News Filter Bubble: How to Get Out of One

A news filter bubble is what a feed becomes when a ranking system keeps showing what you already clicked. You get out by changing the input, by seeing why each story is there, and by resetting the ranking — not by reading harder. Pew Research Center says 53% of US adults at least sometimes get news from social media.

That figure is from Pew's social media and news fact sheet, published September 25, 2025 and built on a survey of US adults run August 18–24, 2025. A social feed is the surface where you decide the least about what comes in, which is why it is the usual example. This guide covers what a bubble is, why the usual advice fails, the three levers that work, and how they apply to a news app that is itself algorithmic — including ours.

Key takeaways

A filter bubble is built by a ranking system reacting to your clicks. An echo chamber is built by the people and sources you chose to follow. The fix is different for each.
Reading "the other side" harder inside the same feed does not help: every extra tap is another training signal for the same ranking.
Three levers get you out: pick a fixed set of newsrooms yourself, see the reason each story is in front of you, and reset the ranking when it drifts.
GetMyNews ranks with an on-device model learned from your swipes, so it can drift too. A card names the weight that put it there, a screen lists what the model learned, and Settings has a "Reset learning" button.

Filter bubble vs echo chamber: what is the difference?

A filter bubble is made by software; an echo chamber is made by choices. That is the two-line version, and the research literature agrees with it.

The Reuters Institute's literature review on echo chambers, filter bubbles and polarisation (Ross Arguedas, Robertson, Fletcher and Nielsen, January 2022) quotes Eli Pariser's original definition of the filter bubble: "a unique universe of information for each of us," produced by personalization that shows "more and more of things we like, while things we are not prone to like are hidden from us." The same review defines an echo chamber as "a bounded, enclosed media space that has the potential to both magnify the messages delivered within it and insulate them from rebuttal." Ohio Wesleyan University's news literacy guide gives the plain-English test for the first: a filter bubble is "when a website's algorithm selectively assumes the information that users want to see based on past actions."

The distinction matters because the exit is different. You leave an echo chamber by following different people. You leave a news filter bubble by changing what the ranking is allowed to learn from, and by being able to see what it learned.

The same review should lower the temperature: in the United Kingdom it found around 2% of people in left-leaning echo chambers and around 5% in right-leaning ones, and that "most people have relatively diverse media diets." The feeling that your feed shows one side is real; the mechanism is more often what you chose to follow than what a machine hid, which is why the first lever below is about input.

Why reading harder does not get you out

Every ranked feed runs the same loop: stories come in, each gets a score, the winners are shown, your reaction updates the scores. How news apps choose what you see walks through the first three steps; the part that matters here is the fourth one, the feedback.

Inside a feed you did not configure, every tap, pause and share is a training signal. Opening a story you disagree with does not teach the system that you want balance. It teaches the system that you engage with that kind of story — and if the story was chosen to provoke you, that means more provocation. You also cannot audit the ranking from inside it: presentation hides the score.

The third problem is comfort. The Reuters Institute's 2026 Digital News Report page on the United States (Lucas Graves and Joy Jenkins, June 16, 2026) puts trust in news at 25%, the lowest US figure since it began tracking trust in 2015, with 45% sometimes or often avoiding the news. When trust is that low, a feed that flatters is easier to keep open than one that widens, and a ranking optimized for time in session will find that comfort.

The three levers: input, visibility, reset

1. Pick the sources yourself, as a fixed list

Ohio Wesleyan's guide says to "search out news sources that do not match your own political leanings." Right, and vague. Make it concrete: choose pairs of desks that cover the same beat from different rooms, and write the list down. A list you cannot recite from memory is a stream, not a list.

Pairs that work:

National politics: NPR and Fox News. A public radio desk and a cable network rarely lead with the same story.
Washington: Politico and The Hill. Two Capitol desks, two ideas of the day's development.
New York: The New York Times and the New York Post. Same city, different front pages.
Markets: CNBC and MarketWatch, with Bloomberg or The Wall Street Journal added if you pay for one.

The point of a pair is that a disagreement between two desks is visible. Once the list exists, the input to any ranking is bounded: it can reorder those newsrooms, but it cannot swap them for whatever a platform found engaging this hour. Why social media is a bad news source is the longer argument for keeping the input out of a platform's hands.

