LinkedIn

LinkedIn says the home feed works in two stages. Retrieval selects the candidate posts. Ranking decides the order that a member sees [1]. LinkedIn says the feed serves more than 1.3 billion professionals and ranks millions of posts at any given time [1].

LinkedIn says its 2026 retrieval uses large language models (LLMs). It can surface posts from authors a member does not follow, on topics the member cares about [1]. LinkedIn also says exceptional content may reach members who do not follow the author [5]. For a new reader, relevance can matter more than follower count.

LinkedIn says its 2026 ranking uses Generative Recommenders with LLMs. The LLMs read what a post is about [3]. Before critical model launches, LinkedIn tests two views: how posts from different creators compete for the same space, and whether the scroll feels consistent across audiences [2].

• Professional signals and engagement patterns. LinkedIn says ranking does not use age, race, gender, or other demographic data [1][2][11].
• Hundreds of signals, including the post's context, and signals from the member's profile, network, and activity [11].
• The author's professional identity, such as experience, content on the profile, and skills [5]. LinkedIn has not said that company pages rank differently from personal profiles [137][138].
• Profile details the member chose to share, such as industry, experience, skills, and geography [138].
• The member's preferences, interests, connections, and past engagement [5].

LinkedIn trains models to predict click, react, comment, and share. It then combines the predictions into one score using a weighted sum [4]. LinkedIn says a comment is much more valuable than a click [115]. LinkedIn does not publish numeric weights for these actions.

• LinkedIn added a P(skip) objective. It predicts the chance that a member's dwell time on a post is below a threshold. The model lowers the post's score in proportion to that chance [4].
• Adding P(skip) raised the model's area under the curve (AUC) by up to 10% over several trainings [4].
• LinkedIn calls dwell time "always measurable" and a more reliable signal than binary clicks [4].
• Dwell has two parts: time "on the feed" (from when at least half of a post is visible) and time "after the click" [244].
• In 2024, LinkedIn set dwell thresholds by percentile. The thresholds update daily, to limit "dwell bait" [245].

• Richard van der Blom's 2025 study of 1.8 million posts says: "Saved posts now drive future reach. A like means little. A save means everything" [84].
• The same study says saves, comments, and clicks on polls or carousels carry 3 to 5 times more weight than likes [86].
• AuthoredUp reports that one save gives a post 5 times the reach of one like, and twice the value of a meaningful comment [249]. Hootsuite reports a similar ratio from a 3 million post analysis [250].

These ratios come from third-party analyses. LinkedIn does not publish them.

• LinkedIn's creator-side model says the first few pieces of feedback matter most. Each response after the first few adds less value [233].
• LinkedIn adjusted candidate selection by post freshness. Creators get feedback sooner, and viewers stay in active conversations [232].
• LinkedIn has not published a timed window for reach. See Part 4.

FollowFeed is LinkedIn's main first-pass source for feed posts. LinkedIn says more than 80% of feed updates come from FollowFeed, and those updates contribute to more than 95% of members' conversations [6]. The candidate model also predicts a member's intent to share, comment, or react, in addition to clicks [6].

• First-pass rankers select the candidates. A second-pass ranker scores them and makes one personalized list [6].
• LinkedIn says its generative recommender reads more than 1,000 of a member's past interactions [1].
• Converting numeric features into percentile buckets raised retrieval recall@10 by 15% [1].
• A 2016 engineering post says FollowFeed powers all LinkedIn feed experiences. It gives no share of feed updates [12].
• LinkedIn's research model 360Brew has 150 billion parameters and handles more than 30 prediction tasks [214]. The arXiv paper is withdrawn [214]. A conference talk describes its design [215]. Treat this as background, not as a posting rule.

Four US patents describe feed ranking methods. Their priority dates run from 2011 to 2017. They show how the feed was designed in those years. They do not prove how the current feed works.

