Echo Chambers

Online polarization is a well-researched, reported-on problem in the United States and all over the world. It has been leveraged to sway elections and public opinion. It has been used to sell you things you don’t need (guilty here). I have a theory to mitigate online polarization. I think that I may have stumbled upon a weakness in the armor of social media algorithms, and I want to try to exploit it in an effort to build understanding and solidarity, and I will explain it here. Fig. 1 shows a simulation I ran of online polarization on Twitter. The red and blue dots represent people across the political spectrum. The lines between the dots indicate the two users are following each other. When lines are removed, that indicates that the users unfriended each other. This model, described in a 2021 paper on Twitter users by Sasahara et al., allows you to tune the following knobs: Tolerance: How open Twitter users are to new opinions. Influence: How likely users’ opinions are to shift given the new information from shares. Unfriending: How frequently users sever connections with people sharing posts they disagree with. In this model run, I set all of these parameters to “mild” settings: so I said that users are reasonably open to new opinions, that they are weakly open to being influenced by opinions shared with them, and that they occasionally unfollow users they disagree with. The exact settings were: Tolerance: Medium Influence: Weak Unfriending: Sometimes Even with these “mild” settings, the users still separated into disconnected echo chambers. This demonstrates the evolution of echo chambers. In the next model run, shown in Fig. 2, I set the following: Tolerance: High Influence: Strong Unfriending: Sometimes This model run shows that with users who are more tolerant of new ideas, who are more open to being influenced by new opinions, and still sometimes unfriend other users, a general agreement can be built. Under these circumstances the population converges rather than separating into echo chambers. You can visit the EchoDemo site, where there is an online version of the model, and tune the parameters yourself to model your own social media environment. Other research, led by Kiran Garimella at Rutgers, gives us a look into how polarization occurs across example topics (see Ref. 2). Fig. 3 shows polarization of users with respect to the benign #tbt (Throwback Thursday) hashtag, and the more polarizing topic of Obamacare. Garimella’s research indicates that there are “gatekeepers,” or users who consume a wide variety of opinions, but who only share content they agree with. This work also indicates that users who share bipartisan material meet friction; these users are less likely to maintain central positions in the social network than the partisan gatekeepers. Together, these two research studies show that, left to our own devices, we humans like and share information we agree with, and we avoid information we disagree with. We create information silos and echo chambers, all on our own. In order to correct the polarizing trends above, we need to deliberately make an effort to reach across disconnected social media silos and polarization corners. Counterintuitively, one tool for doing that may be the same technology that causes polarization: social-media recommendation algorithms. And the way we can do so is by leveraging the advertising used by social media networks. The Meta paid advertising algorithm is extremely good at optimizing to present your ads to Facebook and Instagram users who are likely to buy a product. If I advertise an essay on the biology of human sexuality and gender development, Meta would likely find a liberal-leaning audience of subscribers, and keep showing the ad to users with similar interests and Facebook behaviors. The bridges between the information ecosystems are, in fact, hashtags such as #tbt and interests such as Game of Thrones. Throwback Thursday, Game of Thrones, gym selfies, pictures of food, and our pets are all examples of interests that are independent of politics. So, if I advertised an essay about the separation of church and state to Game of Thrones fans, which I can, I would be advertising to users across the political spectrum. My strategy takes another step: I am looking to share my essays with a deliberately apathetic or opposing audience, who are members of communities which do not normally get to honestly review the ideas in my work. In other words, what if I advertise an essay on the separation of church and state to NASCAR fans in swing states? I am, in fact, doing this. Fig. 4 shows the theory of my strategy. I select audiences with interests and behaviors that normally code as against or uninterested in the ideas I present. Because the Meta algorithm is great at what it does, it