Social Media Algorithms in 2026: How Each Platform Decides What You See

Social media algorithms are the ranking systems that decide which few posts you see first out of everything you could see. Every large ranked feed runs the same four steps. What separates the platforms is how much of that machinery each one admits to in its own documents, and whether you can switch the ranking off.

I read the platforms’ own paperwork instead of the guides written about them. Between October 15 and 17, 2026, I went through Meta’s index of 30 system cards and the ranking code X published on GitHub, the public code-hosting site. I read the help pages of Instagram, Facebook, Threads, TikTok, YouTube, LinkedIn, Pinterest, Snapchat and Reddit. I also read the European Commission’s page on the Digital Services Act, the New York attorney general’s release on final rules for minors’ feeds, Florida’s statute, five appeals-court opinions on state laws and the studies behind the claims people repeat. Where a company says nothing, I say so. If you only want to reset your feed, jump to the reset table.

What is a social media algorithm?

A social media algorithm is a set of models that scores every post you’re eligible to see and shows the highest scores first. The score is a prediction: how likely you are to like, comment, share, watch to the end, or report this particular post. Meta’s public explainer for the Facebook Feed, which it calls a system card, describes a lightweight first pass that keeps “approximately 500 of the most relevant posts,” then scores only those. X says the algorithm behind For You, its ranked tab, narrows “the roughly 500 million posts made daily down to a handful of top posts.” Neither company ranks everything for everyone. Each builds a short list for you, predicts your reaction to every item on it, and sorts.

Meta’s own reason for ranking at all is short. Its Facebook ranking page says that “most people have more content in their Feed than they could possibly browse in one session,” so its systems try to put first “the content that is most valuable to them.” That is the case for social media algorithms in one line: there’s too much to read, so something has to choose. What it chooses, and on whose behalf, is the rest of this article.

Six terms do most of the work from here on. Candidates are the posts eligible for your feed before anything is ranked. A prediction is the model’s estimate that you’ll take one action, such as a reply. The ranking score combines those predictions into one number per post. Personalization means the score uses what the platform knows about you, and European law calls that profiling. A chronological feed skips the score and sorts by time. A system card is Meta’s name for a public page that explains one of its ranking systems in plain language.

The models are large, but the idea is plain. Meta says its Facebook Feed ranking “has over 100 different prediction models,” and X’s 2023 code shows how it combines such predictions: it multiplies each predicted probability by a weight and adds up the results. When people talk about “the algorithm,” they mean that sum and the filters wrapped around it.

More Read

X (Twitter) Algorithm 2026: What’s Confirmed, Contested, and Myth

Substack Alternatives in 2026: What Leaving Actually Costs

Community Platform Pricing in 2026: Who Takes a Cut of Your Growth

beehiiv vs Substack in 2026: Which One Actually Costs Less

Buffer Alternatives in 2026: What Each One Bills For, and What It Keeps When You Leave

From chronological feeds to ranked ones

The companies said in their own words why they moved to ranking. YouTube wrote in 2012 that it had spent the past few months making changes “to encourage people to spend more time watching,” and added that longer viewing “also opens up more opportunities to generate revenue for our partners.” Twitter, now X, began ranking its timeline on February 10, 2016, as a setting you switched on yourself. A month later, on March 15, 2016, Instagram justified ranking its feed with one line: “people miss on average 70 percent of their feeds.”

The share of posts chosen by a machine kept growing after that. On Meta’s July 2022 earnings call, Mark Zuckerberg said “about 15% of content in a person’s Facebook feed” came from AI recommendations of accounts the person doesn’t follow. He expected the number “to more than double by the end of next year.” On April 24, 2024 he gave the update: “about 30% of the posts on Facebook feed are delivered by our AI recommendation system,” and on Instagram it was more than half of the content people see.

