What goes viral on X: we reverse-engineered 414 top posts

November 8, 2026

Every week someone posts a formula for going viral on X: use a thread, open with a number, write in lists, shout the first line. We tested those claims on real data instead.

Method

For each topic we pulled X's Top results (three pages, about 60 posts), kept posts from the last 14 days, and split each topic into its top quarter and bottom half by views. Then we compared what the two groups have in common. Posts older than 14 days were dropped because they keep collecting views, which made long posts look better than they are.

What separates the top quarter

What does not

Six findings

1. Bookmarks are the tell. Top posts collect almost one bookmark for every like; the rest get one for every three. People save things they plan to use: tutorials, tool lists, tactics. The most-viewed AI agents post, a 23-minute tutorial, has 29,120 bookmarks against 9,692 likes post. Write something worth coming back to.
2. Followers help, but reach comes from strangers. Top posts reached about four times their author's follower count. Over a quarter came from accounts under 5,000 followers. One AI agents clip from an account with 1,791 followers passed 1.5M views post. And the like rate falls as reach grows, which is what you expect when most viewers do not follow you.
3. The formatting hacks are noise. Lists, numbers in the hook, ALL CAPS and thread markers show up equally often in the top quarter and the bottom half. They are not why posts spread.
4. A tighter first line helps a little. Top posts open with a slightly shorter first line, 67 characters against 76. The best one-line hooks state the payoff: "this how we run one-person company with 100s ai agents" drew 1.1M views on a 10-second video post.
5. Topic beats craft. The median post about Claude Code got 53,745 views; about AI agents, 37,317. The median post about influencer marketing got 103. Same platform, same week, a 500x gap before anyone writes a word.
6. Post in US working hours. Two thirds of top-quarter posts went out between 11:00 and 21:00 UTC, which is 6am to 4pm in New York. Saturday had the fewest top posts; the other days were close.

Caveats

These are X's Top results for each keyword, which favor video: 85% of the posts were videos. The sample is 414 posts in eight topics over two weeks, enough to rule things out, not to promise a result. Repeat it on your own topic before you change how you write.

Run it on your topic

Each topic is three calls:

curl "https://scrapeio.com/v1/twitter/search/posts?query=cold%20email" -H "x-api-key: $SCRAPEIO_API_KEY"
# then pass data.cursor back as &cursor= for pages 2 and 3

Every post comes back with views, likes, reposts, replies, bookmarks and the author's follower count, in the same field names we use for TikTok and Instagram. Or install the ScrapeIO skills and ask your agent to "find what the top posts about my topic on X have in common". The social-trend-research skill does the rest.

Pulled through the ScrapeIO API on 8 November 2026 (UTC). Views and counts change by the hour. Independent analysis of public posts.