In 2023, the website then known as Twitter partially open sourced its algorithm for the first time. In those days, Tesla billionaire Elon Musk had only recently acquired the platform, and he claimed to be on a mission to restructure the social media platform to make it more transparent.
However, the algorithm’s code release was swiftly critiqued for being “transparency theater,” with critics noting that it was “incomplete,” and that it didn’t reveal much about the inner workings of the organization, or why the code worked the way that it did.
Now the site (rebranded as X) has open sourced its algorithm again, fulfilling a promise made by Musk last week. “We will make the new 𝕏 algorithm, including all code used to determine what organic and advertising posts are recommended to users, open source in 7 days,” he’d said. Musk also promised to provide transparency into the algorithm every four weeks for the foreseeable future.
In a post on GitHub on Tuesday, X provided an accessible write-up about its feed-generating code, along with a diagram of how the program works.
What has been revealed isn’t particularly earth-shattering—but it does provide a peek behind the algorithmic curtain. The diagram shows that, when sifting about for content to feed a particular user, the site’s algorithm considers their engagement history (what posts they’ve clicked on, etc.) and surveys recent in-network posts. It also conducts a machine-learning-based analysis of “out-of-network” posts — as in, content from accounts that the user doesn’t necessarily follow — that it believes the user might also find appealing.
Image Credits:Screenshot
The algorithm then filters out certain kinds of posts, including ones that come from blocked accounts or that are associated with muted keywords, as well as content that has been deemed too violent or spam-like. The algorithm then ranks this content based on what it thinks the user will find most appealing. This process considers factors like relevance and content diversity so users don’t just get a bunch of posts that are all alike. The algorithm also considers content according to the likelihood that the user will like it, reply to it, repost it, favorite it, or otherwise engage with it in some way.
Image Credits:Screenshot
This whole system is AI-based, according to X. The GitHub write-up released Tuesday notes that the system “relies entirely” on the company’s “Grok-based transformer” to “learn relevance from user engagement sequences.” In other words, Grok is looking at what you’re clicking and liking and feeding that information into the recommendation system. The write-up also notes that there is no “manual feature engineering for content relevance,” meaning humans don’t manually adjust how the algorithm determines what’s relevant. It adds that the automation “significantly reduces the complexity in our data pipelines and serving infrastructure.”
Techcrunch event Join the Disrupt 2026 Waitlist Add yourself to the Disrupt 2026 waitlist to be first in line when Early Bird tickets drop. Past Disrupts have brought Google Cloud, Netflix, Microsoft, Box, Phia, a16z, ElevenLabs, Wayve, Hugging Face, Elad Gil, and Vinod Khosla to the stages — part of 250+ industry leaders driving 200+ sessions built to fuel your growth and sharpen your edge. Plus, meet the hundreds of startups innovating across every sector. Join the Disrupt 2026 Waitlist Add yourself to the Disrupt 2026 waitlist to be first in line when Early Bird tickets drop. Past Disrupts have brought Google Cloud, Netflix, Microsoft, Box, Phia, a16z, ElevenLabs, Wayve, Hugging Face, Elad Gil, and Vinod Khosla to the stages — part of 250+ industry leaders driving 200+ sessions built to fuel your growth and sharpen your edge. Plus, meet the hundreds of startups innovating across every sector. San Francisco | WAITLIST NOW
... continue reading