X Experiment: How X Algorithm Works

EP ran an experiment on X recently to determine what X prioritizes in its algorithm.

He did a test where he tested likes, replies, reposts, and bookmarks while involving his audience.

The Test

EP made 4 tweets in 4 minutes. Each asked for one thing.

• Tweet #1: Like the tweet.

• Tweet #2: Reply to the tweet.

• Tweet #3: Repost the tweet.

• Tweet #4: Bookmark the tweet.

He wanted to see which one got the most views.

Tweet #1: "Like It"

This tweet asked for likes. It was posted at 5:48 PM IST.

Here’s how it did:

• Views: 4807

• Likes: 126

• Bookmarks: 8

• Reposts: 1

• Comments: 2

It got some views and a lot of likes.

Tweet #2: "Reply To It"

This tweet wanted replies. It went up at 5:49 PM IST.

Look at the results:

• Views: 7057

• Likes: 11

• Bookmarks: 11

• Reposts: 2

• Comments: 66

The views jumped a lot. Comments were also way up, but likes were down.

Tweet #3: "Repost It"

This one asked for reposts. It was posted at 5:50 PM IST.

Here’s what happened:

• Views: 2519

• Likes: 7

• Bookmarks: 6

• Reposts: 11

• Comments: 6

Views were low. Even with reposts, it didn't get seen as much.

Tweet #4: "Bookmark It"

This tweet wanted bookmarks. It was posted at 5:51 PM IST.

Here's how it did:

• Views: 3592

• Likes: 7

• Bookmarks: 51

• Reposts: 2

• Comments: 4

It got some views and bookmarks, but nothing special.

Final Results

Replies got the most views. And likes were second while reposts and bookmarks didn't do so well. So, replies help a lot on X.

But here's the thing. Nobody likes to retweet boring content but likes and comments are maybe okay so it might be one of the reasons why other scenarios were flawed.

In any case, it was a good experiment to see. Someone with an even bigger audience can test this much better. Plus you can get free engagement which is how social media algorithms pay nowadays.

You might think, "Wait, isn't the X algorithm open source?" Well, it is open source. But, X hasn't updated that code in over a year. And anyone on X could see Elon Musk pushing his own tweets during the election.

So, seeing what actually works matters. Remember, what people say and what people do are 2 different things.

Top Tweets of the day

1/

In the long-term, your direction matters more than your velocity.

You can ship as fast as Open AI but if you don't own TPUs like Google, you are going to be left behind.

Call it foresight but Google just shipped the most mind-blowing AI update (Llava) and they are giving away the house for free.

Google and X AI have a chance just because they have billion-dollar cashflows and lots of GPUs and TPUs.

Elon Musk doesn't have the TPU advantage but he is known to make his teams ship fast every day including weekends and holidays.

2/

Devin charges $500/month. It is incredible as a cheap intern.

In a year, you could probably build a PC for $3000 that does what Devin does 24x7 while you go to sleep but Devin has a chance to be AWS for those who don’t like to host their own servers.

Google just announced Jules AI which is a Devin competitor. Google's biggest advantage is they own the TPUs while others are buying GPUs so Google's margin game is huge.

For context, TPUs use less energy than GPUs because they're optimized for energy efficiency and are considered better than GPUs for specific AI tasks as they are specifically designed and optimized for such computations, making them more efficient and faster when dealing with large-scale machine learning models.

3/

Facebook pays you to post on FB Groups. And then it gamifies itself by adding bonuses.

Facebook is reinventing itself before everyone's eyes but not enough people are noticing.

They want Facebook to be not just for boomers.

Rabbit Holes

1. Gemini 2.0 BEATS Claude. No Models have these NEW FEATURES! by Mervin Praison

2. AI Glossary by Case Done By AI

3. Maximize your social media reach with these crucial tips by Jack Friks

What’d ya think of today’s newsletter? Hit ‘reply’ and let me know.

First time? Subscribe.

Follow me on X.

More Startup Spells 🪄

1. How iScanner Makes $4 Million Per Month: Onboarding Secrets (LINK)

2. The Negative Ad That Made $60k in Sales for Crayo AI (LINK)

3. SEO Blueprint Behind Omnisend's $55M ARR (LINK)

4. The Blueprint for Mirroring Billion-Dollar SaaS (LINK)