Editorial Note: The three people in this piece are composite characters drawn from real patterns of behavior; specific details have been recombined for narrative purposes. The research discussed in the article is cited in the References below.
What an Information Cocoon Is — and Why the Usual Advice Backfires
Here is something easy to overlook: much of what we read each day is not chosen entirely by us.
The feed remembers what we respond to and keeps supplying more of it. We, for our part, drift away from anything that makes us uncomfortable. Those two forces compound quietly, and after a few years the field of view has narrowed — not because anyone locked the door, but because we stopped walking toward it. Researchers use several overlapping terms for this kind of narrowed information environment, including information cocoons, echo chambers, and filter bubbles. The terms are not interchangeable, but they describe related ways in which information can become less varied or more self-reinforcing.
The advice that circulates most widely is also the simplest: go read the other side. Deliberately seek out views you disagree with. It sounds like a reasonable prescription, and a lot of people follow it faithfully. The trouble is that in practice, deliberately seeking opposing views does not always produce the openness people expect.
Three stories illustrate the different facets of that problem. One has written for a living for a decade and recently noticed she can no longer produce anything surprising. One followed the "read the other side" protocol for five months and grew more certain with each week that the other side was wrong. One consumes more information than almost anyone she knows, and still could not parse what a young viewer was saying about her work. They never met. None of them knew the others existed.
Three Signs You're Stuck in an Information Cocoon
Dana Reyes is thirty-four and has been an editor at a technology publication in Austin for ten years.
That afternoon she received a rejection email. The wording was courteous: interesting angle, but the data doesn't support it, projected retention is low, consider a different framing. She knew every variant of that sentence, and she knew what it meant underneath — this topic will not travel through the feed.
The rejected piece represented four months of work and a judgment she had come to trust: the category of product she covered was fragmenting people's attention into shorter and shorter units, and that cost would not appear in any quarterly report. She believed it. She also half-recognized that the last time she had written something the data did not support, three years had passed.
She copied the draft out of the company system into a private folder. She thought for a while about what to call it, and settled on unpublished.
Marcus Feld is twenty-eight, an independent developer in Portland, maintaining three open-source projects on subscription income.
His desk has two monitors. The right one holds a dozen or more tabs at all times, color-coded by category: blue for views he agrees with, red for views he does not. He set himself a rule — read one piece of uncomfortable content every day before shutting the machine down. Five months in, he considered the experiment a success.
Nora Halloran is forty-one, a documentary director in Chicago, with three films that made the festival circuit.
Her method is depth: three to five years on a subject, dozens of interviews, footage cut down to bone. She had never considered herself constrained by anything — her information intake was larger than most people's. Then her newest film went live, and in the comments she found a two-thousand-word read from someone in their early twenties, using a framework she had never encountered, arriving at a conclusion close to the opposite of the one she had intended.
She read it twice. The first reaction was that the viewer had misunderstood. The second reaction unsettled her more: she could not work out where that framework had come from.
Three people, three forms of being stuck. One cannot produce the unexpected. One hardens with every read. One cannot hear what other people are actually saying. None of them had yet noticed that these were three faces of the same condition.
Why Reading Opposing Views Can Make You More Extreme
Marcus's protocol worked well at first.
The first three weeks felt productive. He read op-eds he disagreed with, listened to podcasts from the other camp, clicked into accounts he would normally swipe past. He framed it as training — cognitive muscle, strengthened by resistance. He even wrote a script to log the share of "red content" in his weekly intake. The number climbed steadily, which he took as evidence of progress.
The shift showed up in week four.
He noticed that after reading a view he disagreed with, his conclusion tended to be firmer than before he had read it. Earlier he would think, that person's premise is wrong. Now he thought, that person's premise is wrong, and they clearly know it, and they're saying it anyway. The labels he attached to the other side grew more specific, and less charitable. One evening he spent forty minutes drafting a rebuttal, then left it in the drafts folder unsent. The impulse itself surprised him.
That night he searched for a description of his own state and found a study.