The same companies also built ways back. Instagram’s Following and Favorites views arrived in March 2022, and Meta said both “will show you posts in chronological order.” Facebook’s Feeds tab followed in July 2022 with “no Suggested For You posts,” although ads stayed in. In January 2023, Twitter replaced its Home and Latest tabs with For You and Following. The ranked feed stayed the default in each app. The chronological one became a tab you have to choose.

The latest turns came from regulators and from Meta itself. In August 2023, as Europe’s Digital Services Act took effect, Meta and TikTok launched non-personalized feeds for European users. On October 13, 2026, Facebook added a place to tell its ranking, in your own words, what you want more or less of.

How do social media algorithms work? The four stages

Social media algorithms work in four stages: gather candidate posts, predict how you’ll react to each, filter or demote what breaks the rules, and assemble the final feed with ads and recommendations mixed in. Meta’s Facebook documents describe each stage in the company’s own words, and X and YouTube add details of their own. The stages below use only what the companies themselves wrote.

1. Gather the posts you could see.
2. Predict how you’ll react to each one.
3. Filter out or push down what breaks the rules.
4. Assemble the feed, with ads and suggested posts slotted in.

Stage 1: gather the candidates

The first job is to throw most posts away cheaply. In a 2021 explainer, Meta’s engineers walked through one user’s feed and wrote that “a lightweight model narrows the pool of candidates to approximately 500 of the most relevant posts,” so that heavier models only have to score those. It’s the same figure the Feed card gives today. X describes the same funnel at a much larger scale, starting from roughly 500 million posts a day. Depending on the feed, the candidates come from accounts you follow, from accounts you don’t, or both. Meta, for one, ranks recommended Facebook posts with separate systems, as the last stage shows.

Stage 2: predict your reaction

Each candidate gets a set of predictions, and each prediction gets a weight. Of the platforms in this article, X is the only one I found that has printed the weights of its ranking model, first in 2023 and again in its 2026 repository. A README, the explanatory file in its repository, dated April 5, 2023, weights the predicted chance of a like at 0.5, a reply at 13.5, a reply the author then engages with at 75, and a report at -369. The same file warns that the weights “can be adjusted at any time” and that there “may be small differences from the production model.”

Those numbers multiply probabilities, not counts. Say the model gives you a 2% chance of replying to a post and a 60% chance of liking it: the reply adds 13.5 times 0.02, or 0.27, and the like adds 0.5 times 0.6, or 0.3. In that case the like counts for slightly more. Even though 13.5 is 27 times 0.5, a reply isn’t “worth” 27 likes: it adds more than a like only when your chance of replying is more than a 27th of your chance of liking. Meta prints the categories of its predictions without weights. Its Instagram Feed card lists signals such as “How recently the post was created (within 90 days),” which makes freshness one input among many.

Meta sorts its predictions into four groups: “actions you’ll take on the post,” “how you’ll spend time viewing the post,” “your interest in the post or person, Page or group that shared the post” and “how others will interact with the post if you take a certain action, such as commenting or sharing a post.” The last group is the least intuitive part of social media algorithms. A post can rank higher in your feed because of the conversation it might start among other people.

Stage 3: filter and demote

Before anything is shown, posts pass rule checks. X’s help page says that “Prior to showing you content,” it “filters content that might be harmful, abusive or spammy.”

Meta’s 2021 walkthrough describes a step that decides “which integrity detection measures, if any, need to be applied to the stories selected for ranking,” its phrase for checks against the rules. Some demotions are specific to who you are: YouTube says it is “limiting repeated recommendations” of videos that idealize body types or fitness levels “for teens globally.”

Stage 4: assemble the feed

The last step mixes in content the ranking model didn’t choose. Meta’s Facebook Feed card says suggested posts and ads in your Feed are “not powered by the AI that we describe in this system card.” They come from other systems with their own cards. X says promoted posts can appear in For You. So the feed on your screen is several rankings stitched together. If you want to see how a different kind of model predicts the next item, our explainer on how a model like ChatGPT predicts covers the same idea for text.

