LinkedIn Dwell Time and the Four Phases That Decide Your Reach in 2026
Contents
- What 360Brew actually changed
- Phase one: the score nobody sees
- Phase two: the golden hour, confirmed and unconfirmed
- Phase three: dwell time, saves, and the comment myth that needs killing
- Phase four: the late reactivation window
- How to engineer dwell time on purpose
- What earns a save
- The 72-Hour Protocol
- Five mistakes that kill distribution before it starts
- What I don't know
- What this means Monday morning
- FAQ
- Sources
I have been posting on LinkedIn since April 2004, and I have sat through at least four moments when the platform was declared to have "changed everything." This one is different, and the difference has a name: 360Brew, a foundation model that reads your post roughly the way an editor would instead of counting reactions. At the center of its scoring sits dwell time, the number of seconds a reader spends on your content before scrolling past, a metric LinkedIn's own engineers have written about publicly since 2020. Every post now moves through four distribution phases: a semantic score before anyone sees it, a small-sample test, an expansion driven by dwell time and saves, and a late reactivation window most creators don't know exists. Some of what follows is confirmed by LinkedIn itself. Some of it is practitioner data with real limits, and I will tell you which is which as I go.
In short: LinkedIn's ranking model, 360Brew, scores a post semantically before it distributes it, then tests it on a small slice of your network. LinkedIn's own engineering blog confirms dwell time, the time spent before scrolling past, as a direct ranking input; independent trackers estimate a save carries roughly five times the weight of a like, and posts that pick up saves or substantive comments 24 to 72 hours after publishing can see performance multiply four to six times. One popular claim doesn't survive scrutiny: comments do not outweigh likes fifteen to one. The current data points closer to two to one.
What 360Brew actually changed
This wasn't a tuning pass. LinkedIn replaced a patchwork of specialized ranking models, one for the feed, one for recommendations, one for matching, with a single model built to reason across all of them. The research paper, published on arXiv in January 2025 by Hamed Firooz and colleagues on LinkedIn's FAIT team, describes a 150-billion-parameter, decoder-only foundation model that reads post text, member profiles and interaction history as natural language rather than as hand-built numeric features, and applies that reading to more than 30 predictive tasks across the Feed, Job Recommendations, Search and Ads.
The practical effect shows up in reach trackers. AuthoredUp, which built its numbers on more than 600,000 posts from over 63,000 personal accounts, put average reach down 34% in 2025, with 98% of accounts in its sample seeing a decline. Botdog's write-up of the same rollout, citing an earlier AuthoredUp pull of over 3 million posts, reported a steeper 47% year-over-year drop with video hit hardest at 72%. Both numbers come from the same source measuring at different points, which tells you something on its own: the model is still moving, and any single percentage you read today has a shelf life.
What separates the accounts that lost reach from the ones that gained it, according to AuthoredUp's account-level data, is topical consistency. Creators who scattered across unrelated subjects and chased likes through generic comments lost the most. Creators with a legible area of expertise and content built to be used, not just read, gained.
Phase one: the score nobody sees
Before a single connection sees your post, 360Brew runs a scoring pass. The arXiv paper doesn't publish exact weights, and neither does any source I trust, so I'm not going to hand you invented percentages. What it does confirm is the shape of the inputs: the model reads your profile text, your recent publishing history and the post itself as language, and estimates relevance from that reading rather than from a feature table.
Two things follow from that design, and both are testable by anyone. First, an executive who posts consistently about pricing strategy and sales process builds a legible topical signature; the same executive's post about a weekend hobby gets a lower relevance estimate, not a penalty, just a lower score for that one post. Second, a post that reads as vague or purely emotional scores lower on textual quality than one with a clear point and usable information, because the model is, in the end, a language model grading language.
Phase two: the golden hour, confirmed and unconfirmed
This is where I want to slow down, because it's the part of the LinkedIn algorithm conversation that gets stated with far more confidence than the evidence supports.
Neither of LinkedIn's own engineering posts on dwell time, nor the 360Brew paper, publishes a figure for a test-audience percentage or a fixed testing window. The "golden hour," the idea that your post gets shown to a small slice of your network for 60 to 90 minutes before LinkedIn decides whether to expand it, is a pattern practitioners have reconstructed from watching impression counts climb in stair-steps after publishing, cross-checked against numbers various trackers report independently, usually somewhere between 2% and 10% of a network, converging around 5% to 8% for mid-sized profiles. It is a reasonable model. It is not a LinkedIn-confirmed one, and Pettauer's own January 2026 analysis of 360Brew goes as far as calling the "first 60 minutes" framing a leftover from the previous algorithm generation rather than a feature of the current one.
