Curated news vs algorithmic feeds

Curated news vs algorithmic feeds

Curated news vs algorithmic feeds is the question every reader answers without realizing it. A curated news service has a human editor picking what runs, ordering it by significance, and publishing a finished product to a fixed list of subscribers. An algorithmic feed has a recommendation system picking what each user sees, ordering it by predicted engagement, and serving a different product to every viewer. The two formats answer different questions. Curated news answers "what should a citizen know today?" An algorithmic feed answers "what will keep this person scrolling longest?" Those are not the same question, and the choice between them shapes what reaches a reader every day.

This post compares the two formats: how each one works under the hood, what each is built to optimize for, what the research says about outcomes, and how to tell which format is serving you.

What is the difference between curated news and an algorithmic feed?

The difference is the selector, the standard, and the audience.

• The selector. Curated news has a person. An algorithmic feed has a model. The person can name a reason for picking a story. The model cannot, except in the post-hoc sense that the story scored highly on a numeric engagement target.

• The standard. Curated news judges stories against a written editorial standard: significance, magnitude, lives affected, public-interest weight. An algorithmic feed judges stories against an engagement target: clicks, dwell time, replies, shares, watch completion. Whether the story informs anyone is a downstream metric the system does not directly score.

• The audience. A curated news brief is the same for every subscriber. A reader in Phoenix and a reader in Boston get the same five stories in the same order. An algorithmic feed is different for every viewer. Two people opening the same app at the same moment see different content because the model is making a per-user prediction.

Those three differences cascade into very different reading experiences. A curated brief ends. An algorithmic feed does not. A curated brief has an editor accountable for the picks. An algorithmic feed has a model that even its own engineers cannot fully introspect. A curated brief tells a reader what was important. An algorithmic feed tells a reader what they would click.

What does an algorithmic feed actually optimize for?

An algorithmic feed is a closed-loop optimization system. It observes micro-actions (what gets clicked, how long the scroll pauses, what gets skipped, what gets scrolled back to), builds a predictive model of behavior, ranks content to maximize a target variable, serves the ranked content, and retrains on the response.

The target variable across major platforms is some version of engagement. The major platforms have published architectural details that confirm this: a candidate-generation step that pulls from a large pool, a ranking step that scores each candidate against a neural network predicting likely actions (likes, replies, shares, dwell time, watch completion), and a heuristics layer that applies safety and inventory rules before serving. The optimization target is the predicted action, not the value of the content.

That choice has downstream effects. Two are worth naming.

1. Engagement and accuracy diverge. A University of Copenhagen study published in May 2026 compared engagement-based ranking against several alternative algorithms (reverse chronological, balanced insertion, diversity-weighted). The engagement-ranked feed produced more polarized and less accurate beliefs than every alternative tested, while at the same time receiving higher "insightful" ratings from the users consuming it. People felt informed and were less so.

2. Stated preferences and revealed preferences diverge. A 2026 CSCW study of young adult social media users found a systematic gap between what users say they want from a feed (accuracy, diversity, high-quality information) and what their behavior actually engages with. When the same users were asked to curate an ideal feed by hand, they produced feeds they themselves rated as higher quality and more satisfying than the engagement-ranked one they had been using.

The pattern is consistent across the empirical literature. The engagement target is not a proxy for reader benefit. Optimizing for it produces a feed that performs well on revenue metrics, gets rated highly on subjective satisfaction, and underperforms on the outcomes (accurate beliefs, awareness of significant events, time well spent) that readers themselves say they want.

What does a curated news service optimize for?

A curated news service optimizes for significance. The output is a small set of stories the editor judges important enough to take up a reader's attention today.

The selection bar is explicit and external. Editorial standards usually name a few criteria: the size of the change a story represents, the number of lives affected, the public-interest weight of the underlying decision, the depth of supporting evidence. Stories that are popular but not significant (a celebrity feud, an inflammatory post, a viral video) are not by themselves grounds for inclusion. Stories that are significant but unpopular (a regulatory filing, a foreign-policy development, a quietly important scientific finding) make the cut.

The output has structure most algorithmic feeds do not. A curated brief is finite. It ends. It is bounded by a real editorial commitment to a reader's time. There is no infinite scroll because the editor decided what was worth a reader's morning and stopped there.