optimizes within that population of users to find people within that community who might nevertheless be receptive to the ideas I'm presenting. Instead of fighting Meta’s optimization algorithm, I’m changing the population within which I ask it to optimize. I’m just directing the algorithm by effectively saying: “hey, look for readers over here instead.” This is the social media equivalent of walking door-to-door to persuade voters. It’s analogous to the Democratic politician Pete Buttigieg going on Fox News to debate, so Fox viewers actually get to hear his well-reasoned arguments, which they wouldn’t hear otherwise. It’s also somewhat like those massive Christian billboards you see on the highway, but for public science literacy! Why do this? Because evidence suggests introducing people to new ideas from outside their information bubble softens their stances and moderates political views (see Broockman & Kalla, 2025). Furthermore, if someone in your friend or family circle shares an idea (or a post) with you, you are much more likely to trust that individual. By deliberately sharing my essays within user communities that have interests that do not typically align with the person usually interested in the ideas I present, I’m reaching into polarized information silos, and at least creating the opportunity that interested individuals can connect with and share my work within their communities. In terms of the Sasahara model, I’m asking the Meta algorithm to search an “opposing” audience for people whose tolerance is high enough to not reject seeing my message outright. The goal is not for everyone in the opposing audience to read my work; it is for the algorithm to find the people within that audience who are willing to consider it. I have some preliminary results to share from this experiment. I advertised my essay titled “The U.S. Constitution as a Fault-Tolerant System,” which is an essay about the dangers of oligarchy, to an audience with the following characteristics: Locations: United States: Miami (+25 mi) Florida; Iowa; Maine; Michigan; North Carolina; Ohio; Texas; Palm Beach County Florida, Broward County Florida, Hillsborough County Florida Age: 30 - 65+ Detailed targeting: Interests: Small business, National Football League, North Carolina, Pennsylvania, Michigan, Finance, Personal finance, Pittsburgh Steelers, Texas, Stock, Cleveland Browns, Bass Pro Shops, Scheels All Sports, Ohio, Houston Texans, Maine, Investment, NASCAR Nation, Cincinnati Bengals, Real estate investing, Wisconsin, Dallas Cowboys, Cabela’s, Nevada, Country music, Politics, Economics, Barstool Sports, Personal Finance books, Field of study: Theology This campaign reached more than 12,000 people in the last month and generated 120 free subscribers. The ad achieved a 5.67% link click-through rate (CTR), meaning that roughly one in 18 ad presentations resulted in a link click. For comparison, the ad performance benchmark site Superads reports CTRs of roughly 2–3% for U.S. Facebook traffic campaigns in the media industry during June and July 2026. My campaign’s click-through rate is therefore roughly two to three times this recent industry benchmark. This was done for about $12 per day, and I have also observed that increasing advertising spending increases the number of people reached and subscribers acquired. I have seen Frank’s Free Substack earn 1-3 new subscribers per hour depending on my rate of advertising, which is metered by my level of advertising spending. In the last month I’ve reached over 100,000 people with my ads, and the ads have generated thousands of likes, shares, and comments. Over 1,500 people have initiated free subscriptions. Now, consider the fact that 20,000-100,000 people in a handful of communities decided the 2016, 2020, and 2024 elections, according to the Federal Election Commission (see Ref. 3 below). In my experiments, I have shown my ads to over 100,000 people. In this way, I am much more effective sharing these essays than if I were to cold-call voters or go door-knocking. Showing my essays to 100,000 people doesn’t mean that I’m reaching the few hundred thousand people that will decide our upcoming elections. All I mean to say, is that I am reaching quite a few people, and hopefully those people reach some other people. And if we all take ownership of making connections outside of our information bubbles, we can start to lessen polarization. The reason I think using Meta, and therefore Facebook and Instagram, can successfully reach people is because they will see my essays in the “privacy” of their social media scroll. What I mean by this is that when people are silently scrolling, they are in their own world. Yes, the algorithm is tracking their actions, but they’re “free” to engage with the content without being watched by their peer