|---|---|---|---| | With a question mark | 2,550 | 11 | 19.8 | 43 | 2.5% | | Without one | 17,587 | 8 | 15.0 | 32 | 1.1% |
Posts that asked something beat posts that did not at every level I looked at: the median by 3 likes, the 90th percentile by 11, and the share reaching 100 by more than double. The zero rate was about the same for both, 3.7 percent with a question against 3.6 percent without, so questions are not protecting the floor. They are lifting the middle and the top.
My read is simple. A statement asks a reader to agree. A question hands them a job. Even a rhetorical one invites a reaction, and a post people react to tends to travel further. That matches what the research on sharing keeps finding: content that pulls an active response out of people spreads better than content that only informs them.
And here is the embarrassing number. Only 12.7 percent of my text posts, 2,550 out of 20,137, asked anything at all. The best-performing habit in my own data was one I used about once in every eight posts.
While I was in there, I tested one more piece of internet folklore: that writing in all lowercase reads more casual and does better. It does not, at least for me. All-lowercase text posts had a median of 8; posts with any capital letter had a median of 9. Aesthetic, not strategy.
Finding 4: The Picture Still Wins
None of this touches the biggest lever in my timeline, which is still a photo or video. My 169 media posts sit on a completely different scale: every length group of media posts had a median between 596 and 687.5 likes, and 159 of the 169 kept their caption at or under 140 characters, with a median of 663. Length did not matter much there either. A picture with three words and a picture with thirty landed in the same neighborhood. If anything, the media numbers say the caption's job is to stay out of the way.
Limitations
This is one account, mine, during one stretch of one platform's history, 2022 and 2023. The audience, the feed ranking and the culture of X have all changed since, and I make no claim that the same numbers hold today. Everything here is correlation. Posts with questions might differ from other posts in topic, mood or timing, and I did not control for hour of day or day of week in this piece. Word counts treat an emoji as a word and ignore punctuation. Likes are the only measure; replies and reposts may tell a different story. Deleted posts are not marked in the export, so anything I removed back then is invisible here. Treat this as a well-measured diary, not a law of nature.
What This Means
For me, three changes. First, I am done arguing about length. Short-but-real is my best shot at a breakout and longer posts are my safest floor, and I can pick on purpose depending on what the day needs. Second, I am going to ask more questions, deliberately, because the habit that helped most showed up in barely one post out of eight. Third, captions on photos stay short, because the picture is doing the work and the words are there to frame it.
For anybody else building an audience: measure your own medians before you take anyone's rule, including mine. The mean will lie to you about your typical post, and the typical post is the one you are actually going to write tomorrow. If your data looks like mine, the cheapest upgrade available is not a new strategy. It is a question mark.
Sources
- Hellcat Blondie X archive export, complete window March 1, 2022 to September 23, 2023; 20,306 original posts recomputed on November 6, 2026 (first-party data).
- Kristina Gligoric, Ashton Anderson and Robert West, "How Constraints Affect Content: The Case of Twitter's Switch from 140 to 280 Characters," Proceedings of the International AAAI Conference on Web and Social Media (ICWSM), 2018.
- Jonah Berger and K. L. Milkman, "What Makes Online Content Viral?" Journal of Marketing Research, vol. 49, no. 2, 2012.
- Bongwon Suh, Lichan Hong, Peter Pirolli and Ed H. Chi, "Want to be Retweeted? Large Scale Analytics on Factors Impacting Retweet in Twitter Network," IEEE International Conference on Social Computing, 2010.
- Aliza Rosen, "Tweeting Made Easier," Twitter Blog, November 7, 2017, https://blog.twitter.com/official/en_us/topics/product/2017/tweetingmadeeasier.html