Much of what people see online is chosen not by human editors but by algorithms: automated systems that decide which posts, videos and articles appear in a feed and in what order.

How feeds are ranked

Platforms usually earn money from advertising, so they benefit when people spend more time on them. Their ranking systems predict what each user is likely to engage with, measured by signals such as clicks, likes, shares, comments and time spent watching. Content that is predicted to keep you engaged is shown higher.

This has side effects. Content that provokes strong emotions, especially outrage, surprise or fear, tends to get more engagement, so it can be rewarded with wider reach. Creators who learn what the algorithm rewards may shape their content to match. None of this requires anyone to intend harm: an engagement-maximizing system can spread sensational material simply because people react to it.

Filter bubbles and echo chambers

Two ideas dominate debate about these systems:

• A filter bubble, a term popularized by the writer Eli Pariser in 2011, is a personalized information environment created by algorithms. Because systems learn what you like, they may show you more of the same and less of what challenges you, often without your noticing.
• An echo chamber is a community where like-minded people mostly hear their own views repeated and amplified. Echo chambers can form through people's own choices about whom to follow, which groups to join and which outlets to trust, with or without algorithms.

The difference matters: filter bubbles are mainly about what systems choose for us; echo chambers are mainly about what we choose for ourselves.

What the evidence shows

These ideas are intuitive, but the research is mixed. Some key findings:

• Many studies find that most people's news diets are more varied than the filter bubble idea suggests. Tightly sealed echo chambers seem concentrated among a relatively small group of highly engaged, strongly partisan users.
• People's own choices often matter as much as, or more than, algorithms in narrowing what they see.
• Large experiments run with researchers on major social media platforms around the 2020 US election found that changing people's feeds for about three months changed what they saw, but did not measurably change their political attitudes over that period.
• One experiment that paid people to follow an automated account reposting messages from the opposing political side found that some participants became more entrenched, not less. Simply exposing people to the other side does not guarantee open minds.

At the same time, researchers continue to document cases where recommendation systems promote sensational or misleading content, and effects may differ across platforms, countries and groups. A fair summary is that algorithms matter, but they are not the only cause of polarization, and the size of their effect is still debated.

What you can do

• Diversify deliberately: follow some reliable sources that see the world differently from you.
• Notice your reactions: if a post makes you angry or thrilled, that is when to slow down.
• Use the controls you have: many platforms let you switch to a chronological feed, mute topics or see why a post was recommended.
• Go to sources directly: visit news sites or newsletters rather than relying only on what a feed delivers.
• Remember the feed is not the world: what is popular online is not a reliable measure of what most people think.

Worked example: two feeds

Imagine two invented users who both start using the same video platform.

Alex watches a few videos about home cooking, then one angry rant about a local political controversy. He watches the rant to the end and comments on it. Over the following days, his recommendations shift. He still sees cooking videos, but more and more political content appears, increasingly heated in tone. He has not chosen to follow any political channels. The algorithm has simply learned that this content keeps him watching.

Jordan follows only channels from one political viewpoint, joins a group where members mock the other side, and blocks anyone who disagrees. Her feed is narrow, but mostly because of choices she made herself.

Alex's experience resembles a filter bubble drifting toward sensational content, driven by engagement signals. Jordan's resembles an echo chamber, largely self-built.

What does research suggest about how much this changes their views? Probably less than many people assume, at least in the short run. Large experiments found that changing feeds did not quickly change attitudes. But Alex is spending more time angry, and Jordan is seeing a distorted picture of the other side, which may make her think opponents are more extreme than they are.

What could each do? Alex could check why videos are being recommended, clear his watch history or use a subscription-only feed. Jordan could follow a few thoughtful voices from the other side, not the most outrageous ones, and remember that exposure alone does not guarantee understanding. Both could remind themselves that their feeds are not a mirror of public opinion.