The End of Social Media and the Rise of Recommendation Media

Why friend graphs can‘t compete in an algorithmic world.

Photo by Jan Tinneberg on Unsplash

Hi friends, Nathan here! Today we’ve got another essay by the one and only Michael Mignano, co-founder of Anchor and former Head of Talk at Spotify, who published an essay here last week about how open standards can stifle innovation. This time, Michael is back to talk about how algorithmic feeds (like the one that drives TikTok’s “For You” page) are going to largely replace traditional friend graphs across all social networks. It’s a fascinating topic and one we’ll all be affected by over the coming years. Mike’s take on this is incredibly smart, and I am stoked to share it with you. Enjoy!

Last week, Meta announced that the Facebook newsfeed would be shifting towards an algorithmic, recommendation-based model of content distribution. This announcement marked the most recent example of a major platform to formally make this shift, while other major platforms, including Meta’s Instagram, have been headed in this direction for a while. Given Facebook’s relevance as the world’s largest social network, this change signals the end of social media as we’ve known it for the past decade and a half.

Want to watch or listen to this essay instead of reading it? I’ve also published it as a podcast on Spotify and YouTube.

There has been backlash. Kylie Jenner, one of the world’s most influential users of social media, recently posted about her displeasure with Instagram prioritizing recommended videos over photos from friends. With more than 360 million followers on Instagram, Jenner’s influence can’t be ignored; the last time she complained about a change to a social network, Snap’s stock price fell by 7%. It’s therefore likely no coincidence that Instagram’s CEO, Adam Mosseri, posted a video discussing some of these recent changes and plans for the future. In it, Mosseri acknowledges that the world is changing, and that Instagram must be willing to change along with it.

And yet, these shifts towards algorithmic feeds over friend feeds make sense. Platforms like the massively popular (and still growing) TikTok and YouTube put far less emphasis on friends and social graphs in favor of carefully curated, magical algorithmic experiences that match the perfect content for the right people at the exact right time. This is recommendation media, and it’s the new standard for content distribution on the internet.

But first…what is social media?

Social media is content (text, photos, videos, audio, etc) that is distributed primarily through networks of connected people. This means that some level of distribution is guaranteed for creators based on the creator’s social network of friends or followers. This dynamic puts an enormous amount of power in the hands of creators because it means they have built in audiences to which they can broadcast content. In social media, creators have the programming power. As a result, social media is effectively a competition based on popularity, not on quality of content. It favors the creators with the biggest followings; the bigger the following, the bigger the potential for distribution and influence.

Through this distribution dynamic, social media platforms are able to scale extremely quickly. If a platform can build a social graph (which, in the earlier days of social media, was extremely challenging for platforms but has become increasingly less so over time), it can automatically have a built in distribution system for serving engaging, highly relevant content to massive audiences.

The cost of social media

But just as massively as social media platforms have grown and changed the way we all consume content, they have also wreaked havoc for platform companies, the internet, and more broadly, our world.

Built-in distribution for content to social networks has meant that people can share and spread problematic content just as easily as they spread good-natured content. If a bad actor wants to share problematic content on social media, the content can spread fast because of the guaranteed distribution to the person’s network of friends. Furthermore, because content is primarily distributed to clusters of connected people, there is huge potential for echo chambers of groupthink on social media. Diversity of thought is, by design, at a disadvantage in social networks. When it rarely finds its way in through open comment sections, it’s often met with fierce opposition and resistance, creating polarizing arguments and conflicts, sometimes among some of the most powerful people in the world.

Social media has also proven to simply not be that efficient in terms of matching high quality content with a relevant audience. Just because people can easily distribute content to their friends or friends of friends doesn’t mean that that content will be interesting or relevant to the consumer. This is why, over time, social networks have started not only distributing content based on social graphs, but also based on how much engagement content has received within those social graphs.

The above problems with social media in turn generate massive costs for platforms, in the form of gigantic moderation teams made of tens of thousands of people, severe damage to platforms’ brands, and openings for competition to find more efficient means for distributing content. And no platform has been better at exploiting the weaknesses of social media than TikTok, the platform which popularized algorithmic content distribution and gave birth to what I call, recommendation media.