group. My hypothesis is that, because of the relief of pressure to publicly demonstrate social conformity, they’re more likely to engage with my ad. Classic conformity experiments demonstrate just how powerfully social context can affect people’s expressed judgments (see Ref. 5). In the privacy of their Facebook scroll, a Christian questioning a fundamentalist interpretation of their faith might be able to read an essay on Biblical literalism without the watchful eye of their family. A person interested in finance might be able to learn that there is, in fact, a business case for the separation of church and state. In this way people are less influenced by social conformity pressures to quickly scroll past one of my ads, especially if I’ve proposed an interesting question, such as “What is, exactly, the business case for separation of church and state?” along with a ridiculous image, such as my dog as Benjamin Franklin. People love pets. Fig. 5 shows a graphic I’m running with an ad on the economics of separation of church and state to the same deliberately selected audience described above. In this campaign, the ad was shown 5,500 times to 2,947 people. It achieved a 4.53% outbound click-through rate, while 6.99% of unique users reached generated an outbound click. Meta also recorded 45 completed registrations from $90.04 in advertising spending, or about $2.00 per registration. These results are particularly interesting because this audience was deliberately selected not for its agreement with the subject, but as a potentially less-interested population I wanted Meta to search within. If we are to lessen the polarization we’re seeing in our public discourse, it will be because we make an effort to communicate with those we disagree with. That means we have to get some skin in the game, and get out of our comfort zones. This is my way of doing so. I’m an engineer, and a pilot. I’m not trained in social media advertising, but I’m trying to learn about it, because I’ve always believed in advancing public science literacy. The motto of my Substack is “If knowledge is power, sharing access to knowledge is sharing power.” Education is the rope we can throw people to help them escape from cynical traps of polarization. If you are interested in learning more, I am willing to schedule a call to describe my efforts. Further, I’m looking for help! If you'd like to help, here are several ways: Help Me Run the Experiment! I don’t yet know whether this strategy can meaningfully bridge online information silos, which is why I’m testing it. Advertising costs money. Every paid subscription allows me to share these essays with people who would otherwise be unlikely to encounter them. If you believe that careful, evidence-based ideas should travel beyond the audiences already inclined to agree with them, please consider becoming a paid subscriber. I’ll keep publishing what I learn, including what doesn’t work. Subscribe or give a gift subscription! Other ways to help include: Share the essays on Frank’s Free Substack on Substack, Facebook, Instagram, X, Bluesky, etc. If you’re a social media advertising expert, I would love to talk sometime to learn more. I’m self-educated to this point, and can use assistance with creating artwork, crafting the ads, and eventually turning my essays into scripts for videos. If you are a writer from another Substack who would like to run ads, I can help walk you through setting them up. I am looking to collaborate with individuals and organizations who might have more resources to expand the strategy and effort. Finally, I am looking for funding. If you know of a grant or other source of funding I should apply for, please let me know! You can get in touch with me with the button below: References: Sasahara, K., Chen, W., Peng, H. et al. Social influence and unfollowing accelerate the emergence of echo chambers. J Comput Soc Sc 4, 381–402 (2021). https://doi.org/10.1007/s42001-020-00084-7 Garimella, K., De Francisci Morales, G., Gionis, A., & Mathioudakis, M. (2018). “Political Discourse on Social Media: Echo Chambers, Gatekeepers, and the Price of Bipartisanship.” Proceedings of the 2018 World Wide Web Conference, 913–922. https://doi.org/10.1145/3178876.3186139 Federal Election Commission reports for 2016, 2020, and 2024 Broockman, D. E., & Kalla, J. L. (2025). “Consuming Cross-Cutting Media Causes Learning and Moderates Attitudes: A Field Experiment with Fox News Viewers.” The Journal of Politics, 87(1), 246–261. https://doi.org/10.1086/730725 Asch, S. E. (1956). “Studies of Independence and Conformity: I. A Minority of One Against a Unanimous Majority.” Psychological Monographs: General and Applied, 70(9), 1–70. https://doi.org/10.1037/h0093718 Superads. “Facebook Ads CTR Benchmarks for Media.” Superside. Accessed August 29, 2026.