How Substack Should Use Algorithms
Since it seems to be happening anyway, we should design our feeds differently this time.
Let’s face it, social media algorithms suck. At least they have so far.
Algorithms have oversimplified and encrypted the Internet, and with that, people have seen fewer ideas and less connectivity over time. The growth of algorithms has correlated with a decade-long erosion of what many believed the Internet could be. The good news is that the core experience hasn’t changed, only the doorways and windows that limit what most people can see.
The problem is that current algorithm design has left many of us searching for the sense of optimism, collaboration and authenticity that was present in the web’s early years. But we can’t go back, and we can’t completely avoid algorithms. Even Substack, a place for the eccentric curators and outcasts, is now experimenting with one for its Notes feature. If they need to do this, I really want them to get it right.
5 Rules for Doing Algorithms Better than Social Media
1) Algorithms should remember and reward diversity of thought from its users.
Algorithms oversimplify our interests, and we hate being put into a box. The only way to align with this reality is to avoid short term recommendation loops in favor of constructing multiple, long-term threads that can more effectively follow someone’s varied and evolving interests. This would help avoid your feed assuming too much, which causes algorithms to hide off large parts of the Internet from us. We are in need of an algorithm that rewards individuality, not conformity.
2) Algorithms should optimize for replies and saved posts rather than likes and reposts.
“I really don’t want to forget this.”
“I want to refer back to this for something else I am working on.”
“I know a friend who would love this.”
This one is simple. More of this type of user experience would make Substack’s algorithm a powerful tool for knowledge management and connection.
3) Algorithms should connect individuals of similar network size, and limit broadcasting its largest voices.
Current algorithms have an inequality issue. There are a top percentile of posters who dominate “For You” feeds and then there are the rest of us. Now, I’m not conceited enough to think that Substack should give me an algorithmic boost. Quite the opposite. The algorithm should surface my work to other people who are shouting into the void in a similar way. Those people would be more likely to follow back, or have an open conversation in the comments. This is a necessary prerequisite for developing small online communities, and it would give Substack something unique from other platforms.
4) Algorithms should be as customizable as possible, and this customization should be encouraged by the platform.
The option to see more of one thing or less of another is as far as most social media algorithms have gotten on this topic, and it rarely works that well. If algorithms are based on user activity parameters, why not give users the option to further tweak what they’re seeing? This sense of ownership would increase platform loyalty, and engineers will learn even more about user preferences. Think video game settings, but more robust!
5) Algorithms should be disposable and transportable.
I don’t expect this rule to win over anyone who holds a business degree, but algorithms are our digital echoes, and we should be able to reset them, or apply them to other platforms, without being punished. A single business cannot own your virtual shadow. The pro-business way of looking at it is simple. Instead of optimizing for user addiction which is a chaotic and zero sum game, you would simply compete on different battlefields.
Did I miss a rule? How are you curating your social media algorithms today? Is Substack making a mistake by becoming too much like other platforms? Let me know in the comments.