One app, many algorithms

There’s no single Instagram algorithm or Facebook algorithm. When I counted on October 16, 2026, Meta’s transparency index listed 30 system cards: 15 for Facebook, 10 for Instagram including the Threads feed, and 5 for its VR products. Meta launched 22 cards in June 2023, 14 for Facebook and 8 for Instagram, and said each shows “the top ten most important prediction models, rather than everything in the system.”

Each card covers one surface. Feed, Reels, Stories, Explore, search, notifications and suggested accounts each have their own ranking system. Meta’s Threads card even lists signals measured over different windows, from the posts you liked “in the last 15 minutes, up to a maximum of 100 IDs” to your engagement “over the past week.” That’s why your Reels can feel nothing like your feed in the same app on the same afternoon.

Controls cut across those surfaces in different ways. Instagram’s help page calls its newer “Your algorithm” setting “one unified system,” so a topic you add changes Feed, Reels and Explore together. TikTok goes the other way: its Manage topics setting “only applies to content in your For You feed.” When you find a control, first ask which feeds it touches.

The smaller platforms describe the same split in fewer words. Reddit’s help page calls its Best sort “a personalized sorting option” for your home feed. Pinterest says it “recommends Pins to you based on your saves, searches, and browsing activity.” Snapchat’s page on Spotlight, its feed of short videos, says that if you “skip, report or hide content,” it reads that as a sign “that you don’t enjoy the content.” It also mixes topics on purpose: “We think diversity of content is important to avoid echo chambers.” Each of those is a separate social media algorithm for one surface of one app.

What platforms actually disclose about their algorithms

What each platform discloses depends less on the company than on the document. I sorted every page I read into four levels: what a law requires, what a company published by choice, what sits only in help pages, and what is guesswork. Most platforms sit on several levels at once, and the level of a claim tells you how much weight it can bear.

Required by law

Of the laws I read, only the European Union’s makes a non-personalized feed a legal duty for adults. Under Article 38 of the Digital Services Act, very large platforms, those with at least 45 million monthly users in the EU, “must offer an option to turn off personalised content,” in the European Commission’s words. On its page last updated May 19, 2026, the Commission names TikTok, Facebook and Instagram as currently offering one. Europe shows up in odd places. LinkedIn’s help page says its Most recent view doesn’t use interest data to sort the feed. Yet it adds that outside the European Economic Area that view shows “the updates most relevant to you in reverse chronological order,” while inside it you “will see all updates from your network.” The European Economic Area is the EU plus Iceland, Liechtenstein and Norway.

In the US, the laws written so far cover only minors, and New York’s, which names feeds directly, hasn’t started yet. New York’s SAFE for Kids Act requires platforms “to restrict algorithmically personalized feeds and nighttime notifications for users under the age of 18 unless they obtain parental consent.” The law takes effect on January 25, 2027, under final rules the state attorney general released in July 2026. California’s law on personalized feeds for minors survived an early challenge on September 9, 2025, when the Ninth Circuit agreed with a lower court not to block its personalized-feed provisions and ordered only the rule on like counts blocked.

Some state laws define their target by the algorithm itself. Florida’s 2024 law covers a platform that “[e]mploys algorithms that analyze user data or information on users to select content for users” and has features such as infinite scrolling. In Europe, the duty is now being tested. On July 10, 2026, the Commission said it had “preliminarily found Meta in breach” of the Digital Services Act, in an investigation that “focuses on features such as infinite scroll, autoplay, push notifications, and the platforms’ highly personalised recommender systems.”

Published by choice

X goes furthest. It published its recommendation code on GitHub in March 2023, with a README listing ranking weights that it revised on April 5, and opened a new repository for its For You feed in January 2026. Meta publishes system cards that name its predictions and signals but print no weights. Reddit published the code of its whole site, including the “hot” formula that sorts posts by votes and age, until it archived the repository in 2017. The archived code is still public, but the “hot” rule is a fixed formula, not the weights of a trained model. Bluesky lets anyone build a feed: its 2023 post says “anyone can create a feed” and calls the result “algorithmic choice.”

