Removing Social Media Algorithms: What Experiments Show

Experiments that temporarily replaced algorithmically ranked feeds with chronological or user-selected feeds found mixed, measurable effects—not a simple cure for misinformation or polarization. They show that ranking changes what people see and can shift some outcomes, while effects on beliefs and behavior are often small, inconsistent, or hard to separate from other changes.

What “removing the algorithm” means

Social media feeds are not simply either algorithmic or non-algorithmic. A ranked feed uses automated systems to order posts according to predictions such as relevance, likely engagement, or other platform objectives. A chronological feed generally places eligible posts in time order. Even a chronological feed still depends on choices about who a person follows, which posts are eligible, and what content is removed or downranked under platform rules.

Researchers therefore study specific interventions: switching a feed to chronological order, letting participants choose between feed types, or changing recommendations for a limited period. These tests do not remove every automated decision from a platform. They estimate what happens when a particular ranking system is altered for a particular group, platform, and length of time.

What the major experiments found

One prominent set of studies took place during the 2020 U.S. election. Researchers working with Facebook and Instagram conducted experiments that included a chronological-feed option for some participants. The chronological feeds changed what participants encountered and how they used the platforms. In the Facebook study, people assigned to a chronological feed spent less time on the platform and saw different material, including more content from friends and sources they followed.

However, changing the feed did not produce a clear, broad improvement in several measured political outcomes over the study period. The researchers did not find that switching to chronological order substantially changed political knowledge, polarization, or the other central attitudes they examined. That is not proof that ranking has no influence: it means the measured intervention did not create large detectable changes in those outcomes under those conditions.

Other findings from the same research program illustrate why there is no single “algorithm effect.” In experiments that reduced exposure to content from like-minded sources, participants saw less such content, but the intervention did not produce corresponding large changes in several political attitudes. Separate work on Facebook and Instagram during the election found that deactivating accounts changed political knowledge and some measures of well-being, but those results concern taking a break from the platforms—not merely removing ranking.

Why feed changes can affect exposure more than beliefs

A feed ranking system has a direct influence on the order and visibility of posts. It does not control all the factors that shape a person’s information environment. Users choose whom to follow, what to click, whether to share, and when to leave. Friends and followed accounts also continue to supply content, and news or political discussion takes place outside the platform.

That creates a useful distinction between exposure and downstream effects. A study may reliably show that a feed change reduces the number of posts from a particular source or changes time spent scrolling. It is harder to show that this change alters durable beliefs, voting behavior, or social relationships. People may encounter similar material elsewhere, ignore posts they see, or hold views shaped by long-standing commitments rather than a few weeks of feed exposure.

Researchers also measure different outcomes. “Polarization” might mean stronger dislike of political opponents, greater distance between policy preferences, or more extreme views. A result for one measure does not establish an effect—or lack of effect—on all the others.

What chronological feeds do not guarantee

A chronological feed is a different ordering rule, not a neutral window onto the internet. It may give more visibility to accounts that post frequently, while posts from less active accounts quickly move down the screen. It can make it easier to follow a sequence of updates, but it does not by itself verify claims, balance viewpoints, or ensure that important posts appear.

• Following still matters: A chronological feed only displays content from accounts or sources the user can access under the platform’s rules.
• Virality can remain: Users may still share and react to posts, spreading them beyond their original audience.
• Moderation remains separate: Removing ranking does not automatically remove harmful content or change the platform’s enforcement policies.
• Recommendations may persist elsewhere: Search, suggested accounts, notifications, and video recommendations can still be algorithmically selected.

So “remove the algorithm” is often shorthand for a narrower proposal: change one feed’s ranking while leaving much of the service intact.

How to interpret the evidence

Random assignment is valuable because it helps researchers compare people who were offered different feed experiences. But the findings still have boundaries. The election studies covered a particular country, period, platform design, and group of consenting participants. A temporary switch may not predict what would happen after years of use, after a platform-wide redesign, or in a different political and media environment.

There are also practical limits. Participants may not use the assigned option consistently. A platform’s interface can affect whether a setting is noticed or selected. Researchers may measure what is visible in a feed more directly than what users remember or believe. And results from one platform should not automatically be generalized to every service, since their recommendation systems and user communities differ.

These caveats do not make experiments useless. They help specify what a result supports: for example, “this feed option changed exposure and time spent over this period,” rather than the much broader claim that algorithms are harmless or that chronology fixes social media.

Frequently Asked Questions

Do experiments show that chronological feeds reduce political polarization?
Not consistently. In prominent U.S. election experiments, chronological feeds changed exposure and platform use, but researchers did not find large, clear changes in several measured political attitudes, including key polarization measures.

Does removing recommendations stop misinformation?
No experiment establishes that simply switching to a chronological feed stops misinformation. Posts can still be shared by followed accounts, and ordering changes do not amount to fact-checking or content moderation.

Is a chronological feed truly free of algorithms?
Not necessarily. The displayed order may be chronological, but eligibility, moderation, notifications, search, and other features may still use automated systems or platform rules.

Why do studies find changes in screen time but not beliefs?
Screen time and feed exposure are close to the intervention, so they can change quickly. Beliefs have many influences and may be stable; a short-term change in posts seen may be too limited to shift them detectably.

Can results from Facebook or Instagram apply to every social network?
Only cautiously. Platforms differ in their feeds, recommendation systems, users, and content. A finding on one service is evidence about that intervention in that setting, not a universal result.

What the evidence supports

The strongest conclusion is narrower than either side of the public debate often suggests. Feed ranking can shape exposure and use, and changing the ranking can produce measurable effects. Yet experiments to date do not show that replacing ranked feeds with chronological ones reliably transforms political beliefs or solves misinformation and polarization. The results support evaluating specific design choices and outcomes separately, rather than treating “the algorithm” as one switch with one predictable consequence.

Related

• Monitors for Architects: What to Look For
• Reopen Closed Programs & Folders with Hotkeys
• Record Macro VBA: How to Record & Use Macros