Chris Bail and colleagues at Duke ran an experiment (Bail et al., 2018, PNAS, 115(37): 9216–9221) in which more than a thousand Twitter users were paid to follow and retweet accounts expressing the opposite political view, for a month. Their positions did not move toward the center. They moved away from it. The effect was statistically clear among conservative participants; among liberals it was small and not statistically significant.
Marcus closed the tab and sat in the dark for a while.
He went back and read the methods section again, and caught a detail: the accounts were bots, and they were used to expose participants to messages from political figures and opinion leaders on the opposing side. He told himself this was not a clean analogy — he was reading long-form essays by actual people. That defense held up about halfway. The other half was that his posture when opening a red tab was not much different from that of the people the experiment had provoked. Every time he loaded a page, the rebuttal was already half-written in his head.
What he was missing had little to do with volume.
Is the Filter Bubble Real? What the Research Actually Shows
Following the thread further, Marcus found that the scholarly picture is considerably more cautious than the popular one.
He started with Emily Dubois and Grant Blank (Dubois & Blank, 2018, Information, Communication & Society), who tested how many people actually satisfy all the conditions an echo chamber requires, using a national UK sample. By their estimate the share is small — on the order of eight percent. Most people's information environments are messier than the discourse assumes; the cracks are there even when people believe they are sealed in.
Then Seth Flaxman, Sharad Goel, and Justin M. Rao (Flaxman, Goel & Rao, 2016, Public Opinion Quarterly, 80(S1)), who analyzed web-browsing records from roughly fifty thousand US-located users. They found evidence that people tend to consume ideologically similar sources, while social media and search also affected the mix of perspectives people encountered. Both effects were modest in overall magnitude.
Read together, those two studies point somewhere uncomfortable: attributing everything to the algorithm lets the personal half of the bargain off the hook. Marcus had opened the red tabs himself. He had written the logging script himself.
Next came a concept Bail developed further in book form (Bail, Breaking the Social Media Prism, Princeton University Press, 2021). He compares social platforms to a prism: they amplify not the real distribution of opinion but its two ends. The predictable result is that most people overestimate how extreme their opponents are — you picture a crowd of fanatics, while the majority standing opposite you are considerably closer than that. Bail calls this false polarization.
Looking back over his five months, Marcus realized that a large fraction of the "opposing views" he had consumed came from the two brightest corners of that prism. He had believed he was surveying the whole landscape. He had been studying its magnified edges.
The study that stopped him was the last one. Sara B. Hobolt, Katharina Lawall, and James Tilley (Hobolt, Lawall & Tilley, 2024, American Political Science Review) ran something closer to lived conditions: they put people into groups to discuss contested topics, some groups homogeneous in stance, others mixed. Afterward, the like-minded groups had grown more divided. The mixed groups had not.
Exposure to differing views produced different outcomes depending on how people participated in the exposure.
Marcus wrote a line in his notebook: what he lacked was not opposing information but a posture capable of holding it. Read defensively, volume only reinforces. Read some other way, and something might give. He could not yet describe that other way. He had, however, moved the question from what to read to how to read.
What an Information Cocoon Actually Takes From You
Around the same time, Dana was auditing her own dead draft.
She did something she had never done: she laid out three years of her published work by headline, then re-sorted it by the question who thought of this angle first. The second sort revealed the pattern. The pieces that had come easiest were the ones whose angles she had not generated. They were angles she had seen repeatedly in the feed, validated by engagement data, and then judged as viable. She had believed she was making editorial calls. Much of the time she was restating an answer the metrics had already produced.
The cost showed up exactly there: she had lost the ability to write anything she was not sure about.
This has less to do with willpower than it appears, and a solid body of cognitive science explains why. Experiments by Henry Roediger and Jeffrey Karpicke (Roediger & Karpicke, 2006, Psychological Science; Karpicke & Roediger, 2008, Science) found that active retrieval can strengthen long-term retention more than repeated study alone. Restating material in your own words or explaining it from memory requires a similar shift from passive exposure to active retrieval. Passive intake can raise familiarity, and familiarity is not the same as durable understanding.
