Will AI End the Newsfeed?

When I started reading this report, I was skeptical about the potential for AI to disrupt social media feeds wholesale.

I can see AI tweaking how those feeds operate, in the way that AI-generated summaries are now prominent in the Google Search interface, and how Google re-engineered its algorithm from PageRank to powering searches through Gemini. I believe further overhauls to the interface are on the way, effectively converting Google Search into a chatbot agent. Ultimately, people are still using “Google Search”, in the same way I’d expect them to continue using Facebook, Instagram, TikTok, YouTube and X. After reading the report, my skepticism remains, but I’m excited both to investigate some of the premises they put forward, and the emerging tooling and initiatives that may provide credence to those premises. The authors claim the social media feed is no longer the default interface (or more generously, that it soon won’t be). People will still go to the newsfeed interface but increasingly look elsewhere for their social lives and information diet. I would at least draw out four distinct functions people look to on a newsfeed (and I’m sure there are more): information seeking, entertainment consumption, content creation, social connection and bonding. The first two are inherently consumption-based for users, the third is production-based, and the last is interactive.

AI certainly influences both consumption and production on social media platforms in disruptive ways. The piece makes compelling points that AI both makes content creation and participation more accessible, but in doing so undermines both quality and trust in that content. It is also plausible that it undermines feelings of connections here people believe they are talking to bots – given the difficulties in verification the report describes, it is close to impossible for users to actually know if they are speaking to bots. However, the perception they might be is damaging enough to trust in relationships, in the same way that the liar’s dividend undermines epistemic trust in information more broadly.

However, people continue to use social media platforms despite negative experiences with them. 64% of Americans said social media had a mostly negative effect on the country in 2020, 48% of teens in 2025 said social media had a mostly negative impact on people their age, and 44% of parents thought social media was the single factor that most negatively affected their children’s mental health. Despite this, ~50% of U.S. adults used Facebook and YouTube daily in 2025, and 45% of teens said they spent too much time on social media. Connection is cited as a leading reason for continued use, with 74% of teens saying the platforms made them feel more connected to friends.

Connection, alongside ingrained habits, are likely to make it extremely difficult to shift people off established platforms. The piece speaks to the promise of AI and emerging initiatives to allow users to shift their networks from existing platforms, though I think the barriers to this are worth taking seriously. As the report itself alludes to, defaults are sticky, and profit-seeking platforms invest heavily in research to intentionally set those defaults and increase the friction of changing them. Simple behaviors, such as deciding which app to open to start consuming a newsfeed become deeply reinforced and engrained after thousands of repeated actions, so much so that we often start browsing our Instagram, TikTok, or Facebook feeds before we are even conscious of what we are doing. Moreover, social graphs are two-way, too; not only do we have to shift, but so do our connections on these graphs. Even if it becomes technically straightforward to migrate our networks – and my current understanding is that it is not – this is still a significant coordination problem, even to create the smaller communities of 500 or so users the piece conceives of. Try asking any Android users persuading their iPhone owning friends and relatives to message them on WhatsApp instead of via text!

That said, there are several promising avenues for my own learning and research to explore here. I want to investigate the extent to which micro-communities are emerging, under what conditions, and who is building and joining them. I see this as a potentially to come back full circle to the internet of my childhood, where people spent time on decentralized forums and websites before the tech giants consolidated everything into the platform oligopoly. I want to learn more about the specifics of decentralized protocols, which may provide technical solutions to address some of my questions and concerns in the previous paragraph. There is a lot of exciting work leveraging the BlueSky protocol I want to look into further. I want to understand if there is potential for these micro-communities to connect with each other to support bridging and trust-building across diverse groups. Furthermore, I want to build on my community moderation work to understand if AI can support constrained professional and volunteer moderation efforts, especially given the increasing scope and scale of threats that AI also enables.

This is emergent work that will start off with a review of the literature and industry landscape. Alongside traditional academic work, I will also look at trade publications, speak with industry experts, and talk to developers of emerging tools and initiatives, given the fast pace of development. Alongside this, I would like to document the features and tools I come across, with the aim of eventually systematizing it into coherent form. After identifying what I believe are the most promising design features and understanding the mechanisms through which they can foster prosocial outcomes, I will look to contribute to an online field experiment to test this framework. If any of this sounds interesting to you, please get in touch.

In sum, I want to understand if emerging AI-enabled tools and initiatives constitute something that is incremental or disruptive to existing platforms. Whether Facebook and X go the way of Google Search, building in interface changes and affordances that leverage AI, sometimes to the benefit of the user, but always in pursuit of greater engagement and profits. Or alternatively that we see a genuinely new, more decentralized age to the internet.