Code you can read still can’t show you your own feed. X’s 2023 documentation says it “cannot release the real training data due to privacy restrictions.” Its 2026 repository leaves out some files on purpose, including some anti-spam rules, which X says it held back to make the system harder to game. That repository also adds a warning of its own: the weights apply to the predicted probability of an action. Open code tells you the recipe, not the ingredients.

Help pages only

TikTok, YouTube, LinkedIn, Pinterest, Snapchat and Reddit explain their ranking mostly through help pages: which signals count, which buttons exist and what they change. Those pages are useful and usually undated, and they describe controls more than models. TikTok’s US joint venture, announced in January 2026 with ByteDance keeping 19.9%, adds a different kind of disclosure. Its January 23, 2026 announcement says it “will retrain, test, and update the content recommendation algorithm on U.S. user data,” secured in Oracle’s US cloud, which tells you who controls the model without telling you how it ranks.

Guesswork

The rest comes from vendors who sell scheduling and analytics tools. Sprout Social’s guide, published July 1, 2026, says LinkedIn’s models predict engagement “with the first hour carrying the most weight,” with no source attached. Of the 14 pages ranking at the top of Google for this topic that were audited for this article on October 17, 2026, none mentioned the EU option or a US state law on minors’ feeds, and none gave step-by-step reset instructions. Treat any rule that names no platform document as a hypothesis.

Platform by platform

Each of the big social media algorithms below gets the same three questions: how it ranks, what the company admits in writing, and what you can change. The disclosure level at the end of each block comes from the ladder above, and the link goes to our deeper guide for that platform.

Facebook

Meta documents Facebook’s feed in detail. Its Feed card describes the gather, predict and rank steps and says tapping “Interested” or “Not interested” will “temporarily increase or decrease the ranking score” of a post and posts like it. On October 13, 2026, Meta added “Your algorithm,” a place in the three-dot menu where you “add and manage the topics you want to see in Feed and Reels.” Meta says you can describe what you want “in your own words.” Its Feeds tab, Meta says, “allows you to see the newest posts first.” About 30% of Facebook feed posts come from its AI recommendation system, by Meta’s own April 2024 count.

Disclosure level: published by choice, plus the EU option. For what drives reach on the platform, see how Facebook ranks your feed.

Instagram and Threads

Instagram runs separate rankings for Feed, Reels, Stories and Explore, each with its own card. You can reset suggested content, and the help page says the reset “can’t be undone, but your suggestions will become more personalized again as you like, share and interact.” Instagram still describes its own “Your algorithm” setting, which shares the name of Facebook’s, as “in testing.” Threads, which Meta files under Instagram, has its own card, and it says choosing Following shows posts “in reverse chronological order. When you select this option, you won’t see recommended content.”

Disclosure level: published by choice, plus the EU option. The creator side is in Instagram’s algorithm in detail.

TikTok

TikTok’s For You feed is its recommendation feed, and the company explains it through help pages rather than model cards. The Refresh option shows “some popular content” and, the help page warns, “can’t be undone.” In the EU and Kazakhstan, turning off personalized feeds means “The For You feed will become a Popular feed and serve popular content from your region and internationally.” Its keyword filter lets you “remove content containing specific words and their variations, and hashtags” from the For You, Following, Friends, LIVE and STEM feeds. In the US, the joint venture says it will retrain the recommendation model on US data inside Oracle’s cloud.

Disclosure level: help pages, plus the EU option. For how the For You feed scores video, see how TikTok ranks video.

YouTube

YouTube ties its Home recommendations directly to your watch history. Its help page says: “If you don’t want video recommendations on Home, you can delete and turn off your watch history.” A second page goes further. With history turned off and no relevant recent history, “recommendations on the YouTube homepage are removed.” If you pause history only for a while, the page adds: “Remember to resume your history when you’re done.” Your “Not interested” and “Don’t recommend channel” clicks “may be used to tune your recommendations” until you clear them. The Commission’s page doesn’t list YouTube among the services offering a non-personalized feed.

