How Google DeepMind launched Gemini 3.5 on X

What this launch shows

The post went out from an account with 1.6M followers and was seen by fewer people than follow the account. Accounts of that size usually reach 2.4 times their following.

This one did not get far past the people who already followed it. It ranks 406 of 611 launches from accounts that size. Against every launch in the directory it sits at 1.2 times the median of 688K views.

Reposts led, 381 at 4.7 per 10,000 views against a median of 2.5. A repost is an endorsement without a comment, and a launch with an unusual repost rate was being passed along on purpose.

Bookmarks: 459. Replies: 126. One bookmark for every 8.1 likes is a low ratio for a launch, and reads as entertainment more than intent.

The post opens with “Introducing Gemini 3.5: our newest family of models combining frontier intelligence with real-world action”, the same first word as 25% of launches in the directory.

It carries an image and no video. 92% of launches in the directory carry a video, so this is the rarer choice, and the few image launches that travelled did it on the strength of the image.

The copy runs 30 words, about the usual length for a launch post here, where the median is 41. The body carries a link, so the post had somewhere to send the reader and paid a little reach for it.

It went out on a Tuesday. Launches posted on a Tuesday have a median of 651K views in the directory, and Friday is the strongest day at 880K.

Why posts from large accounts under-reach covers this: a post seen by fewer people than follow the account was held back by the ranking model.

For the next one, read whether a launch needs a video: 99% of the launches in the directory carry one, and the guide is about when the 1% is the right call.

What viral AI launches share is the guide for it: model launches are the largest category in the directory and travel on the demo. This launch sits in the Developer Tool and Frontier Lab directories.