People assume a timeline with hundreds of thousands of followers runs on a secret. So I did the least glamorous thing possible: I exported every post I have ever made on X, 32,307 rows across four years, and counted. This is a report on what that count shows. The short version is that attention is not spread out, it is stacked. One percent of my posts earned a third of all the likes I have ever received, the median post earned 9, and the things that move a post from 9 to 900 are measurable and boring: a picture, a short line, and the hour it goes up. Everything below is from my own data, with the method and the limits spelled out so you can argue with it.
The Question
Every creator has a story about why a post went off. Mine were mostly wrong. The useful question is not why one post did well but what the whole population of posts looks like, because the shape of that population is what you are actually managing when you post every day. If attention were evenly distributed, consistency alone would be the strategy. If it is concentrated in a few outliers, then the job is to raise the floor and buy more lottery tickets at the same time, and those are different behaviors.
So the question for this piece is simple: across everything I have published, how is attention distributed, and which observable features of a post move it?
Data and Method
The dataset is my own archive export from X, merged with later read-only scrapes of my timeline, stored as one row per post. Totals: 32,307 rows, of which 2,382 are retweets of other people and 29,925 are my own posts (replies included). The first row is dated March 26, 2022 and the last July 31, 2026. The export is complete from March 2022 through September 23, 2023 (22,415 rows); everything after that date comes from partial scrapes, so counts of how many posts I made per year after 2023 are not reliable and I do not use them below. Where the analysis depends on a complete population, I restrict it to the complete window: 20,797 of my own posts over 571 days, about 36 posts per day.
I counted likes only. Views were not recorded for most of the export and reposts are too sparse to be useful. I did not remove deleted posts, because the export does not mark them. Times were converted to Pacific time because that is where I live and post. Word counts exclude links.
Finding 1: Attention Is Stacked, Not Spread
Across all 29,925 of my posts, total likes come to 664,353. The mean is 22.2 likes per post, but the median is 9, and the 90th percentile is only 37. The gap between mean and median is the whole story: a small number of posts pull the average up while the typical post sits in single digits.
The top 1 percent of posts, 299 of them, earned 32.9 percent of all likes. Thirty-seven posts cleared 1,000 likes. Six hundred and nineteen cleared 100. One thousand eight hundred and seventy-four earned zero. The single largest post, from May 22, 2022, earned 19,285 likes, which is more than 2,000 times the median post and about 2.9 percent of every like I have ever received on its own.
This is the shape economists call a power law and it is the normal shape for cultural attention. Vilfredo Pareto described it in income data in 1896. George Zipf found it in word frequencies in 1949. Chris Anderson named the back half of the curve the long tail in 2006. My timeline is not special; it is a textbook example.
Finding 2: The Picture Does the Work
The export recorded an attached-media field on 271 of my posts. Those 271 posts average 586.8 likes with a median of 490. Posts with no recorded media average 17.0 likes with a median of 9. That is a 34x gap in the mean and a 54x gap in the median.
Two cautions. First, the media field in the export undercounts: many posts carried a photo through a link that the export stored as text, so the no-media group is contaminated with image posts, which if anything understates the real gap. Second, the posts I attached media to were not random. They were the posts I had already decided were worth a photo, so some of the gap is selection, not causation. Even discounting heavily for both, nothing else in the data comes close to this effect.
Finding 3: Short Wins, But Not Too Short
My median post is 11 words. Bucketing the complete window by length: posts of 1 to 5 words average 21.4 likes, 6 to 12 words average 27.4, 13 to 25 words average 17.7, and 26 or more words average 15.4. The sweet spot is one sentence with a little texture. Posts under six words underperform the middle bucket, which surprised me, because I assumed shorter was always better. The medians tell the same story more quietly: 6 for the shortest bucket, 9 for everything else.
Replies are a separate species. Within the complete window I posted 491 replies that begin with another handle; they average 3.7 likes with a median of 1, against 22.3 and 9 for original posts. Replies are for relationships, not reach.
Finding 4: When I Post Versus When It Lands
I post in the morning. In the complete window, 32.5 percent of my posts went up between 6 and 11 AM Pacific, and they earned 31.6 percent of the likes, almost exactly their share. The afternoon block (noon to 5 PM) is where I over-perform: 27.0 percent of posts, 30.9 percent of likes, with the highest block mean at 25.0. Evening posts (6 to 11 PM) have the lowest mean, 19.0, but the highest median, 11, against a morning median of 7. In plain English: mornings are reliable volume, afternoons produce the outliers, and evenings give the typical post its best floor.
By weekday, Sunday posts average 29.0 likes against roughly 19 to 23 for every other day, on fewer posts (2,616 Sunday posts versus 3,202 on Tuesday). My biggest post ever was a Sunday afternoon post. I would not build a religion on one day of the week, but if I had to pick a slot for something I cared about, it would be Sunday between noon and five.
What the Literature Says
None of this contradicts the research. Salganik, Dodds and Watts (2006) built an artificial music market with 14,341 participants and showed that when people can see what others liked, inequality in outcomes rises sharply and which songs win becomes much less predictable; quality sets a floor and a ceiling, but social signals decide the rest. That is what a timeline is. Byron Sharp's How Brands Grow (2010) argues that growth comes from reach among light buyers rather than from loyalty among heavy ones, which matches what the media effect shows here: the posts that travel are the ones a stranger can understand in one glance. Kevin Kelly's 1,000 True Fans essay (2008) is the counterweight: the long tail of my own posts is not wasted, because the median post is where the people who already know me show up, and they are the ones who convert.
Limitations
This is one account, so nothing here generalizes on its own. Likes are a weak proxy for value; a post that earns 9 likes and three paying subscribers beats one that earns 900 likes and nothing. The export has no view counts and almost no repost data, and it does not flag deleted posts. The post-2023 rows are partial scrapes, which is why I did not report per-year volume or any before-and-after comparison around the months my account was private in late 2023. The media effect is partly selection. And the date range covers the period in which X changed its ranking system more than once, so the hour-of-day pattern may not hold next year.
What This Means
If I were advising a younger version of me with this spreadsheet in hand, it would be four lines. Expect the median: nine likes is normal, not failure. Put a picture on anything you want to travel. Keep it to one sentence with texture, six to twelve words. Post your best thing on a Sunday afternoon and your daily things in the morning. Then stop reading the numbers on individual posts, because the single post is noise and the distribution is the signal.
The unglamorous truth is that I did not grow because of a secret. I grew because I posted 36 times a day for a year and a half, put photos on the ones that mattered, and let a power law do the rest.
Sources
- Hellcat Blondie, X archive export (data download, 2022-03-26 to 2023-09-23) merged with read-only timeline scrapes through 2026-07-31; 32,307 rows, analyzed November 2026.
- X Help Center, "How to download your X archive", https://help.x.com/en/managing-your-account/how-to-download-your-x-archive (accessed 2026-11-05).
- Salganik, M. J., Dodds, P. S., and Watts, D. J. (2006). "Experimental Study of Inequality and Unpredictability in an Artificial Cultural Market." Science, 311(5762), 854-856. https://doi.org/10.1126/science.1121066
- Sharp, B. (2010). How Brands Grow: What Marketers Don't Know. Oxford University Press.
- Anderson, C. (2006). The Long Tail: Why the Future of Business Is Selling Less of More. Hyperion.
- Kelly, K. (2008). "1,000 True Fans." The Technium. https://kk.org/thetechnium/1000-true-fans/
- Zipf, G. K. (1949). Human Behavior and the Principle of Least Effort. Addison-Wesley.
- Pareto, V. (1896). Cours d'economie politique. Lausanne: F. Rouge.