Disclosure level: help pages. The creator side is in YouTube’s recommendation system.

LinkedIn

LinkedIn shows Top posts by default and lets you switch to Recent, or make Most recent your permanent view in settings. The help page promises that in Most recent “Your interests and activity data will not be used to sort the feed,” then adds the regional exception above: outside the European Economic Area, you get the most relevant updates in reverse chronological order. Time sets the order there, but you don’t see every update. On October 15, 2026, LinkedIn said it is testing a separate Network feed in the US, India, the UK and Australia.

Disclosure level: help pages. For what changed in reach, read what LinkedIn changed in its feed, and for the account risks of automation, what LinkedIn actually allows.

X

X publishes more about its ranking, including code and weights, than any other platform in this article, and it offers a plain alternative. Its help page says you “can always choose to see posts that are not personalized for you by selecting the Following timeline,” which shows posts from accounts you follow “in reverse chronological order.” X also says that when you close the app on For you or Following, “you will return to whichever timeline you had open last.” Not every user sees that: in July 2026 a Reddit user wrote that the app opened on For You despite being closed on Following. Promoted posts can still appear in For You, X says.

Disclosure level: published by choice.

Pinterest, Snapchat and Reddit

The three smaller platforms in this article each publish one clear control. Pinterest’s “Refine your recommendations” page lets you “Turn off recommendations based on Pins you have engaged with,” and adds a toggle to reduce AI-generated Pins in your feed. Snapchat’s page on Spotlight explains ranking in plain terms: what you watch, share or favorite counts for a video, and skipping, reporting or hiding counts against it. Reddit lets you turn off “Show recommendations in home feed” in your preferences on the web, or under Privacy in the app’s account settings. Disclosure level for all three today: help pages, plus Reddit’s archived site code.

Can you reset or turn off the algorithm?

Most platforms let you retrain your feed in some way, but a reset isn’t an off switch. Instagram and TikTok both say their resets can’t be undone, meaning you can’t restore the old feed with a click, and Instagram says personalization rebuilds as you use the app. You can turn ranking off in fewer ways. One is a chronological tab, which you may have to pick again each time you open the app. Another is a YouTube setting tied to watch history. In the EU, the law also requires an option to “turn off personalised content.”

How to reset or switch off each feed

These are the paths the platforms’ own help pages gave when I checked on October 16 and 17, 2026. Menus move between app versions, so look for the same words if a step has shifted.

1. On Instagram, tap Profile, then Menu, then Content preferences, then Reset suggested content. It works only in the app. For time order instead, tap Instagram in the top left of your feed, then Following.
2. On TikTok, tap Profile, then the Menu button, then Settings and privacy, then Content Preferences, then Refresh your For You feed.
3. On Facebook, open the three-dot menu on a post to reach Your algorithm and add or remove topics, or open the Feeds tab to see the newest posts first.
4. On X, select the Following timeline instead of For you.
5. On LinkedIn, click the Me icon, then Settings & Privacy, then Preferred Feed View, and choose Most recent posts.
6. On YouTube, delete and turn off your watch history if you don’t want recommendations on Home.
7. On Reddit, turn off Show recommendations in home feed in your preferences or in the app’s Privacy settings.

The resets work as described, and users report the same limit. In August 2026, a Reddit user who reset Instagram daily for a week wrote that “the effect quickly wears off” and that “a couple of minutes and it’s back on the content I like. So, the resetting needs to happen daily.” TikTok’s Refresh drew a similar report in October 2026: “I feel like I have to do this every day now!”

Chronological views are steadier, but only some of them remember your choice. Facebook’s Feeds tab and Instagram’s Following view show recent posts from accounts you follow. Instagram’s help page doesn’t say the choice survives a restart, and a European user wrote in 2023 that to stay in Following “you’ll have to do it each time you open it.” X promises to remember your last tab, and LinkedIn lets you save Most recent as a permanent preference.

