The algorithm doesn't have to destroy us

We can build internet platforms that actually work.

This is an interview with Hamish McKenzie for Internet Sovereignty, nine writers exploring the future of the internet through an essay collection and print pamphlet. Support the series by collecting the digital or print pamphlet. 👇🏻

Hamish McKenzie is the co-founder of Substack. Here is our interview about how we can build better internet platforms.

Elle Griffin: Everyone is on Substack now. Politicians are here—like Pete Buttigieg. Celebrities are here, like Lizzo. People are moving from social media platforms like Instagram and Twitter and writing long-form posts and videos directly for the people who follow them. How does it change the world when we change our platform?

Hamish McKenzie: There’s a Charlie Munger quote, “show me the incentives, and I’ll show you the outcome.” And for at least 20 years, we’ve been stuck with media platforms that have bad incentives and warp our culture.

Substack, of course, is the only pure and perfect media platform, and that will lead to a cultural renaissance. That’s a tongue-in-cheek statement, but what I do believe is that Substack is an attempt to set up a media system that has better incentives, where the people participating in the in ecosystem—the publishers—are incentivized to win and hold on to the trust of their audiences, and then the audiences actively participate in helping the culture and the makers of that culture flourish.

The platform undergirding it all, in this case Substack, is forced by its business model to act in service of the publishers and the audiences. We only make money when publishers make money. And publishers own their audience relationships, and those audience relationships exist on a mailing list that they can take with them anytime they want, anywhere. They own all their content, which they can export with a click of a button.

As a result, the culture of the platform becomes something that’s much more about trust, relationships, and quality and depth. Which is not to say it’s going to be perfect, but it’s going to be completely different from the media platforms that have dominated for the last 20 years.

Elle: Being sold on the economic model already, what do you think changes when a politician doesn’t make their next announcement on Twitter, but on Substack instead?

Hamish: When a politician or an artist shows up to speak in the space, they have a much better chance of being understood with nuance and context. With other platforms, nuance and context are ripped from the conversation, and it’s forced into a soundbite that is thrown against the wall.

And it’s not because Substack is a team of geniuses, although there are many brilliant people who work at the company. And it’s not because the technology of Substack is somehow superior to all other technologies. It’s because the incentive system is different.

I don’t think Tiktok, Twitter, Instagram, YouTube, Facebook, and LinkedIn are going to go away, at least not overnight. But increasingly, I think the places that aggregate audiences and control the relationships and force people into a feed that is optimized for advertising are going to produce different types of culture than what you see on Substack. It’s going to be much more superficial, fun, and entertaining, but also lightweight.

That’s not where we think deeply about something and spend time with an idea. But now there is a place where you can spend time with an idea and think deeply about something and engage in genuine, thoughtful discourse with other people. I think you’ll see more thoughtful people and more thoughtful culture being rewarded.

Elle: I agree. Recently, a Utah citizen decided to run for Congress and posted an introduction video on Twitter. This person was of Somalian descent, and Twitter just went insane—2 million hateful comments were calling on ICE to deport him. I have to imagine that if he had published his video to Substack instead of Twitter, the response would have been very different. How can our platforms solve for hate?

Hamish: The first thing to note is that Substack is not perfect—it’s not Disneyland. It’s not like there are never bad things being said, or that there are never fights. But it is, I think, at least an order of magnitude different in terms of civility and the way people engage with each other.

For years, we have labored under the illusion that the problems related to speech online could be addressed or solved by a more sophisticated content moderation apparatus. And I think what we’ve seen is that as companies and platforms have invested more in trying to refine their content moderation apparatus, as they’ve hired more people to act as content moderators, as they’ve made their policy documents more complicated and precise, and as they’ve spent billions and billions of dollars in trying to get the system to solve or address the problems, we’ve only seen those problems get worse. Or at the very least, they haven’t gone away and they haven’t gotten better.

So we suspect that the problem has to do with the incentives of the systems that they have set up. A system like Twitter incentivizes the types of actions and behaviors that have everything to do with getting attention at any cost and nothing to do with helping us understand each other or build trust in each other, or seek truth. If you create a different system where you’re incentivized to help each other understand each other, which I think direct subscriptions and direct relationships create the best proxy for, then you’re going to get a totally different outcome than the other systems.

What Substack has shown over many years now, for many millions of users, and for an extreme diversity of ideologies and perspectives, is that you can have a hands-off approach to censorship while also having a healthy culture and productive discourse. That doesn’t mean we don’t do any content moderation. We do do some, and there are rules that protect the platform at the extremes, but it has been shown, I think, that the incentives matter even more.