• US10762583B2 ranks items by metrics such as click-through rate. It also weights first-degree interactions by connection strength in the second-degree network [7].
• US11514115B2 scores each feed item from the chance of interaction, the chance of downstream engagement by others, and the value of that engagement. It lists click, like, share, comment, job view, and connect as interaction types [8].
• US10504191B2 selects content for a member segment when its current engagement passes quantile thresholds. It then shows the content to members who have not seen it [9].
• US9674136B2 shows a post to a target member when a member who engaged with it shares an endorsed skill and experience level with the target [10].

• LinkedIn will show less content with a video that does not match its text. That video "should no longer gain additional reach" [3].
• LinkedIn will show less engagement bait, such as "Comment 'Yes' if you agree" [3].
• LinkedIn will show less recycled thought leadership that lacks substance [3].
• LinkedIn says it is making engagement pods ineffective and curbing comment automation [3].
• New members see an Interest Picker at sign-up, to choose topics for their feed [3].
• Richard van der Blom calls the 2025 reach drop "the biggest shift in the LinkedIn algorithm since 2016" [83]. Creators reported sharp view drops that they took for shadowbanning [254].
• For Suggested posts (from outside a member's network), LinkedIn lists quality, professional relevance, and feedback signals. It does not list post age [234].

Notes on the table:

• Multi-image posts generate the most impressions per post on pages with up to 50K followers [63].
• Multi-image posts and PDF carousels each get 1.7 times the impressions of a typical post [73].
• Metricool's 2025 study reports carousels at a 45.85% engagement rate, the highest of any format. It reports video posts grew 53% in a year [65].
• Posts with at least one image get two times more comments than posts without one [59]. LinkedIn also says video gets 1.4 times more engagement than other formats [55].
• Video reach fell sharply. Median reach dropped over 72%, from 1,764 to 477 [67]. Socialinsider reports video views fell 36% year over year across all page sizes [63].
• Video results conflict. Van der Blom reports a 3.4 times reach multiplier and 0.9 times engagement [69]. He also reports video reach fell 300% in three months to February 2025 [70]. LinkedIn reports video is watched 36% more year over year [60].
• Vertical, captioned videos of 30 to 60 seconds did best in one study (1.49 reach amplifier, up 23%). Horizontal video did worst [71]. LinkedIn also recommends 30 to 60 second videos [59].
• Polls allow 2 to 4 options. The default duration is one week [58].
• Polls reach more people per post, but their median likes are low. A median poll reaches 1,154 people, against 760 for other posts. A median poll gets 6 likes, against 34 for an image post [72]. Poll use fell from 1.14% of posts in 2025 to 0.54% in 2026 [72].
• Standalone long-form articles reach about 0.4 times a normal post's reach [98].
• LinkedIn's API lists Carousels as supported for sponsored posts only, not organic posts [62].
• A reach multiplier compares a format's reach with each profile's own median post [265].

• Documents can be up to 100 MB and 300 pages. Supported types: PPT, PPTX, DOC, DOCX, and PDF [57].
• Van der Blom's 2025 analysis of 1.2 million posts found 8 to 12 slides works best. The first slide carries about 40% of the result [241].
• Socialinsider says audiences value documents as free value they can keep, such as reports, frameworks, and templates [242].
• Socialinsider suggests educational carousels, step-by-step frameworks, and lessons from experience. It also suggests re-cutting popular videos and posts into document format [242].
• Documents lead on relative reach, but their median reach fell 25% year over year [243]. Relative leadership and absolute reach moved in opposite directions.

• The sweet spot in one 2025 study was 800 to 1,000 characters [82].
• A separate study of 372,126 posts found posts of 1,301 to 2,500 characters earned 27% more engagement than posts under 400 characters [246]. The two studies conflict. Test both lengths on your own audience.
• The standard post limit is 3,000 characters. LinkedIn raised it from 1,300 in June 2023 [246].
• Over 91% of LinkedIn browsing happens on mobile [82].

No study found with a stated sample and method measured organic reach for LinkedIn Live. Published Live figures are advertising benchmarks, not organic reach [218].