Dana's working life was the inverse of that mechanism. She read enormous volumes, then produced in whichever form had already been validated as likely to win, and rarely wrote anything she had not resolved. Information flowed in; it was almost never forced out. After three years, her intake was probably as broad as ever, but the part of her output that felt genuinely hers had begun to shrink.
A second form of attrition was subtler. Sophie Leroy's concept of attention residue (Leroy, 2009, Organizational Behavior and Human Decision Processes) describes what happens when you switch tasks: residue from the previous task travels with you and occupies bandwidth the next task needs. Feeds can encourage frequent switching between pieces of content. Dana noticed that reading an eighty-page industry report, she would reach page twenty and feel a pull toward something else to skim — not fatigue, just too much residue and not enough bandwidth.
That evening she started a document no platform would ever see. No headline, no word count, no retention projection. The only rule: it could be bad, its conclusions could be overturned, it could stop after three lines.
The first week produced four entries, mostly nonsense. In the second week, one of them contained a sentence she had not anticipated. She looked at it for a long time, then kept writing.
She left the rejected draft on a personal site with almost no traffic — plain title, no promotion, no social card. She uploaded it without seriously expecting anyone to find it.
The Missing Half: From Output to Reception
Nora did the least intuitive thing of the three: she went to find the student who had misread her film.
Her intention had been to clarify. She brought material, prepared to explain why the edit was structured that way, why a particular shot sat where it did. Once they were sitting together, she found she barely got a turn.
The student talked for two hours. About where he grew up. About how old he was when he first saw imagery like hers. About why he had read a supporting character as the film's center of gravity. About what he had gone on to do because of what he took from it. Nora interrupted herself four times, internally, and did not do it out loud.
She was not persuaded. She still thought her edit was right, and the film's intention had not moved. But at some point she registered something more basic: outside her framework there were other frameworks, and those were not errors — just different entrances.
On the way home she located her own problem. Her cocoon sat on the output side.
Her workflow contained a stark asymmetry. Input was rigorous: dozens of hours of footage, dozens of interviews, repeated cross-checking. Output was entirely one-directional — cut, submit, release, done. She had never treated "how the audience understands this" as a stage worth researching. She gathered an enormous quantity of information and collected no feedback at all. High bandwidth in, near zero in the other direction.
That was the same coin as Marcus's problem, flipped: one could only read without loosening; the other could only speak without hearing.
Back at the studio she changed the process. The research phase of every new project now includes a fixed step: a rough cut goes to ten people spread as widely as possible across age, gender, and professional background, forty minutes each, with one rule — no explaining, only listening. She wrote it into the project documentation under a plain title: listen before cut.
Assembling research material, she needed a tool that could aggregate scattered sources. Commercial products did not fit. In an open-source community she found a personally maintained project whose author had documented the design clearly: no recommendations, just subscription feeds flattened onto a timeline, with fetch rules anyone could modify. She spent two weeks adapting it to her needs, left a note of thanks in the documentation, and attached her changes.
She never learned who the author was. She only remembered a name in the code comments: Marcus.
How the Loop Closed Without Anyone Meeting
Marcus found Dana's dead draft on an ordinary late night.
He had been searching for a marginal detail on a technical problem and landed on a personal site with almost no traffic. The article described how recommendation systems substitute data-driven judgment for a creator's own, and one passage made him read it twice: the author admitted that most of what she had written over ten years were angles the algorithm had already validated as winners, and admitted that discovering this had been unpleasant.
The piece did not argue with him. It did not address his positions at all. It did something of a different order: it showed him a peer's failure process — how she had noticed, how it had felt, and what she had done next.
That night Marcus did not draft a rebuttal. He opened his own logging script, deleted the "red content share" metric, and replaced it with a clumsier rule: after reading something he disagreed with, restate the other side's argument using the other side's premises, until the other side might nod — only then is judgment permitted. He gave himself three weeks. The first week his restatements were ugly. In the second week he produced one version he thought the other person would recognize as fair, and he paused at that moment. It was the first time in five months.