Elle: Does that change artist and celebrity culture? Do we still need social media and paparazzi when artists and celebrities can speak directly to us with full nuance and context?

Hamish: That’s what we are seeing—Lizzo or doechii or charli xcx being able to express themselves in ways they simply can’t on other platforms. The other platforms flatten artists and celebrity voices, as well as academic voices, but Substack can give context and color and contours and result in something just fundamentally better for anyone who cares about quality and understanding. I don’t know if that’s going to become the dominant way that people interact with culture, but it’s going to get larger than it is today, and it’s going to be more present than it has been in the past.

Elle: Substack announced an app for smart TVs, which I think is an incredible replacement for YouTube. If YouTube were subscriber-based, and I could pay for my favorite people right there and watch the things that they’re producing for me, rather than whatever the YouTube algorithm wants to surface to me next? I think that could change the culture a lot. Do you agree?

Hamish: It can change the culture a lot, yeah. Even inside Substack, we underestimate the potential for reshaping culture as a result of the direct relationships model. I don’t even think we need to replace YouTube to change the culture in a positive way.

There are three big powerhouse media models at the moment. There’s Netflix, which is an aggregator—you pay for the platform directly, and then the platform decides what you see and where the money goes. It’s a successful thing, serving hundreds of millions of people and making lots of money. Then there’s YouTube, and anyone can publish there, but it’s still an aggregator. Its revenue model is split between platform advertising and platform subscriptions via YouTube Premium. But YouTube still controls the relationships and still owns all the communities that happen there. People show up in the comments underneath your videos. But that’s not your community, that’s YouTube’s community.

What has been missing all along is this direct relationships platform or ecosystem. Substack has made it work for blogging, and writing is arguably the hardest thing to commercialize. Being a writer was the hardest way to make a living when we started Substack in 2017, now we see writers becoming millionaires based on this direct relationships model. So I’m really excited to see what can happen with the direct relationships model when it’s applied and scaled to things like video, including TV shows, films, news, studios and curators, independent filmmakers, and so on. Where it’s not just a format type that unites people, but it’s a model type that unites people. It’s direct relationships, and an artist ownership model.

Elle: I could see a world where a lot of platforms start to head in that direction, and that excites me. Whether it’s video or reading or social media, changing the economic model and the relationship model changes what you see. But Substack has introduced an algorithm that surfaces people we aren’t following, which I’m sure will be part of the TV as well. Even in the Substack app, the video feed shows me people I don’t know, and that I don’t subscribe to. What have been the risks and benefits of adding the algorithm, and what have you learned about how to make that work for the good?

Hamish: The algorithm is a bit of a boogeyman. It’s a scary word, right? But the important thing isn’t whether there is an algorithm, it’s what that algorithm has been asked to do.

The algorithms everyone has become skeptical of are based on the last 20 years of social media, where the algorithms were asked to addict you to the feeds so that you can see more ads. But the only way Substack makes money is when publishers make money. The only way publishers make money is by getting subscriptions from their audiences. And so the algorithm is trying to drive people into those deeper relationships so that they might result in subscriptions, so that they might result in money to the publishers, so that might result in money to the platform, which can then improve the whole experience for everyone.

So the algorithms of Substack are very different from the algorithms of other social media platforms. They’re doing a different job.

Elle: Since Substack Notes debuted, what have you learned about making the algorithm work in a way that drives relationships?

Hamish: We started out with the mindset that people wouldn’t want short posts or videos or images. Here, they’re serious readers. I’m a serious reader and I just want the good stuff. I’m sick of those other spaces that are vying for my attention. I just want to see the posts that are going to force me to be on my best behavior.

It turns out that is not the best way to help people enter into deep relationships with writers and creators, and that a browsing experience helps. Introducing someone to an idea or a person that can deepen over time through repeated exposure will actually lead to more subscriptions for writers down the line. So that was a key insight.

And the types of activity that convert into subscriptions are not just relentless promotion of a specific piece. It’s, “Oh, I’m interested in this person’s mind. They’re showing up in an interesting way in this feed. Can I find out more about that person? Maybe I’ll follow them. After following them for a while, maybe I’ll invite them into my inbox with the subscription. Later on, they’ve convinced me so much that I’m going to pay them.”

The algorithm for Substack is trying to do that job.

Elle: I will frequently see somebody hating the Substack feed for this reason or other. And then Substack design head Mills Baker will pipe in and say, “We keep trying to make that kind of content not show up, but you guys keep making it show up because you want it.” There’s a feedback loop where the things people say they don’t want are also what they’re clicking on and liking, which makes the algorithm hard to work for everyone. How do you solve for this?