• Organic company content is about 2% of what members see in the feed. Top creators make up 31% and other creators 28%. These figures come from van der Blom's October 2024 analysis [135][136].
• AuthoredUp states that organic company page posts are "invisible in the feed". This is the author's opinion, not a measured result [135].
• Organic company page updates reach few people. Copying personal posts to the company page adds little. Write native updates for the page [135].

• On personal profiles, text posts have a 0.88x reach multiplier. On company pages, text posts have 0.42x [135].
• On company pages, documents lead at 1.40x. Images reach 1.21x. Polls fell from 1.64x to 1.19x. Video reached 1.05x [66].
• For pages with 1 to 5K followers, Socialinsider reports average impressions per post: native document 525, multi-image 490, poll 340, video 355, image 315, text 300, and link 235 [134].
• Page growth slowed. Pages with 1 to 5K followers grew 24.5% in 2025, against 40.75% in 2024. Pages with 100K to 1M followers grew 6.4% [134].

• LinkedIn's marketing blog says employees can have up to ten times the followers of a company page. It says content shared by an employee gets twice the click-through rate of the same content shared by the company [122].
• LinkedIn says employees who enroll in its advocacy program share content five times more often [121].
• A 2018 survey of over 2,500 employees and 400 leaders reports that branded messages reached over 5 times (561%) further when employees shared them [140][141].
• Personalized employee shares drew about 64% more engagement than non-personalized shares [140].
• A Sprout Social case: 87 employee shares of one article reached nearly 233,000 people [140].

These figures come from vendors and from LinkedIn's own marketing. Treat them as reported, not tested.

• Since March 2024, Follow is the default profile button. Members can choose Connect instead [101].
• The About section moved to the top of the profile in March 2024 [101].
• Creator mode was removed as a toggle in March 2024. Sharing tools and analytics stayed [101].
• You can highlight a newsletter in the Featured section of your profile [102].
• Creator analytics are available to all members. There is no follower threshold [106].
• The profile custom button is available only to a select group of Premium Business, Sales Navigator, and Recruiter Lite subscribers [104].
• Top Voices is an invitation-only program. Active members get a blue badge. LinkedIn reviews Top Voices twice a year. Members who mainly promote paid services are not considered [109].
• The Creator Accelerator Program was closed to applications at the time of research [107].

• Every comment usually creates a Comment Viral Update. It shows to the commenter's own connections [220].
• Comment Viral Updates get 2.5 times the interactions per impression of connection updates. They get 1.8 times the interactions of like-based viral updates [117].
• Comment ranking is per viewer. A model with close to 100 features predicts each viewer's engagement with each comment. The features include the commenter's reputation, profile views, shared attributes, and past interactions [220].
• The top 1% of long comment threads draw more than 40% of member visits [220].
• Comments are allowed. Automation and pre-arranged engagement are not. LinkedIn may hide automated comments from "Most relevant" and sometimes from people outside the commenter's network [222]. Pre-arranged engagement breaks the rules in the policy [112].
• The default sort is "Most relevant". Some comments show only under "Most recent" [219].
• A post's author can turn comments off, or limit them to connections [219].
• Trade press reports a 2026 change: comments display by relevance to each user. Trade press also reports LinkedIn's Q2 2026 figure of 18% more time spent in comments [224]. LinkedIn has not published details of this change [231].
• One member saw a "Has verifications" comment sort option in April 2026. This is an observed product state, not a LinkedIn announcement [230].

• No controlled study found shows that commenting on large accounts adds followers or profile views.
• One named two-week test, with 10 comments a day, saw profile views fall 17%. Follower growth rose 0.9%. The author reported no baseline to judge a rise [226].
• Van der Blom claims a good comment strategy raises profile views 6 times, follower growth 3 times, and invites from the ideal audience 2 times. He gives no sample or method for these figures [227].
• Comments from second-degree connections have 2.6 times more impact on a post's reach than comments from first-degree connections. Comments after the first 90 minutes have 1.8 times more impact [228]. These figures describe a comment's effect on the post's reach. They do not describe a commenter's follower gain.
• A vendor synthesis reports a correlation of 0.545 between commenting frequency and follower growth. Top 1% creators leave 286 replies a week, against 34 for average creators [229]. A correlation is not proof that commenting causes growth.
• A 2017 trade article argued that simply commenting is no longer enough. Profile views and relevant comments that earn replies also matter [223].