2. See why a story is in front of you

If a surface cannot tell you why an item is there, you cannot correct it. Three things to look for:

A reason per item. "Because you follow X" — anything specific. A generic "For you" label is presentation, not explanation.
A page that lists what was inferred. Topics, sources, words. No such page means the model is a black box by design.
A way to delete one inference without deleting everything. That is the difference between steering and starting over.

Where the model lives decides whether any of this is possible: a model on a company's server shows you what the company chooses to show, while a model that never leaves your phone can be listed on screen in full. On-device vs cloud personalization sets out that difference.

3. Reset when it drifts

A reset is maintenance, not failure. The signs of drift are consistent: the same three topics every morning, headlines whose shape you recognize before you read them, a feeling that the feed "knows" you. When that happens, reset the ranking and keep the sources. The list is the part you chose; the weights are the part that accumulated.

GetMyNews is algorithmic too — here is what you can see and undo

This site belongs to a news app, and the app ranks stories with a model. Leaving that out of a guide about filter bubbles would be dishonest, so here is how the three levers apply to it, read from the app's code.

The ranking. The model is a set of weights stored on the phone — per category, theme, newsroom, country and headline word — learned from your swipes. A swipe right adds weight to the story's keys; a swipe left takes some away, and counts for less, because a left swipe is ambiguous where a right swipe is a choice. The next cards are scored from those weights plus freshness, with two counterweights: an exploration term that never falls to zero, and a re-ranking pass that penalizes repeats of the same category, theme, newsroom or country inside a batch. That is still a ranking system reacting to what you tapped, and it can drift.

The input. 92 public RSS feeds from 37 US newsrooms, fixed in the app's registry. The model can reorder those newsrooms; it cannot reach outside that list, and no amount of swiping adds a source to it.

Seeing why. When the model has a reason for a card, the card says so in one line: "Shown because you follow Science," "Shown because you liked Politics stories," or "Shown because you asked to see Economy first." After ten swipes, a card with no positive signal reads "Outside your usual — keeping your feed varied." On the Discover tab, a strip under the header shows "12 swipes learned ›"; tapping it opens "What I learned about you," which lists the categories, themes, newsrooms and headline words the model weights, plus your rules. Rules set in the Tune feed tab — hide a topic, see one first — appear there and on the strip as "Hidden: Politics" or "First: Science," and a tap removes the rule and rebuilds the deck.

Resetting. In Settings, under Algorithm, the button is "Reset learning" — "Restarts from your onboarding preferences, without touching your history." The same button sits at the bottom of the "What I learned about you" screen. The newsroom list is untouched by a reset, which is the whole point of lever one.

GetMyNews is free on Google Play for Android, paid for by a banner and one full-screen ad about every 20 stories.

Keep the list, watch the reason, reset without guilt

Getting out of a news filter bubble is three settings, not a reading assignment: a source list you can recite, a surface that tells you why each story is there, and a reset you are willing to use before the feed starts to feel like the world. Do the first one today.

FAQ

Is a filter bubble the same as an echo chamber?

No. A filter bubble is produced by a ranking system inferring what you want from your past clicks; an echo chamber is a media space you assembled by choosing who and what to follow. The Reuters Institute's 2022 literature review put the UK at around 2% of people in a left-leaning echo chamber and around 5% in a right-leaning one, and found that relying on search engines and social media is "in most cases associated with more diverse news use." Change what you follow for one; change what the ranking can learn and show for the other.

Does incognito mode or a different search engine fix a news filter bubble?

Only for search. The Ohio Wesleyan guide suggests searching incognito to "remove your personal belief from the algorithms," which works for checking how a topic looks without your profile attached. It does nothing to a feed you are signed into on a social platform or in a news app, because that ranking is driven by your account history and reactions, not by browser cookies.

Can a news app that learns from swipes still be a filter bubble?

Yes. Any model trained on your reactions will tend to show you more of what you accepted, wherever it runs. What changes with an on-device, inspectable model is your leverage: the source list is fixed, each card can name the learned reason it is there, rules can be removed with a tap, and one button resets the weights without touching the list.