Some controls nudge rather than reset. Facebook’s card says “Interested” and “Not interested” change a post’s score only temporarily. TikTok’s “Not interested” promises “we’ll show you less content like that.” YouTube’s route is the bluntest: with history off and no relevant recent history, it removes Home recommendations. In April 2026, a Reddit user who kept history paused wrote: “Getting no recommendations on YT home page on the iOS app this morning.”

Europe’s non-personalized feeds look different from platform to platform. When Meta launched them in 2023, it said Facebook and Instagram users could view Stories and Reels “only from people they follow, ranked in chronological order, newest to oldest.” TikTok said its non-personalized For You feed would “recommend both locally relevant and globally ‘popular’ videos.” US law doesn’t require either option for adults.

Two apps also tell you why you saw a post. TikTok describes its “Why this video” option as a way “to check why they are seeing a certain video in their feed.” Instagram’s “Your algorithm” page, still in testing, has a “Why you’re seeing this” link for each interest you add, and it labels posts recommended from those interests with tags like “From Running.” So Instagram’s tool explains posts tied to interests you added, not every post.

Which social media platforms don’t use algorithms?

Of the services I checked, Mastodon, a network of independently run servers, is the only one whose own app description promises no ranking at all. The official app from Mastodon GmbH says: “Follow anyone across the fediverse and see it all in chronological order. No algorithms, ads, or clickbait in sight.” The fediverse is the wider set of independently run servers that can follow one another.

Bluesky, another independent network, takes a different route. Its default Discover feed is ranked, “customized for what you like to see” in the company’s words, but the same 2023 post says “anyone can create a feed,” and users can swap the default for feeds built by others. One Bluesky user described its separate Following feed this way in November 2025: “The main feed on Bluesky (Following) is a chronological timeline without any algorithm.”

Inside the big apps, the closest thing to “no algorithm” is a tab. Facebook Feeds, Instagram Following and Favorites, Threads Following, X Following and LinkedIn’s Most recent view sort by time, though outside the EEA LinkedIn shows only the updates it rates most relevant. In the EU, TikTok says that with personalization off, your Following and Friends feeds “may feature content from creators you follow in chronological order.” Newsletters and community platforms reach people without a ranked feed in between, which is part of why creators move there; we covered what that move costs in newsletters that reach readers without a feed and in community platforms.

Chronological isn’t neutral either. Arvind Narayanan of the Knight First Amendment Institute wrote in 2023 that “‘reverse chronological’ is an algorithm, albeit a simple one,” with its own “rich-get-richer effects.” My count covers the ten platforms in this article plus Mastodon and Bluesky, which I checked on October 16 and 17, 2026. It leaves out messaging apps and smaller networks.

What research says about algorithms’ effects

Research on social media algorithms says that ranked feeds change what you see and how long you stay, and that switching a feed off for a few months rarely changes what people believe. The best studies are large field experiments with consenting users, and they disagree less than headlines suggest once you note what each one measured, for how long and on which platform.

Two large experiments came from the 2020 US election. In a study published in Science in 2023, researchers moved consenting Facebook and Instagram users to chronological feeds for three months. Time on the platforms “substantially decreased,” and users saw more political and untrustworthy content, yet the chronological feed “did not significantly alter levels of issue polarization, affective polarization, political knowledge, or other key attitudes.” A companion study in Nature cut exposure to like-minded sources by about a third for 23,377 users and found “no measurable effects on eight preregistered attitudinal measures.”

X tells a different story. In a seven-week experiment run in 2023 and published in Nature in February 2026, switching users from a chronological to an algorithmic feed “increased engagement and shifted political opinion towards more conservative positions.” For users moved the other way, from the algorithmic feed to the chronological one, the switch “had no comparable effects.” The authors traced the gap to accounts people started following and kept following. Twitter’s own researchers had already reported in PNAS that “the political right enjoys higher amplification compared to the political left” on its ranked timeline.