Hamish: Historically, one of the limitations of a system like this was a response to the user’s activity. So if you have been enjoying a lot of horse photos recently, you’re probably going to be shown more horse-related things in the next little bit. And sometimes, people just browse shitty shit and then they get lured into something that might be a guilty pleasure at first, but isn’t content they’d choose as their best self.

But with machine learning and improvements in machine learning and LLMs, we can actually get more sophisticated than that. We’re at such an early stage with this, but we’re trying to make sure that we give people more and more tools so they can set their own conditions within the app. For example, you can choose the feed that is purely machine-driven, or you can choose the “following” feed which is only going to show you the people you follow or subscribe to. And it stays persistent so the next time you pick up the app again, it will remember the feed that you said you preferred.

We’re still testing features like this, but we’re rolling it out more and more. If you don’t like a note for whatever reason, you can click an X in the top corner and choose a reason why you didn’t like it. For instance, “I want to see less political stuff or slop or rage.” Then there are extra signals that give Substack an advantage over other platforms, like the subscription signal. If you subscribe and pay money to a particular writer who’s an expert in a particular thing, it’s quite likely you’ll want to hear from that person more than other people that you’re following. And there’s a good chance that you’ll be interested in who they recommend.

So there are concentric circles of subscription interests, and though there are some limitations to machine learning based algorithmic feeds, Substack’s infrastructure provides the opportunity for refining a feed that really works for the end user with greater depth and possibility than what other systems can offer.

Elle: Let’s talk about Cory Doctorow’s Enshittification. Everybody is entirely stressed out that it’s going to happen to Substack, because Substack is everybody’s favorite app and all our past favorite apps are no longer our favorite apps. How do you avoid it?

Hamish: If you look at the top revenue earners on the leaderboard across all categories, you’re not getting people who have turned up to win internet points; you’re getting something of depth and where there’s real community. I think this is a system that’s much more resistant to the enshittification that is rampant on every other platform.

And maybe one day Substack will become vulnerable to the forces of enshittification, but it’s not going to happen at its current scale, which is millions of users in the app every day. Facebook went for decades before the word “enshittification” was ever uttered, and with a model that was focused on advertising and had billions of users. So there’s still plenty of runway and time before Substack reaches that point.

People get scared when things change and new things arrive because it’s messing with the original conception of what the thing was. But over time, I think largely people have discovered that they can live in the way they want to live on Substack. They don’t have to watch short-form video if they don’t want to. They don’t have to read Charli XCX’s essay if they don’t want to hear from celebrity voices.

Elle: I love Substack’s expansion into video and podcasts because I think the world needs Substackification. But I’m also nostalgic for and miss when Substack was just writing and reading and commenting and that was it. So I both know that Substack needs to grow and reach more people and have more features for different kinds of creators, and I’m nostalgic for a time when it was just focused on writing and my own niche. How can Substack be the app for everyone, as much as it is the app for each individual niche?

Hamish: For people like me who are reading-oriented and writing-oriented, I can still have that experience on Substack. In fact, I have several escape hatch options. I can still be in the app if I don’t mind Notes—and I don’t, because I like discovering new things that are cool to read. But I can also keep my inbox as the place where I have my experience with writers and reading a text. I can even make my inbox the primary tab in the app, so that I’m being nudged to focus on reading the writers I really deeply care about, and then Notes becomes an afterthought.

Or I don’t have to be in the app at all. I can go to individual websites and follow writers only through email. There’s no other platform that allows that flexibility or diversity of engagement styles. So while I’m somewhat sympathetic to the people who would rather Substack remain the purest text platform, my message to them is you can still have that, and unlike other platforms, you have the choice.

Elle: Last question. This is for an issue called Internet Sovereignty, it’s a bunch of writers exploring the future of the internet. Even if the whole internet worked like Substack and we’d achieved your nirvana, what would you still want to change about the internet? How would you want it to be better?

Hamish: I still want real-life, person-to-person interactions to be the primary mode of social engagement, and the internet and online discourse to be a distant second to that.

Online communities should activate real-world interactions. There are certain things you could build into platforms that would actively encourage that, like selling tickets to events. Or helping them run good meetups, happy hours, or gatherings.

When I used to work in magazines in Hong Kong, I was a music editor and part of the job of making a good music section was to throw good music events and show people that indie music was something to love and appreciate. Hong Kong was not an indie music town back then, so we held a competition for indie bands. People would send in demos and the winning band got world-class producers, engineers, and studio time to make this beautiful album. We threw a big party for them at the end of it, and that launched musicians onto the scene. One went on to become a bestselling pop act in Hong Kong.

That’s the sort of thing that conveners of culture can and should have outsized influence over.

So more of that.