What I can tell you with more confidence: responding to comments quickly during the first hours of a post's life is consistently associated with stronger downstream reach across every tracker I checked, whatever the underlying mechanism turns out to be. That correlation has held since long before 360Brew existed. Treat the specific numbers, the percentage tested, the exact minute count, as directional, not as inputs to a formula.
Phase three: dwell time, saves, and the comment myth that needs killing
Dwell time is the one piece of this system LinkedIn has explained in its own words, twice. In May 2020, engineers Siddharth Dangi, Johnson Jia, Manas Somaiya and Ying Xuan published a post describing two measurements: on-feed dwell time, which starts counting once at least half your post is visible on screen, and post-click dwell time, the time spent after someone taps to expand it. In October 2024, a second post by Fengyu Zhang, Mohit Kothari and Birjodh Singh Tiwana described an "Auto Normalized Long Dwell" classifier that predicts, per post and per viewer, whether time spent will exceed a rolling percentile threshold, adjusted daily so the bar doesn't stay fixed as behavior shifts. The companion LiRank paper, published by LinkedIn's engineering team on arXiv in February 2024, confirms "long dwell" sits alongside like, comment, share and click as one of the actions the core feed model predicts for every post-viewer pair.
None of that is speculation. It's published, peer-reviewable engineering.
| Dwell time | What trackers report | What's actually confirmed |
|---|---|---|
| Under 3 seconds | "Limited distribution" | Falls below the Long Dwell threshold (LiRank, LinkedIn Engineering) |
| 4 to 30 seconds | "Restricted to extended" | Directional only, no published rate |
| 31 to 60 seconds | "Maximum distribution" | Directional only, no published rate |
| 60 seconds and up | Roughly 15.6% engagement rate | Richard van der Blom's Algorithm Insights Report, versus 1.2% under 3 seconds |
Saves sit next to dwell time as the strongest available signal, though here too the exact multiplier depends on who's counting. AuthoredUp's post-level analysis, cited by Botdog, puts a save at roughly five times the reach of a like and twice a comment. Pettauer's separate estimate widens that to a five-to-ten-times range. Nobody publishes an exact figure because LinkedIn hasn't, but every independent tracker I found agrees on the direction and the order of magnitude, which is about as close to consensus as this field gets. LinkedIn began surfacing save counts directly in post analytics, which at minimum tells you the platform wants creators paying attention to the metric.
Now the correction, and it's the kind of finding that makes an article worth writing rather than repeating. A widely shared claim holds that comments carry fifteen times the weight of a like. I went looking for where that number comes from and couldn't find a primary source behind it. Meet Lea's dwell time analysis, updated in May 2026, states it plainly: the effective weight of a comment "after the quality-scoring layer LinkedIn now applies is closer to ~2x a like, not 15x," referencing the same AuthoredUp dataset. Fifteen-to-one is the kind of number that sounds authoritative because it's specific, gets repeated across a dozen LinkedIn "algorithm guide" blogs without anyone checking the original, and turns out to be unsupported. If you've built a content strategy around chasing comment volume at fifteen times the value of a save, you've been optimizing for the wrong asset.
Length still matters for the comments you do get. Multiple 2026 trackers converge on 15 words as the rough line between a comment that reads as a genuine contribution and one the model discounts as low-effort, "great post" and an emoji included. I'd treat 15 as a useful mental threshold, not a hard cutoff coded somewhere in production.
Sends, a post forwarded by direct message, round out the signal set. They're the hardest of the four to fake because forwarding something privately costs the sender real social capital if the content turns out to be weak, and LinkedIn added save and send counts to creator analytics dashboards in late 2025, effectively telling every creator what it now measures.
Phase four: the late reactivation window
Most creators treat a post as finished after 24 hours. Under 360Brew, that assumption costs reach.
AuthoredUp's own language on this is unambiguous: "LinkedIn is no longer a 24-hour content platform." The company reports that posts sparking genuine conversation stay in circulation for two to three weeks rather than one day. Botdog's write-up of the same shift is more specific about the mechanism: posts that pick up saves and substantive comments in the 24-to-72-hour window after publishing can see performance multiply four to six times relative to their initial trajectory, because the model reads delayed engagement as a signal of durable value rather than novelty-driven noise. One case cited in that same write-up describes a post crossing 100,000 views after a save spike at the 72-hour mark, three days after most creators had stopped checking it.
The mechanism is coherent with everything else 360Brew rewards. A like in the first ten minutes says "I saw this." A save or a thoughtful comment on day three says "this was still worth my time after the scroll had moved on," and that is precisely the durability signal the model is built to detect.
How to engineer dwell time on purpose
Dwell time isn't luck. It's built into how a post is structured.
