How Social Media Algorithms Work

In 2009, Facebook made a quiet change. Until then, its News Feed showed posts in reverse chronological order, the most recent thing your friends posted was at the top. The change replaced that with a ranked list, reordered by a score Facebook calculated in real time. The company called the system EdgeRank. Most users didn’t notice. Over the following decade, every major platform followed the same logic: Instagram in 2016, Twitter/X in 2016, LinkedIn long before most people realized it. By 2025, there is no major social platform that shows you content in the order it was created. For a broader look at how these recommendation algorithms work across different platforms, see our explainer.

The short answer

Social media algorithms are ranking systems that score every piece of content and decide what order to show it to each user. They’re trained to predict which content will generate the most engagement from you specifically, likes, comments, shares, saves, and time spent, because engagement correlates with time on platform, which drives advertising revenue. The result is a feed that feels personalized but is optimized for the platform’s financial interests, not yours.

The full picture

Signals: the raw material

Every interaction you have with a platform generates data that feeds its algorithm. When you like a post, that’s a signal. When you scroll past something without stopping, that’s a signal too. When you watch a video for 8 seconds and then keep scrolling, that’s different from watching the whole thing and rewatching the last 10 seconds.

Meta’s internal research, revealed in documents published by The Wall Street Journal in 2021, showed that Facebook’s algorithm classified user interactions into more than 200 distinct signals. These include obvious ones (likes, comments) and subtle ones: whether you hovered over a profile photo, whether you clicked a link and then came back immediately, whether you watch a Live video with the sound on.

The signals aren’t equal. TikTok’s algorithm, described in a leaked internal document first reported by The New York Times in 2021, weights completion rate, the percentage of a video you watch, as its most important signal by far. A video that 80% of viewers watch to the end will be boosted significantly over a video that 80% of viewers abandon at the halfway point. This single insight explains most of TikTok’s product design: short videos make high completion rates easier to achieve.

Collaborative filtering: the mechanism behind “for you”

The most powerful component of modern recommendation algorithms isn’t what you did, it’s what people like you did. This technique is called collaborative filtering, and it’s the engine behind TikTok’s For You page, Spotify’s Discover Weekly, and YouTube’s recommendation sidebar.

The logic: if you and 50,000 other users have all liked the same 30 videos, and those 50,000 users also liked a 31st video that you haven’t seen yet, the algorithm infers you’ll probably like it too. You’ve never been shown that video, but your “neighbors” in the user graph have, and their behavior predicts yours.

This is why the For You page works for new users. Within the first 3–5 videos you watch on TikTok, the algorithm has placed you into several clusters of similar users. It doesn’t need your long-term history. It needs enough signal to find your cluster, and then it can borrow recommendations from everyone else in it. TikTok’s engineering team has said that a new user’s feed can become “meaningfully personalized” in as few as 6 interactions.

The engagement trap

The problem with optimizing for engagement is that some content gets disproportionately high engagement not because it’s valuable, but because it’s emotionally activating. Anger, outrage, and anxiety are reliable engagement generators. So is content that confirms existing beliefs.

A 2021 internal study at Facebook, described in documents obtained by whistleblower Frances Haugen, found that posts in the “angry” reaction category were given 5x the algorithmic weight of regular “like” reactions. The underlying models use the same large language model-adjacent techniques to classify content sentiment and predict engagement. The company removed this weighting in 2019 after internal researchers found it was amplifying misinformation, but the broader dynamic remains: emotionally activating content generates more engagement than calm content, and engagement is still the primary optimization target.

Researchers at MIT’s Media Lab, including Sinan Aral, have published studies showing that false news stories spread on Twitter/X at roughly 6 times the speed of true stories, and that the primary driver is novelty, not bots. Surprising or emotionally charged false information generates more replies and retweets than accurate but unsurprising true information. Their landmark study was published in Science journal.

How Instagram and Facebook differ from TikTok

TikTok’s algorithm starts from cold: it assumes nothing about what you want and uses completion rate as its primary signal. Instagram and Facebook start from the opposite end, they show content from people you already follow first, then layer in recommendations. The balance has shifted. In 2022, Instagram CEO Adam Mosseri announced that the platform would show more content from unfollowed accounts, a shift toward the TikTok model that was partially reversed after user backlash.

LinkedIn adds a professional context layer: content gets higher scores when the people engaging with it are senior professionals in your field. LinkedIn’s engineering team has described this as “quality voter” weighting.

The creator perspective: what the algorithm rewards

For people publishing content on these platforms, the practical implications are significant. Instagram’s internal documentation, published in 2021, listed the specific signals that affect Reels distribution: video file quality, audio quality, aspect ratio (9:16 preferred), use of original audio, and whether the video was originally created for another platform (with a TikTok watermark, for instance, explicitly down-ranked).

Average engagement rate across Instagram in 2023 was approximately 0.54% for business accounts, according to Rival IQ’s annual benchmarking report meaning that for every 1,000 followers, fewer than 6 people will interact with a typical post. This low baseline makes algorithmic reach amplification the primary growth lever for most creators. The alternative, reaching your followers organically, is no longer the dominant distribution model on any major platform.

Why it matters

The most concrete consequence of algorithmic curation is that your feed is not a reflection of your social connections. It’s a prediction of what will keep you on the platform. These are related but not the same thing.

In 2023, Pew Research Center found that 64% of American adults said social media had a mostly negative effect on the way things are going in the country, even as the same people continued to use it heavily. Their 2023 study on social media and well-being captures this disconnect. The gap between what people say they want from social media and what the algorithm serves them is measurable and persistent.

For businesses, the stakes are financial. Organic reach for Facebook Pages has declined from roughly 16% in 2012 to under 2% in 2023, according to research by Social@Ogilvy and subsequent tracking by Hootsuite. Brands that built audiences of millions of followers found those audiences effectively unusable without paid amplification. The algorithm shift from chronological to ranked feeds transferred enormous value from publishers to the platforms themselves.

Common misconceptions

“The algorithm shows you what you want to see.” It shows you what it predicts you’ll engage with, which is often what makes you feel angry, validated, or anxious, not what you’d consciously choose to consume if asked. The goal is engagement, not satisfaction. Internal research at Meta found that users often report feeling worse after using the platform, even while continuing to use it.

“Posting more frequently increases your reach.” Frequency has little direct effect on algorithmic reach and can reduce it. Most platforms penalize low-quality or low-engagement posts by down-ranking subsequent content from the same account. A post that generates low engagement can “train” the algorithm to show your next post to a smaller initial audience, resulting in lower overall reach despite higher posting frequency.

“The algorithm is a single system.” Every major platform runs multiple algorithms simultaneously. Instagram alone has separate ranking systems for Feed, Stories, Reels, Explore, and Search. Each has different signals and optimization targets. A piece of content can be strongly boosted by the Reels algorithm and simultaneously ignored by the Feed algorithm. Creators who understand this publish content formatted specifically for each surface rather than posting identical content everywhere.