• LinkedIn's model values early feedback most [233].
• LinkedIn has not confirmed a 60- or 90-minute window for engagement [118].
• Hootsuite says LinkedIn rewards quick engagement after publishing. It cites no LinkedIn source [119].
• Early replies help the creator and the post. No fixed window decides reach.

• Buffer says consistency beats bursts. It says posting three times a week, every week, beats posting daily for two weeks and then going quiet [236]. Buffer does not test a silent period.
• Buffer's recommended start is 2 to 5 posts a week, where most creators and brands start to see results [118].

• Buffer's 2026 study of 4.8 million posts found the strongest times on weekdays from 3 p.m. to 8 p.m. Wednesday at 4 p.m. was strongest [118].
• Hootsuite, with Critical Truth, analyzed over 1 million posts. It found 8 to 9 a.m. on Tuesdays and Wednesdays was best [119].
• Buffer's 2025 peak was limited to working hours. Its 2026 peak moved to the evening. The best time moved within one publisher's own data [235].
• Sector data also conflicts. For the education sector, Hootsuite gives 9 a.m. Wednesday and 11 a.m. to 1 p.m. Friday. Metricool gives Friday at 6 a.m. [237][238].
• Choose one posting window, keep your post type steady, and test for several weeks. The studies give no fixed test length.

• Opening lines under 40 characters earned a 2.61% median engagement rate. Lines over 200 characters earned 2.08%. The study covered 323,111 posts from December 2025 to May 2026 [248].
• Story openers had the highest median engagement, at 2.60%. Contrarian openers reached 2.31%, statement openers 2.27%, and results openers 2.19%. Question openers were last, at 2.16%. The study covered 309,614 posts [248].
• Only 7% of posts use a story opener [248].
• Question openers are the most recommended opener on LinkedIn. In this study they are the weakest [248].
• Put the most interesting part first. A LinkedIn team member advises this, but no effect size is measured [252].

• A post truncates after about 210 characters on desktop and about 140 characters on mobile. A "see more" button hides the rest [246].
• The highest-engaging length band, 1,301 to 2,500 characters, sits almost entirely past the fold [246]. The first line must earn the click. The study does not test this link directly.
• Van der Blom calls a "see more" click "worth 10 empty likes". He gives no method for this weight [251].

• Posts under 100 words are the weakest bucket on all four main formats, at 0.72 to 0.79 times the format median [247].
• Posts that take 60 to 120 seconds to read earn 15% to 43% more than posts that take 30 to 60 seconds [247].
• A post above a 10th-grade reading level is associated with over 35% less reach [135].

• All LinkedIn members can create a newsletter [87][88][92].
• A Page needs more than 150 followers or connections to be evaluated for newsletter access. The Page must have a record of original posts and must be in good standing [87].
• An author can run up to five newsletters at a time [89][95]. After you delete or create one, you must wait two weeks before you create another [89]. A newsletter can hold an unlimited number of editions [95].
• A newsletter page is separate from the Page. A newsletter made for a Showcase Page is not linked to its parent Page [94][95].

• When you publish a newsletter's first edition, your connections and followers receive a notification to subscribe [89][92]. A Page's followers receive the same invitation [95].
• When someone follows you, they see an invitation to subscribe. The invitation is not a subscription. The follower must accept it [88][92][98].
• Subscribing also makes the reader follow the author [88][90].
• Readers find newsletters through the Subscribe option in the feed, the "Trending newsletters in your network" section of My Network, and the Interests section of member profiles [88][90].
• Readers can view a newsletter page without logging in. They must log in to subscribe or to share [94][89].
• Anyone can discover, read, and share a newsletter [92][95].
• Even after unsubscribing, a reader may still see your newsletter in their feed if one of their connections shares, comments on, or likes it [88].