The first three lines, the ones visible before "see more," decide whether a reader stays. An opening that names a concrete stake and withholds one piece of information, rather than one that summarizes the whole post, gives the reader a reason to keep reading rather than a reason to stop.
Structure the body so each paragraph adds something the last one didn't have. A post that hands over its full value in the opening two paragraphs gives the reader no reason to finish it, and the model is measuring whether they finish it.
Format carries real weight too. Document carousels average 15 to 20 seconds of dwell time across the trackers I checked, against 8 to 10 seconds for a single text or image post. That's a meaningful gap, though the exact seconds swing by niche and by tracker, so treat the ratio, not the specific numbers, as the takeaway. For text posts, line breaks and short paragraphs slow the eye down; a dense wall of text scrolls, and gets scrolled past, faster.
What earns a save
A save is a promise to come back. Three content types reliably earn it.
Usable content: templates, checklists, scripts, formulas, anything a professional can apply directly rather than merely appreciate. "Seven cold-outreach openers that worked in Q2, with the actual message" gets saved. A general reflection on the state of outreach doesn't.
Reference data: sector benchmarks, comparisons, numbers a reader will want to pull up again in a meeting six weeks from now.
Structured frameworks: decision matrices, step-by-step processes, anything the reader recognizes they'll reuse rather than just read once.
The 72-Hour Protocol
Here's how I structure the three days after publishing, mapped to the four phases above.
Day 0, hours 0 to 2. Confirm your profile headline and About section still match what you're about to post, since alignment is part of the semantic score. Publish when your specific audience is actually online; Buffer's July 2026 analysis of 4.8 million posts found Wednesday through Friday afternoons, 3 p.m. to 5 p.m., now outperform the traditional morning slot, a genuine shift from 2025's working-hours pattern, and worth testing against your own analytics rather than assuming. Answer every comment within 15 minutes, in more than a few words.
Day 0, hours 6 to 12. Answer stragglers with substance. Prioritize replies to commenters whose profile matches your actual buyer.
Day 1, hours 12 to 18. Add a comment of your own with new information, a data point you left out, a clarification. It renotifies everyone who engaged the day before.
Day 1, hours 18 to 36. Send the post by direct message to five to ten people for whom it has real, specific use, not a mass blast. Keep answering comments as they come in.
Day 2, hours 36 to 48. Post a second self-comment: what the responses so far have confirmed, or a data point you're adding based on the conversation. This is squarely inside the reactivation window.
Day 2 to 3, hours 48 to 72. Engage with the posts of the people who commented on yours; reciprocity strengthens the connection the model reads between your two accounts. Answer the last wave of comments. Note the recurring questions, they're next week's post.
| Window | Priority action | Signal targeted |
|---|---|---|
| H0-H2 | Fast, substantive replies | Early engagement velocity |
| H6-H12 | Depth for late commenters | Conversation depth |
| H12-H18 | Self-comment with new information | Renotification, dwell time |
| H18-H36 | Targeted DM sends (5-10 people) | Send signal |
| H36-H48 | Second self-comment | Late reactivation (4-6x) |
| H48-H72 | Reciprocal engagement, last replies | Extended distribution queue |
Five mistakes that kill distribution before it starts
Posting and vanishing. No replies during the early window reads as a broadcast, not a conversation, and the model scores it accordingly.
Farming generic comments. "Great post!" and a clapping emoji no longer move the needle, and a high ratio of generic-to-substantive comments can read as manufactured engagement rather than help it.
Publishing off-topic. Bouncing between sales strategy, vacation photos and motivational quotes flattens your topical signature, and 360Brew has less to go on when deciding who should see the next post.
Posting daily. A new post entering its test window pulls distribution attention away from the one before it. Several 2026 posting-frequency guides converge on two to four posts a week rather than seven.
Treating saves and likes as interchangeable. They aren't measuring the same thing, and content built to be liked, quotable one-liners, isn't the same content that gets saved.
What I don't know
I don't know the exact weight LinkedIn assigns to a save, a send or a qualified comment, because LinkedIn has never published one. Every multiplier in this piece, five times for saves, two times for comments, four to six times for late reactivation, comes from third-party analysis of observed outcomes, not from LinkedIn's own engineering documentation. Treat the direction as solid and the specific number as an estimate with a margin I can't quantify.
I don't know whether "golden hour" is a real, fixed mechanism inside 360Brew or a name the community kept from the previous algorithm generation and retrofitted onto new behavior. The pattern, fast early engagement correlating with stronger reach, is consistent across trackers. The specific window and percentage are not confirmed anywhere I could find.