Nora's new film contained an interview segment on how creators are shaped by recommendation systems. She had talked to seven people; one was a former editor at a technology publication, careful with language, almost adjective-free. After the cut was finished, Dana saw that segment at a post-screening talk in an independent theater.
She sat in the last row and heard the man say: "I'm no longer sure which of my judgments are mine."
She did not go up afterward. She walked out, stood in the parking lot for a few minutes, got back in the car, and opened the folder called unpublished. There were twenty-something entries in it now. She picked the shortest, changed two headlines, and posted it to her own site. The view count stayed in the double digits.
None of the three ever knew the others existed. Dana's dead draft prompted one of Marcus's turns; Marcus's tool supported Nora's research; Nora's film let Dana confirm she was not the only one. The loop closed through work, not coincidence.
During that stretch, each of them also made a small change that rhymed with the others'. Dana turned off auto-refresh in the publishing dashboard and switched to checking metrics manually twice a day. Marcus disabled autoplay in his browser, so "next video" became an action requiring a click. Nora turned off real-time preview in her editing software, watching only after completing a segment.
Three people, independently, switched off something that had been on by default.
What Actually Changed
Dana assembled a source list of eleven outlets, deliberately spread across the spectrum, three of them sources she explicitly disagrees with. She reads through it weekly and writes three lines of notes afterward, notes nobody else sees. Her judgment did not improve overnight. What returned was the ability to write a sentence containing "I'm not sure" — which mattered more to her.
Marcus replaced "read opposing views" with "read process." He started reading other people's failure logs, retrospectives, rejected proposals, abandoned drafts, rather than position statements. He kept the opposing content, but behind the restatement gate. His positions did not soften. What changed was the speed at which he reached them — slower, and somewhat more accurate.
Nora wrote reception into her workflow, so every project now has a mandatory listening stage. She still defends her editorial instincts, but she can now say, before cutting, which entrances an audience is likely to come through. Her films did not become more accommodating. They became more legible.
The common thread across all three is narrow: none of them treated "seeing more" as the goal anymore.
The signal that your thinking has widened does not necessarily feel like certainty. The moment something actually gives is quieter — you read a sentence you would once have dismissed, and you cannot immediately rebut it, and you find yourself a little curious about why the person believes it.
That feeling is curiosity. It is not proof that you have changed your mind, but it can be a useful sign that you are no longer reacting from inside the same frame.
Continue Exploring
References
-
Christopher A. Bail, Lisa P. Argyle, Taylor W. Brown, et al. (2018) Exposure to opposing views on social media can increase political polarization https://doi.org/10.1073/pnas.1804840115
-
Seth Flaxman, Sharad Goel, and Justin M. Rao (2016) Filter Bubbles, Echo Chambers, and Online News Consumption https://doi.org/10.1093/poq/nfw006
-
Elizabeth Dubois and Grant Blank (2018) The echo chamber is overstated: the moderating effect of political interest and diverse media https://doi.org/10.1080/1369118X.2018.1428656
-
Jeffrey D. Karpicke and Henry L. Roediger III (2008) The Critical Importance of Retrieval for Learning https://doi.org/10.1126/science.1152408
-
Henry L. Roediger III and Jeffrey D. Karpicke (2006) Test-Enhanced Learning: Taking Memory Tests Improves Long-Term Retention https://doi.org/10.1111/j.1467-9280.2006.01693.x
-
Chris Bail (2021) Breaking the Social Media Prism: How to Make Our Platforms Less Polarizing https://doi.org/10.1515/9780691216508
-
Sara B. Hobolt, Katharina Lawall, and James Tilley (2024) The Polarizing Effect of Partisan Echo Chambers https://doi.org/10.1017/S0003055423001211
-
Sophie Leroy (2009) Why is it so hard to do my work? The challenge of attention residue when switching between work tasks https://doi.org/10.1016/j.obhdp.2009.04.002