Nobody sees the internet. Each person sees a thin slice of it, chosen moment by moment by ranking systems that sort billions of posts and pages. Those systems are neither neutral pipes nor evil puppeteers. They are tools built to pursue goals, and the goals shape which information wins. Understanding them is part of information hygiene, because you cannot judge your diet without knowing something about who is cooking. (How feeds are designed to hold attention is a subject of its own; this level asks what ranking does to information.)

Ranking by engagement

Most large platforms rank content partly by predicted engagement: how likely you are to click, watch, like, comment or share. Engagement is easy to measure and roughly tracks what people want, so it is a reasonable place to start. The trouble is that it rewards reactions, and anger and fear provoke them reliably. A calm, accurate post and an angry, misleading one may compete for the same slot, and the angry one often wins.

A well-documented case came from Facebook. In January 2018 the company changed its News Feed to favor what it called "meaningful social interactions," giving extra weight to posts that drew comments and shares. The stated aim was to help people connect with friends and family. Internal documents later released by the former employee Frances Haugen, and reported in 2021, showed that company researchers had found the change rewarded outrage and divisive content, and that political parties in Europe told the company they had shifted toward more negative messages to keep their reach. Facebook disputed parts of this account and made later adjustments. The case shows how a reasonable-sounding metric can reshape public speech.

Defenders reply that people choose what to click, and that chronological feeds often show more spam and less of what people value. The honest summary is that ranking strongly affects what spreads, while its effect on what people end up believing is still debated.

Data voids

Search engines face a different problem. In 2018 the researchers Michael Golebiewski and danah boyd described data voids: search terms with little reliable content behind them, so that whatever does exist rises to the top. Voids open up during breaking news, before journalists have written anything, and around obscure or newly coined phrases. Manipulators exploit them because they can fill an empty search with their own content: coin a phrase, publish pages using it, then urge people to "search it yourself." The searcher feels independent, but the results were planted. Hygiene here means treating the first hours of a story, and any unusual phrase you are urged to look up, with extra care.

Context collapse

Feeds also strip posts from their settings. The researchers Alice Marwick and danah boyd used the term context collapse for social media merging audiences that are usually separate, such as friends, coworkers, family and strangers. A joke meant for friends reaches strangers who lack the background, and a clip from a long speech circulates without what came before it. Much misleading content is not fabricated at all. It is real material shown out of context: an old photo labeled as new, a quotation with its qualifying sentence cut off.

Changing the design

Design choices can also slow bad information. In 2018, after rumors spread on WhatsApp were linked to a series of mob killings in India, the company limited how many chats a message could be forwarded to at once, first in India and then worldwide. In April 2020, early in the pandemic, it went further: messages that had already been forwarded many times could be sent on to only one chat at a time. The company reported a 70 percent drop in such highly forwarded messages. The change did not judge any message true or false. It simply added friction.

Another approach is crowdsourced context. Twitter launched a program called Birdwatch in 2021, later renamed Community Notes. Volunteers write notes adding context to posts, and a note is shown publicly only when people who usually disagree rate it helpful. This "bridging" rule aims to reward notes that persuade across divides rather than notes that please one side. In 2025 Meta announced that it would replace its US fact-checking partnerships with a similar system. Critics note that notes often arrive after a post has already spread, and that most proposed notes are never shown.

Worked example

Minutes after an explosion is reported in a city, you search the city's name plus "explosion." The top results are three videos from accounts you have never heard of, one claiming that a named group is responsible. This is a data void: reliable outlets have not yet published, so whatever exists ranks high. One of the videos turns out to be real footage from a different city two years ago, a case of collapsed context. You also notice that the top post in your feed on the subject is the most furious one, not the most informative, which is what engagement ranking predicts. The hygienic choice is to wait an hour, check whether established outlets or local authorities have confirmed anything, and avoid sharing early videos, however dramatic. Doing nothing for an hour is sometimes the most responsible thing a reader can do.