Enter recommendation media

In recommendation media, content is not distributed to networks of connected people as the primary means of distribution. Instead, the main mechanism for the distribution of content is through opaque, platform-defined algorithms that favor maximum attention and engagement from consumers. The exact type of attention these recommendations seek is always defined by the platform and often tailored specifically to the user who is consuming content. For example, if the platform determines that someone loves movies, that person will likely see a lot of movie related content because that’s what captures that person’s attention best. This means platforms can also decide what consumers won’t see, such as problematic or polarizing content.

It’s ultimately up to the platform to decide what type of content gets recommended, not the social graph of the person producing the content. In contrast to social media, recommendation media is not a competition based on popularity; instead, it is a competition based on the absolute best content. Through this lens, it’s no wonder why Kylie Jenner opposes this change; her more than 360 million followers are simply worth less in a version of media dominated by algorithms and not followers.

A better consumption experience

In recommendation media, the best content for each consumer wins. This means that consumers are always being recommended and actively served content best suited for them, creating a superior consumption experience at all times. Whereas in social media, people see content from their friends regardless of the quality of the content, in recommendation media, content distribution is optimized for engagement. This results in very little waste in a feed, and consumption patterns are highly efficient.

Platforms also get to decide what’s popular and when. In social media, creators maintain programming power over what gets seen and when. But in recommendation media, the platform is always in control. This is similar to how cable television networks and radio stations have operated for decades; they program all media based on editorial and business decisions. However, on a platform like YouTube or Instagram which contains billions of pieces of potentially programmable content, programming can occur across a multitude of dimensions, such as any user’s interests, demographic, or location.

Less trust and safety risk

Since a platform is in control of what content gets served to who and when, there’s no expectation that a creator’s social network is guaranteed to see their content. Therefore, platforms can also choose what not to program, and there’s little creators can do or say to counteract this. Long gone are the days where a creator can complain about being de-platformed or shadow banned because their followers aren’t seeing their content; in recommendation media, the algorithm is understood to be the final decision maker about what gains traction and what doesn’t. This gives platforms far more leverage to hide unwanted content and therefore reduce the need for massive moderation teams. It’s not that these teams are no longer needed; they’re simply not needed to the same scale as in social media because distribution for certain types of content can be eliminated from a platform without changing the underlying structure of content distribution.

Massive growth potential for platforms

Since there’s no guaranteed distribution for content through friend graphs in recommendation media, creators are incentivized to seek engagement elsewhere when they’re not getting it from the platform where they created content. Where do they turn for that engagement? Other platforms. This is why you often see so much TikTok content being shared to platforms like Instagram, Twitter, and Facebook. Creators are sharing content to networks where they already have audiences.

This has a second order effect of driving massive growth to the original platform. As an example, each time content from TikTok is shared on Twitter, a user who wants to consume that content clicks through to consume it on TikTok. This not only drives engagement on TikTok, but when the content consumer isn’t already a user of TikTok, it drives new user acquisition as well. Now imagine this dynamic occurring tens of millions of times, each time someone shares content from a recommendation media platform, and it’s easy to see how this can result in sky-high growth potential.

More defensible

In addition to the drawbacks of social media mentioned above, social networks are simply no longer defensible because the underlying data that powers them, the social graph, has become commoditized. By leveraging login APIs from Facebook or Twitter, or even connecting a product to a user’s smartphone address book, teams can now quickly spin up social networks through which they can distribute content based on social graphs.

But in recommendation media, the algorithms that power distribution reign supreme. These algorithms, which are powered by machine learning, are unique, valuable, and grow in power and accuracy as a platform scales. Therefore, only the biggest and most powerful platforms can afford investments in the best machine learning algorithms because they are such expensive and resource intensive assets. In recommendation media, the platform with the best machine learning wins.

What comes next?

With Facebook formally pivoting to recommendation media, it feels like a new era of the internet is upon us, and it’s hard to imagine what might come next. But just as we’ve seen in previous generations of the internet, platforms will always seek more efficiency as technology becomes more advanced. Here are a few predictions for where the world could go next.

