Research
What Got AI Launch Posts Seen on X
A music mashup app launched on X from an account with 1,386 followers. The post was 11 words long. On the day we read it, it showed 11,120 likes and just over a million views.
In the same month, the middle launch post from an account under 5,000 followers got about 235 likes. So we wanted to know what that one did right. We read every AI launch post with a video that went out in the month to 5 October: 383 of them, from What Ships, a public directory of launch videos. For each post we took the counts X showed on 5 October: views, likes, reposts, bookmarks, the account's followers, the length of the text and the length of the video.
Every number here is what the posts reported. We did not see anyone's analytics.
First, do not trust the views
66 posts in the set passed a million views. For 31 of them, about half, the likes, reposts and bookmarks added up to less than 0.3% of the views. Across the whole set, the middle post was at about 1.1%.
We cannot prove that any one post bought its views. But a million views that nobody likes or saves is not attention. So we judged posts on likes and bookmarks, and we read views with care.
And do not buy views. A big number with no likes under it tells everyone what happened.
Short posts did better
Posts of 15 words or fewer had a median of 786 likes. Longer posts had a median of 522. Only 51 posts were that short, so treat it as a lean, not a law.
The mashup post said what the app does in one sentence and then gave its name. That was the whole post.
A number in the first line
95 posts had a number in their first line: how much faster, how much cheaper, which version. Their median views were about 161,000. The rest had about 84,000. Roughly twice.
A number gives a stranger something to repeat. It is 40% cheaper travels. It is a new way to work does not.
Show the product in the first two seconds
We watched eleven of the films frame by frame, with the sound off. The breakouts from small accounts all showed what the product does in the first two seconds. Each one was one idea a stranger could say back to you.
The words worked the same way. One post was seven words: Someone built a FaceID for Mac. The account had 1,962 followers. The post showed 3,371 likes.
Length did not hurt. Films of 60 to 100 seconds had the highest median likes in the set, 780. People will watch a long film if the first two seconds tell them why.
Put the link in a reply
135 posts had a link in the post itself. Their median views were about 65,000. Posts with no link had about 110,000. If you need the link, put it in the first reply.
Post on a weekday afternoon
Weekend posts had about half the median likes of weekday posts: 341 against 701. Posts on Tuesday to Thursday between 15:00 and 18:00 UTC did best, with a median of 792 likes.
One more: 92% of the videos were wider than they were tall. On X, launch films are mostly landscape.
What this does not show
It is one directory's pick, not every launch on X. The patterns are correlations inside 383 posts, not causes. The counts were read once, on 5 October, and they keep moving. We did not listen to the sound, so music and voice are not in it.
It also does not say a good post saves a weak product. A post can get a product seen. It cannot make people want it.
The short version
- Say what it does in one sentence. Fifteen words or fewer.
- Put a number in the first line.
- Show the product working in the first two seconds of the film.
- Put the link in the first reply.
- Post Tuesday to Thursday, in the afternoon UTC.
- Judge it on likes and bookmarks, not on views.
We use this for our own launch posts. If you have a launch coming, Karos for launches plans the posts around it and drafts them every day. Nothing goes out until you approve it. Once you do, a single X post goes out by itself (after a one-time OK), and Results shows the views, the engagement and the link clicks it got.