I don't know why AuthoredUp's own published reach-decline figure moved from 47% to 34% between the version other sites cite and the version live on their blog as of this writing. It could be a later, larger, more representative sample. It could be the market partially recovering. Nobody has published that explanation, so I'm not going to invent one.
What this means Monday morning
None of this changes the fundamentals that mattered before 360Brew existed. A post built around one number nobody else has, a specific outcome you can date and place, a position you're willing to defend, still outperforms a technically well-timed post with nothing in it. The four-phase mechanic changes how fast that difference compounds, not whether it exists.
FAQ
What is dwell time on LinkedIn?
Dwell time is the amount of time a member spends viewing your post before scrolling past it, measured both while it's visible in the feed and after a click to expand it. LinkedIn's engineering team has confirmed it as a direct ranking input since a 2020 blog post, refined further in a 2024 follow-up describing a daily-adjusted "Long Dwell" classifier used across the feed ranking model.
How much does a LinkedIn save weigh compared to a like?
Independent trackers, not LinkedIn itself, estimate a save carries roughly five times the reach weight of a like and about twice that of a comment, based on AuthoredUp's analysis of millions of posts. Other trackers put the range as wide as five to ten times a like. No official multiplier has been published.
Is it true that comments are worth 15 times more than likes on LinkedIn?
No. That figure circulates widely across LinkedIn algorithm guides, but Meet Lea's May 2026 analysis, drawing on the same AuthoredUp dataset commonly cited for it, found the effective weight closer to twice a like, not fifteen times. Treat any specific multiplier you read, including the ones in this piece, as a practitioner estimate rather than a LinkedIn-published fact.
What is the LinkedIn golden hour?
The golden hour describes the idea that LinkedIn tests a new post on a small slice of your network, commonly estimated at 5% to 8%, during the first 60 to 90 minutes, before deciding how far to expand its distribution. LinkedIn has never published a figure for the test percentage or window length; the concept is reconstructed from practitioner observation of impression patterns, and some analysts now treat it as a leftover framing from the pre-360Brew algorithm.
Does engagement received days after publishing still help a LinkedIn post?
Yes, more than most creators assume. AuthoredUp reports that posts receiving saves and substantive comments 24 to 72 hours after publishing can see performance multiply four to six times relative to their initial trajectory, and that posts sparking real conversation now stay in active circulation for two to three weeks rather than one day.
How often should you post on LinkedIn in 2026 to maximize reach?
Most 2026 tracking guides converge on two to four posts a week rather than daily posting. Publishing every day pulls distribution attention away from the previous post as each new one enters its own test window, which works against the extended, multi-day reach that 360Brew now rewards for posts that hold attention.
Sources
- Hamed Firooz et al., "360Brew: A Decoder-only Foundation Model for Personalized Ranking and Recommendation", arXiv:2501.16450, LinkedIn FAIT team, January 2025.
- Siddharth Dangi, Johnson Jia, Manas Somaiya, Ying Xuan, "Understanding dwell time to improve LinkedIn feed ranking", LinkedIn Engineering Blog, May 12, 2020.
- Fengyu Zhang, Mohit Kothari, Birjodh Singh Tiwana, "Leveraging Dwell Time to Improve Member Experiences on the LinkedIn Feed", LinkedIn Engineering Blog, October 1, 2024.
- "LiRank: Industrial Large Scale Ranking Models at LinkedIn", arXiv:2402.06859, LinkedIn engineering team, February 2024.
- AuthoredUp, "How the LinkedIn Algorithm Works", authoredup.com, accessed August 2026.
- Botdog, "5 Biggest LinkedIn Algorithm Changes In 2026", botdog.co, March 16, 2026.
- Meet Lea, "LinkedIn Dwell Time: The Hidden Metric That Controls Visibility", meet-lea.com, updated May 17, 2026.
- Pettauer, "LinkedIn 360Brew and the New Physics of Visibility", pettauer.net, January 26, 2026.
- Richard van der Blom, Algorithm Insights Report 2025/2026, richardvanderblom.com, based on an analysis of over 1.8 million posts.
- Buffer, "Best Time to Post on LinkedIn in 2026: 4.8M Posts Analyzed", buffer.com, July 22, 2026.
All In: reading the machine correctly is table stakes, not the whole job
Understanding what 360Brew actually rewards keeps you from optimizing for a number that turned out to be wrong. It doesn't replace having something worth saying in the first place.
All In is the B2B media that decodes LinkedIn, expert blog, weekly podcast and newsletter for SME leaders and sales directors who want to turn LinkedIn into measurable growth. An original creation by Patrick de Carvalho, on LinkedIn since 2004. Motto: "I Never Lose."
Discover All In and get what matters on LinkedIn, every week.