The Algorithm Isn't the Problem — What It Optimizes For Is

For almost a decade, social media sites have been telling us algorithmic feeds were a gift. Personalization done right sounded like showing you “the most relevant content,” not a firehose of posts. Then, in 2021, Frances Haugen left Facebook’s offices with internal documents suggesting something uglier: the algorithm wasn’t optimizing for relevance. It was optimizing for engagement. And engagement, it turned out, was most efficiently generated by outrage, anxiety and conflict.

Since then the debate has split into two camps. One wants the algorithm out entirely, replaced by simple chronological feeds. The other side says the problem isn’t algorithmic sorting, it’s what social platforms have chosen to sort for. Both have good in them, but neither tells the whole story.

Table of Contents

The Case For Algorithmic Curating

The case against algorithmic feeds is an intuitive one. There is a clear chronological timeline. You see whatever the accounts you selected to follow posted, in the order that it was posted. There’s no black box that says one friend’s political rant should be rated higher than another friend’s birthday photo. User resumes control.

This is not a hypothetical issue. After Elon Musk bought Twitter in 2022, a large number of users joined the decentralized social network platform Mastodon and it has been a common practice on the platform to use a feed of chronological posts by default. Bluesky is another platform whose feed allows users to let themselves decide or even create their personal algorithms, instead of getting a platform-default one, like how it is handled on platforms such as a centralized one. They are based on the AT Protocol, and at its very beginning supported by the Twitter co-founder Jack Dorsey that has implemented the very same idea with the feed philosophy launching of a feed that allows the users’ freedom of decision, if not the creation of algorithms for the feeds.

Somewhat the two companies have been the examples where users are less manipulated when browsing, though the EU has used another way which they see as a more regulatory method. One of the measures provided by the new legislation, the Digital Services Act entered, very large online platform companies, among them the Meta, TikTok and YouTube were required to give their users at least one option that the recommendations provided are not based on personal data (profiling). The result of that is Meta, TikTok and YouTube have to give their European users an option to be able to move from the engagement maximizing curated experience. Whether this regulation really will change the ways of platform operators or it will get lost among other minor settings that will remain unnoticed by most people is a matter of discussion still.

Why Chronological Feeds Aren’t a Silver Bullet

And this is where the optimism gets sticky. Chronological feeds solve one problem, and create another.

For Twitter’s heaviest users, Twitter as a purely reverse-chronological timeline was a thrill. But for most people, more than a few dozen accounts made the feed an undifferentiated torrent. Volume buried relevant posts. That friction is one reason why platforms moved to algorithmic solutions in the first place — users wanted signal, not noise.

You might be interested in:

And then there’s who benefits from chronological ordering. In a time-based feed, it’s no surprise that media organizations, brands, and political campaigns that can post at high frequency dominate the feed. The little personal account that posts once a day is left behind. Ironically, a well-designed algorithm can bring out that quieter voice more reliably than raw chronology ever could.

The Real Variable: What does the algorithm reward?

The more precise question is not whether to sort algorithmically. It’s what tells an algorithm to listen to.

If an algorithm is optimized for watch time alone, for likes or shares, it will always promote the most emotionally provocative content it can find. That’s not a design flaw, that’s the design working as designed, if the design goal is to keep you engaged and sell you ads. Haugen’s leaked internal Facebook research implied the company’s employees knew their systems were boosting divisive content, and that proposed fixes were shelved partly due to concerns about lowering engagement metrics.

An algorithm calibrated for different signals — articulated user preferences, content diversity, time spent reading versus time spent scrolling — would generate a very different feed. Some platforms have experimented on the fringes. YouTube added features to highlight authoritative news sources during breaking events. Instagram rolled out a “Favorites” mode that gives priority to accounts users explicitly mark as such. They are incremental changes, not structural.

What Would Make a Real Difference

Three things probably need to happen simultaneously for meaningful reform: regulatory pressure that provides users with real, accessible alternatives to engagement-maximization; greater algorithmic transparency so that researchers and journalists can audit what is really being amplified; and platform business models that don’t rely solely on advertising revenue tied to time-on-platform.

Any of these conditions are not easy to induce. The DSA is a start, but the reach of regulation is uneven geographically. Transparency audits are only as good as the access that the platforms provide. And diversifying business models — subscriptions, tipping, paid features — has been a slow and difficult transition for companies built around advertising.

Summary

“Scrapping algorithmic sorting would change social media. It wouldn’t make it better. The more fundamental problem is a disconnect between the things that platforms are financially incentivized to optimize for and the things that users, and society, might actually benefit from. The most promising experiments in progress – user-controlled feeds on Bluesky, non-profiling options required by the DSA, alternative business models – all point in the same direction: giving users more real agency, rather than the illusion of it. It is less a question of technology than business and politics whether the major platforms follow.