What you need to know

• Twitter has effectively open sourced its amplification algorithm, sharing the code on GitHub.
• Likes provide tweets with the greatest boost, with retweets close behind.
• Mutes, blocks, spam and abuse reports, unfollows, and getting labelled as misinformation hurt virality - and so do spelling mistakes and including links in a post.
• The algorithm also explains why Twitter is basically an echo chamber. It punishes posts that don't fit neatly into the lookalike box it has crammed your profile into.
• Twitter specifically designates Jack Dorsey, Katy Perry, Stephen Curry and Barack Obama as “testing accounts” for getting random Tweets for testing, with an emphasis on Katy Perry in particular.

Twitter published details over the weekend about how its algorithm boosts or diminishes the amplification of the more than 500 million daily tweets, releasing the relevant code onto GitHub and effectively open sourcing its methodology. The decision is in keeping with owner Elon Musk's previously stated policy on the matter.

In a blog posted on Friday in the US, the company noted, "The foundation of Twitter’s recommendations is a set of core models and features that extract latent information from Tweet, user, and engagement data. These models aim to answer important questions about the Twitter network, such as, “What is the probability you will interact with another user in the future?” or, “What are the communities on Twitter and what are trending Tweets within them?” Answering these questions accurately enables Twitter to deliver more relevant recommendations."

There are three main stages to the recommendation pipeline:

• Fetch the best Tweets from different recommendation sources in a process called candidate sourcing.
• Rank each Tweet using a machine learning model.
• Apply heuristics and filters, such as filtering out Tweets from users you’ve blocked, NSFW content, and Tweets you’ve already seen.

Speaking on a Twitter Spaces session on the issue, Musk said, "Our initial release of the so-called algorithm is going to be quite embarrassing, and people are going to find a lot of mistakes, but we’re going to fix them very quickly. Even if you don’t agree with something, at least you’ll know why it’s there, and that you’re not being secretly manipulated … The analog, here, that we’re aspiring to is the great example of Linux as an open source operating system … One can, in theory, discover many exploits for Linux. In reality, what happens is the community identifies and fixes those exploits."

So how does it work?

So how does the recommendation algorithm play out in practical terms? According to Aakash Gupta, product growth lead at Buy Now Pay Later outfit Affirm in the US:

• Likes (30x) and retweets (20x) are the most powerful boosters.
• Replies barely move the dial, offering only a 1x multiplier.
• Images and videos provide much less of an uplift – 2x for each – despite what all those content marketing consultants have been telling us for years. Video and images may still improve engagement.

And what dampens amplification, according to Gupta?

• Anything labelled as mis-information gets crushed.
• Unsurprisingly, mutes, blocks, spam and abuse reports, and unfollows, have strong anti-viral qualities.
• Surprisingly, so does sharing external links — these are likely to get you marked as spam unless the tweet has enough engagement.
• Subeditors will be thrilled that spelling mistakes will get you algo-savaged. "Making up words or misspelling hurts," per Gupta. "Words that are identified as “unknown language” are given 0.01, which is a huge penalty. Anything under 1 is bad. This is really bad."
• Crypto posts will likewise not enamour you with the machines.

Echo chamber

Social media channels have a history of ignoring the law of unintended consequence, in particular when they optimise for engagement. YouTube for instance, famously pushed users towards more and more extreme content as researcher Zeynep Tufekci has outlined in the past. Twitter's algorithm meanwhile reinforces the echo-chamber effect of social media through its engagement for optimisation. By grouping users into what are effectively lookalike categories it is trying to ensure your tweets are seen by likeminded users but – and it's a very big but – it penalises you for sharing out of network content. What could possibly go wrong?

Additionally, given Musk is desperate to refill the leaky revenue bucket after he poked several giant Nazi-and-white-supremacist-holes in it, sending advertisers rushing to the exit, you won't be surprised to learn blue ticks have a magical viral quality. By default the tick offers a four-fold boost in amplification (though it can be anywhere from 0 to 100 times inside your own network), while offering a two times boost for out-of-network posts.

It might also not surprise you to learn that the algorithm specifically looks out for Tweets by Elon himself, though Musk professed surprise when this was pointed out to him in the Spaces session. Hong Kong based cybersecurity researcher Jane Manchum Wong found that little bon mot in the code, along with the fact that Twitter also checks whether the author is a Democrat or a Republican.

Wong also noted that, "Part of Twitter’s algorithm specifically designates Jack Dorsey, Katy Perry, Stephen Curry and Barack Obama as “testing accounts” for getting random Tweets for testing, with an emphasis on Katy Perry in particular."

She also noted that Twitter tried to hide this last point shortly after initially sharing the code on GitHub. "Twitter force-pushed a new initial commit to hide this, overriding the original commit. In the new commit, the original 'testing accounts'  is now a sad empty array."