Professional media will turn to recommendation media

Given the strength of recommendation media platforms like TikTok and YouTube, and the way traditional social media platforms are chasing them, it seems likely Professional Media platforms (such as Netflix) may try to follow suit (in fact, Netflix’s co-CEO, Reed Hastings, may have even foreshadowed this when he famously stated that his biggest competitors were TikTok and YouTube, both of which are open to any creator). However, in order to be able to match the exact right content with the exact right person, a platform needs an ocean of content, including extremely niche content for every person on the planet. The only way to have that much content is to be an open creation platform where users of the platform are able to create on the platform. So, I expect Netflix and similar platforms to let anyone create, not just the professional studios.

Platforms will seek even more control

If recommendation media is about platforms having more control over the consumer experience, it’s not hard to imagine that platforms will ultimately seek even more efficiency by making their own content. We’ve seen professional media platforms do this on a smaller scale (e.g. Netflix making originals, etc). But to do this at the scale of an open creation platform, such as TikTok or Instagram, platforms won’t be able to rely on humans. They’ll instead need to rely on machines to create AI-generated media, or as my friend Matt Hartman calls it, synthetic media. Recently, OpenAI’s DALL-E 2 has shown the world just how powerful and human-like synthetic media can be, but it’s unlikely these capabilities will stop at still images. As the cost of AI content-creation solutions come down, I expect platforms to create more synthetic media over time to create even more perfect fit content for the right users at the right time.

RIP social media

Recommendation media is here. As a result, we’ll make fewer explicit choices (“these are my friends”) and more implicit choices (“this is where the algorithm recommends I should spend my attention”) about how, when, and why we consume content. In the near term, we may not notice much of a difference, but it’ll be fascinating to look back a few years from now and reflect on how our personal behaviors have changed.

What do you think? Is social media gone for good? Or does this create an opportunity for a challenger to take a contrarian approach and bring social media back from the dead? Get in touch with me on Twitter or LinkedIn to let me know.

Michael Mignano is a technology executive and entrepreneur, most recently serving as Head of Talk audio for Spotify, where he was responsible for leading the podcast, live, and video strategies for the world’s leading audio streaming platform. Prior to this, Michael co-founded Anchor in 2015, which was acquired by Spotify in 2019, and is now the world’s largest podcasting platform. Michael is also an active angel investor and advisor to 50+ early stage, technology-enabled companies.

Like this?
Become a subscriber.

Subscribe →

Or, learn more.

Thanks for rating this post—join the conversation by commenting below.


You need to login before you can comment.
Don't have an account? Sign up!

Hi Michael,

Thank you so much for sharing great insights on social media vs. recommendation media.

I also think that recommendation media will dominate or play a major role in how we consume information in the future. As you mentioned, the keys are how good the ML is and how open the platform is for creators.

One thing that I believe is that "curation can be another creation". If one's curation process is open and can be treated as a creation for other people's consumption, then it'd be a fun way of learning with each other and accessing relevant information (I call it a knowledge circulation economy).

I think users should be able to have the option to choose if they want to find content through ML-based recommendations or from their friends or people who they follow. Though the social graph is no longer defensible, having both options can be a plus in terms of optionality.

Also, in the area of knowledge management, the problem with social media is the ephemerality. We are learning a lot from other people on the current social media or even on recommendation media but they need to have a solid way to manage what they learn. These platforms play a good role in sharing knowledge, wisdom, insights, etc but without having a solid system, we'll forget.

I'm actually addressing this area with Glasp (, a social curation and knowledge management platform. We use both ML-based recommendation and social media types of content distribution.

I don't usually leave a comment but this post really resonates with me so I'm writing down this.

Thank you for sharing this!


Justin Jackson almost 2 years ago

"In recommendation media, the best content for each consumer wins."

Best, in what way?

If you make junk food designed to be addictive people will consume large amounts of it. It might be the "best" in its desirability, but it's not the "best" for producing healthy humans.

I'm seeing AI-driven the same way. When I use an app like TikTok, it “entertains” me for hours. But I wouldn’t describe that experience as healthy or good.

What's even more concerning is *who& will control this AI-driven future. As you said on the Every podcast:

"The platforms that will be most successful here are the ones with A) the most data, and B) the most computing power. If recommendation media is all about making AI-informed recommendations, basically the platform with the most computing power wins."