The discipline costs something. A curated service has to pay editors, build editorial process, and accept the friction of human judgment with all of its bias and inconsistency. An algorithmic feed scales to billions of users on no editorial labor at all. The trade is that one of these formats is built around a question a reader cares about (what do I need to know?) and the other is built around a question the platform cares about (what will keep this person here longest?).

What does the research say about each?

The empirical literature on algorithmic news recommendation has matured. The headline findings.

• Filter bubble effects exist, especially for moderates. Two experimental studies published in 2025 in Germany and the United States built running news recommender systems biased toward each participant's political preferences. The result: ideologically biased recommendation increased polarization among politically moderate users in both countries. The effect was real, bounded, and measurable.

• Recommendation systems create information cocoons. A 2025 multidimensional assessment of seven news recommendation algorithms found that personalization deepening (more user history, tighter behavioral matching) drove higher topic homogenization, lower category diversity, and stronger sentiment polarization. The cocoon effect is structural, not a glitch.

• "News will find me" eroded source evaluation. A May 2026 Penn State study found that readers who rely on algorithms and social networks for news (about one in three Americans, by survey) judged algorithmic recommendations and friend shares as just as credible as professionally edited stories. The mid- and low-NFM groups still distinguished between source types. The high-NFM group did not.

• Alternative ranking algorithms can produce better outcomes. The same University of Copenhagen study tested simple alternative ranking algorithms against engagement ranking. Every alternative produced more accurate and less polarized beliefs than the engagement baseline. The implication is that the outcomes of algorithmic feeds are a function of design choices, not an inevitable property of automation itself.

The literature does not say algorithms are bad. It says engagement-maximizing algorithms produce outcomes that diverge from what readers say they want, and that there are design alternatives, but the major platforms have not deployed them at scale.

When is an algorithmic feed actually useful?

The honest answer.

Algorithmic feeds are useful for discovery in domains where the universe of content is too large for any editor to filter and the goal is exposure to surprise. Music recommendation, niche-interest video, hobby-specific communities, long-tail product discovery: all of these benefit from a model that sees patterns across millions of users and surfaces content a human curator would never get to.

Algorithmic feeds are also useful when the cost of a wrong recommendation is low. Picking a song a listener skips after thirty seconds is cheap. Reading a polarizing story that subtly miscalibrates a reader's view of an election is not.

Where algorithmic feeds fail is in domains where the reader is trying to maintain an accurate model of the world. News is the canonical example. The optimization target (engagement) is structurally misaligned with the reader's actual goal (a sound picture of what happened today). The mismatch is not a bug in any one platform. It is a property of the design pattern.

How do you tell which one is serving you?

A short test for any news source in current rotation.

1. Can the source name its bar? Ask: "What standard does this source use to pick what it shows me?" If the answer is "the algorithm decides based on what I have engaged with," it is a feed. If the answer is a written editorial standard a reader can read, it is curated.

2. Does the source end? Curated products are finite. They have a length. They stop. Feeds do not.

3. Does everyone see the same thing? Curated products are the same for every subscriber. Feeds are different per viewer. If a friend can read the same edition and have a conversation about the same five stories, it is curated.

4. Is the optimization target visible? Curated products optimize for something the source will name openly (significance, accuracy, depth). Feeds optimize for engagement, and the source will rarely say so directly.

5. How does it feel after thirty days? A curated source should leave a reader better informed and not worse-feeling. A feed often does the opposite. The thirty-day check is the cleanest one.

The informed.now angle

informed.now is a curated SMS news service. An editor selects what runs each day. The brief is the same for every subscriber. It ends. There is no algorithm choosing per user. The format is built for the question a citizen actually asks (what do I need to know today?) rather than the question a platform asks (what will keep this person here?). The longer description of the editorial standard lives on the site.

Try one week of curated, one week of algorithmic, then compare

The honest evaluation is empirical. Subscribe to a curated daily news service for seven days. The same week, track how often the algorithmic feeds get opened for news. At the end of the week, write down three things: what stories from each format are memorable, how informed a reader feels about events they did not personally select, and how much time the two formats took.

For most readers in 2026, the curated week wins on all three. The feed week loses on memory because feeds are designed to be forgotten, loses on informed-feeling because feeds optimize for engagement not coverage, and loses on time because feeds are designed to not end. The test is repeatable, cheap, and ends cleanly either way.

The real question is not which format is better in the abstract. It is which format is serving the reader you want to be.