• Subscribers get push, in-app, and email notifications when a new edition publishes [88][92].
• If a subscriber has several notification types on, LinkedIn does not send duplicates. A subscriber who gets an in-app or push alert may not also get the email [88].
• The edition also appears in the feed when the author clicks Post [89][98]. Subscribers see updates in their feed [90].
• Practical rule: publish every edition as a normal feed post as well. Do not depend on the notification.

• Newsletter analytics show impressions, engagements, article views, new subscribers, and subscriber demographics [91].
• Analytics cover editions from the last two years. The date range runs from 7 to 365 days, and applies only to the Trends and Article totals sections [91].
• Article views count clicks from the article page or from a newsletter email. They exclude views from organizations. So the number mixes email and on-platform reads [91].
• Email sends for each edition are on the article analytics page [91].
• The author can see each subscriber's name, photo, job title, and headline, subject to their settings [88].

• LinkedIn recommends sharing editions, or the newsletter page, on other social platforms or by email to increase reach [93].
• LinkedIn recommends changing the profile button from Connect to Follow, so profile viewers can follow your activity [93].
• Add the newsletter to your Featured profile section [102].
• Van der Blom reports newsletters outperform regular posts on reach, engagement, and conversion [100].
• Melanie Goodman's study of the 50 largest LinkedIn newsletters found standalone long-form articles reach about 0.4 times a normal post [98].

You can republish your own original work on LinkedIn. You cannot republish content you do not own. LinkedIn disables accounts that use infringing content [125].

• LinkedIn does not guarantee the order of profile results. Results are tailored to each searcher. One profile can rank on page 2 for one member and page 5 for another [182].
• Results depend on the searcher's activity, the profiles that match the query, patterns from members who ran similar searches, and the searcher's history [183].
• A profile's own activity and network affect where it appears [183]. Closer connections improve the chance of appearing [182].
• A complete profile helps you appear in more searches and improves how the system matches you. This includes current and past positions, schools, and fields of study [182].
• Add skills. LinkedIn says skills are among the most common recruiter queries [182].
• Use standard job titles. LinkedIn says these are what other members search for [182].
• LinkedIn's own advice: keep your profile current, and add relevant keywords to your headline and summary [187].
• Use several relevant keywords. Repeating the same words can trigger spam detection and hurt visibility [183][184].
• LinkedIn reads skills from text anywhere in the profile, including the summary, positions, title, and headline [185].
• LinkedIn does not publish weights for profile sections [183].
• Testing a query from a few accounts does not show how the profile appears to others. Search results are personal [183].
• Profile search appearances update weekly. The data is aggregated and does not show who clicked [187].
• LinkedIn says profile views are a more useful visibility measure than the order of search results [183].

• A post written in a group appears only in that group's feed [197]. You cannot share an existing post into a group [197].
• LinkedIn's systems track social actions inside groups, such as posts, likes, and comments. The data feeds recommendations in the feed and in Groups [198].
• You can find groups by name or keyword, or through groups LinkedIn recommends [194].
• Group types: listed private groups, unlisted private groups (not in search or profiles), and public groups, which are rolling out [196][199].
• Limits: a member can be in up to 100 groups. A member can post up to 15 times in one group per week, up to 20 groups per week, and up to 60 times across all groups per week. Group posts are capped at 3,000 characters. LinkedIn says these limits cannot currently be raised [200].
• Posting the same content in several groups can be flagged as spam and hidden [200].
• A group owner or manager can recommend a post. The post then appears on the group's Recommended tab [202].

• People You May Know suggests people based on shared attributes and on contacts imported from your email or phone [201].
• A profile missing current or past companies, positions, schools, or industries may not appear in these suggestions [201].
• Removing a suggestion stops the feature from recommending the most recent 500 removed suggestions [201].
• LinkedIn says it does not scan the content of messages to suggest people [201].
• Other Similar Profiles shows members with similar company, industry, or function on the right side of a profile [203].
• Joining groups, completing profile fields, and importing contacts cost nothing. No help page reviewed ties these to a fee [194][201].