Several teams have now re-ranked real feeds from outside the platforms. A 2025 Science study used a browser extension on X with 1,256 people for 10 days. Turning hostile political posts up or down shifted feelings toward the other party “by more than 2 points on a 100-point feeling thermometer.” The 2026 Prosocial Ranking Challenge preprint tested five alternative rankers on 9,386 desktop users for six months. On average, they cut polarization by 0.03 standard deviations, a small effect, with slightly less time on Facebook and Reddit and slightly more on X. A 2025 PNAS Nexus audit found that Twitter’s engagement ranking “amplifies emotionally charged, out-group hostile content” and that users didn’t prefer the political tweets it picked.

On YouTube the studies split. A 2021 PNAS study of real browsing found “little evidence that the YouTube recommendation algorithm is driving attention” to far-right content, and a 2020 study of anonymous recommendations argued they favor mainstream channels. The oldest large study is a reminder that people choose too. Science’s summary of a 2015 Facebook study says users “encountered roughly 15% less cross-cutting content in news feeds due to algorithmic ranking and clicked through to 70% less of this cross-cutting content.” The ranking narrowed the news a little, and people’s own clicks narrowed it more.

For teens, a widely quoted number comes from a 2022 test by the Center for Countering Digital Hate. Its researchers set up accounts aged 13 and found TikTok recommended a suicide-related video “within 2 minutes and 38 seconds” at the earliest. That was an advocacy group’s test, not a peer-reviewed study.

What these studies left out: effects over years rather than months, effects on teenagers’ health rather than on what they’re shown, and TikTok, which none of the large experiments above covered.

Myths about social media algorithms

These claims about social media algorithms come up again and again in guides and AI answers. Here I set each one against what the platforms themselves wrote.

“The first hour decides your reach.” It’s a staple of creator guides, and no platform document I read states it. On the related question of when to post, YouTube’s help page says: “Publish time is not known to impact a video’s long-term performance.”

“Newer posts rank higher.” Google’s AI Overview, the AI-written answer at the top of its results, said for “social media algorithms” on October 15, 2026, that “Newer posts usually rank higher than older ones.” Instagram’s own card lists recency as one signal, “How recently the post was created (within 90 days),” among many signals about the post and about you.

“A reply is worth 27 likes on X.” That ratio comes from dividing two weights in X’s 2023 README, 13.5 by 0.5. The weights multiply predicted probabilities, and an earlier version of the same file gave a reply a raw weight of 27, double the revised 13.5. X’s 2026 weights file uses different values again. None of these numbers is a price for one reply.

“70% of YouTube watching comes from recommendations.” The figure traces to a panel at CES, the annual Las Vegas tech show, in January 2018, where YouTube’s chief product officer Neal Mohan spoke. CNET’s report of the panel says “more than 75 percent”; Tubefilter’s report the next day put it “to the tune of 70%.” I found no YouTube transcript or post with either number.

“A reset wipes what the app knows about you.” Instagram’s help page says the opposite: “Resetting won’t delete your data. We will still use your data to personalize your experience on Instagram in other ways, to show you ads.” A reset clears your suggestions and leaves the profile behind them in place.

“I’ve been shadowbanned.” Adam Mosseri, head of Instagram, addressed the word directly: “People often accuse us of ‘shadowbanning’ or silencing them.” In the same post he gave the plainer cause of falling reach: “most of your followers won’t see what you share, because most look at less than half of their Feed.”

If you post: where to go next

This article is for people on the receiving end of the feed. If you publish for a brand or a living, the platform guides go deeper: Instagram, TikTok, YouTube, Facebook, LinkedIn and X. If you’re choosing a social media scheduler, we priced the main ones for a six-channel setup. And if LinkedIn is your main channel, our report on who owns LinkedIn shows what Microsoft’s filings say about the company that runs its feed.

Frequently Asked Questions