So we're forecasting a future where we're all addicted to AI-driven content, where companies like Meta, Google, and Amazon own the algorithm? No thanks.

Giving people a never-ending stream of content designed to keep them on a centralized platform longer where the algorithm is controlled by some megacorp (and its shareholders) isn’t my idea of a great future.

It’s dystopian.

Michael, I don't think you are wrong, but the horse has left the barn. FB has been doing algorithmic recommendations for a while now, just maybe not as explicitly. They have more tuned it to advertising, so people aren't big fans. But Amazon, and Spotify, have been doing recommendations for ages. And you worked at Spotify but my students have found the reco engine is only so-so. (Youtube is worse though.) Also, Google has really stepped up their recommendation game, so that so many things that you get in your search results are recommendations. You don't even need to go to the website. But social media isn't dead, not by a long shot. BeReal is a new example of a much smaller focused social media network that is quite popular. I think the interesting take is the way TikTok is eating everyone's lunch. You, me and many writers are in awe of the way it works even though it really is a mash-up of other social media. I enjoy TikTok a lot, but there are challenges with the algorithm. For example, my TikTok is amazing, but if you have never used it before, the algorithm is terrible at first. It learns, but obviously, the pain for many people will not get them to subscribe. Algorithms work better when they have information, like your social media network, which is why algorithms and social graphs work so well together. Chances are you might like what your friends like. Also, Instagram (owned by FB) is like the poster child for flavours of the month. They have lost their way and they have been slavishly copying other social media, as directed by FB, in ways that have made the platform annoying and obviously upsetting for content creators. It feels like the decline is more at FB and Insta than the sector as a whole. Also, sure, change is good, as Mosseri said, unless it sucks.

Derek Schlicker almost 2 years ago

Great article and appreciate the write up. I think its fair to point out though, that Youtube has been a recommendation platform since practically day 1, and has yet to really solve the problem of bad actors spreading misinformation, hate speech, or politically extremist rhetoric on the platform. So if we compare social media's ability to self-regulate or solve that issue (vs recommendation media), I actually think social media at its purest (i.e. I just want to be connected to friends and family) feels like a slower spread-rate than recommendation platform.

I've seen too many experiments showing how fast it takes (via Tik Tok or YT) to go down a rabbit hole of extremism, hate speech, etc. As a somewhat innocuous example, I somehow got stuck in ex-mormon Tik Tok (no idea how). Its pretty easy for the algo to send someone down a random thread and if that thread is Q anon or flat-earth conspiracy theories, the likelihood that person is groomed into a certain way of thinking is very very high.

If anything this trend tells me that behavior will get worse before it gets better.

@gr.fromage almost 2 years ago

I get the efficiencies and the benefits of moving from the social graph to recommendations. In some ways it's worrisome because of the potential for algorithmic bias, intentional and unintentional. I wonder how or if humans will oversee diversity of ideas, to make sure diverse and even dissenting ideas are discoverable. It's good to evolve past the social graph for the reasons stated, but recommendation engines are definitely going to have equivalent issues and won't be a silver bullet.

Or Dubinsky almost 2 years ago

I'm also concerned with the lack of diversity but even more so with inhibiting innovation. "recommendatio media ... is a competition based on the absolute best content". I think the use of the word 'best' there is horrific. Best for what? Seems like we've completely forgot the value of relying on a trusted source. Also ML prioritizes duplication, not exploration to new areas. YouTube used to be great for exploring new genres. Now all Discovery features there as well as Spotify just self perpetuate users existing preferences. Calling it 'Best' is alarming

@michael_3992 almost 2 years ago


@michael_3992 almost 2 years ago

@michael_3992 oops sorry, I didn't realise I'm logged in. Was curious what happens when I click "post" as an anonymous website visitor. Hence this test post.

Interesting article. I'm working on something related (a writing platform where you choose purposely what you want to read next, just like you pick the next book in a library). It's too early to show something but I'll follow up here once ready. :)

Every smart person you know is reading this newsletter

Get one actionable essay a day on AI, tech, and personal development


Already a subscriber? Login