• Depending on severity, LinkedIn may limit the visibility of content, label it, or remove it. LinkedIn generally sends a notice that explains the violation and the action [126].
• Repeated non-compliant posts can lead to permanent loss of features, such as posting, commenting, or messaging. Restrictions on a personal account also apply to the Pages that account manages [127].
• Automated moderation can filter some violating content within 300 milliseconds of creation. That content stays visible only to its author [129].
• LinkedIn says it blocked over 80.6 million fake accounts at registration from July to December 2024 [133].

• Promoted posts run on an ad auction. The cost depends on your bid and the audience you target [48].
• LinkedIn prices ads by campaign objective. For a website visit objective, you pay when someone clicks through [48].
• Premium Company Page costs from US$99.99 a month per license, or US$839.88 a year [54]. It covers one Company Page. Each additional page needs its own subscription [54].
• Sales Navigator Core starts at US$119.99 a month per license, or US$1,079.88 a year [45]. It is a separate subscription from Premium [45].
• None of these paid options adds organic feed reach. Paid features add amplification, follower invitations, or prospecting tools [51][54].

• Do not reshare other people's posts as your main activity [265].
• Do not use three to five hashtags for reach. Use one or two topic words if they help viewers [156][158].
• Do not join an engagement pod or use comment automation. The account can be restricted [111][132].
• Do not post the same copy to several groups [200].
• Do not post recycled thought leadership or "Comment Yes" prompts [3].
• Do not buy Premium expecting reach [50][51].

• Engagement rate on a Page is interactions divided by impressions. Interactions include clicks, reactions, comments, and shares [142].
• An impression counts when at least 50% of the content is on screen for 300 milliseconds or more, or when someone clicks it [142].
• Members reached counts distinct members and Pages who saw a post. Repeat displays do not count [142][143].
• Engagements include reactions, comments, saves, sends, and reposts [143].
• Combined post analytics shows in-network impressions (from followers and connections) and out-of-network impressions (from everyone else) [143]. Out-of-network share shows whether a post escaped your existing audience.
• LinkedIn says impression numbers are estimates and may not be precise [110][106].
• Your own views, engagements, saves, and sends count in your analytics [144].
• Demographic data is not available for content shared only with connections or in Groups [144].
• Audience analytics shows the top five results for job title, location, industry, seniority, company size, and company [105].
• You can export creator analytics as an XLSX file [105].
• Creator analytics include a date range of 7 to 365 days [110].

• Video performance analytics and video viewer demographics: 180 days [144].
• Article and newsletter analytics: 2 years [91][144].
• Discovery and social engagement counts: 1,000 days. Members reached: 400 days. Demographic breakdowns: 180 days [144].
• Save a copy of any number you need after these windows close.

• Profile appearances are times your name and headline appear across LinkedIn, in search, posts, comments, and recommendations [188].
• Profile views are visits to your profile [188].
• Free members see basic view counts. Premium members see full visitor lists and behavior data [147].
• Full profile view analytics need three or more unique viewers in the past seven days. With fewer, analytics may be limited or empty [147].

Calendly supports UTM parameters on scheduling links (utm_source, utm_medium, utm_campaign, utm_content, utm_term). You can filter bookings by tracking ID and export the list as CSV. Each UTM value can be up to 255 characters [152].

• Impressions and members reached, to see how far posts travel [142][143].
• Out-of-network share of impressions, to see whether posts reach people beyond your audience [143].
• Saves and substantive comments per post, the signals with the most weight [84][86].
• Format, post length, and posting time for each post, so you can compare like with like.
• Profile views and search appearances [147][187].
• Newsletter subscribers and edition views [91].

• LinkedIn does not publish the ranking formula or its weights. Most numbers here come from third-party studies with different samples and methods [1][4].
• No study has tested the same content on a profile and a page in the same window.
• Some sources conflict and are shown side by side: Socialinsider's 2024 document figure (5.85% in the 2025 release and 6.10% in the 2026 release) [64][63], van der Blom's video figures [69][70], and the two timing datasets [118][119].
• Van der Blom's paid report was not reviewed. Its figures are used as published.
• LinkedIn Help pages show relative update dates, so first-party dates are approximate.