What goes viral · Part 3 of 4
LinkedIn told us what its feed counts. Reading is on the list.
LinkedIn's own engineers wrote down how the feed picks posts. We read it. Time spent reading counts like a comment, a post nobody touches counts against you, and two habits really do help.
September 2026 · 7 min read
More engagement when you reply
About 30% better than your usual when you reply in your own comments. Buffer measured it on 2 million posts.
Of profiles do better when they reply
The highest share on any platform Buffer looked at.
More reactions per person tagged
Name the person your post is about and link them.
Per hashtag
Less than a tag, but they stack. Three hashtags beat one tag.
01 · The scorecard
Reading a post counts like a comment
On 12 March 2026, LinkedIn's engineers explained the ranker, the software that decides which posts you see and in what order. A paper from LinkedIn Corp., dated 16 October 2025, says what it is trying to get more of.
The headline is real: the feed scores the time a reader spends on your post, right next to likes, comments and shares. The details knock down most of the advice built on that headline.
Dwell is how long someone stays on your post before scrolling past. The ranker guesses what each reader will do with each post, in two kinds: “passive tasks (click, skip, long-dwell)” and “active tasks (like, comment, share)”. A long dwell gets its own guess, just like a comment does.
The paper says what the ranker is aiming at: more people each day who do something with a post. It lists what counts: “(e.g., long dwell, react, comment, repost, etc.)”. Any one of those is a win.
A win
any one of these
Against you
shown, then nothing
Reading is scored, not just clicking. A post that holds attention counts even with no likes under it.
02 · The stopwatch
What you stop on decides what you see next
The feed keeps a record of every member, LinkedIn's word for a user: more than a thousand past actions per person.
LinkedIn describes the record as “a chronologically ordered sequence of posts they’ve previously engaged with”, where “your actions (long dwells, likes, comments, shares) are embedded and interleaved with post representations”. In plain words: the posts you did something with, and what you did to each.
So dwell is also part of how the feed describes you. Stop on a post and the feed writes that down as something you like, whether you meant it or not.
They stayed. No like, no comment. It still scores.
They didn’t stop. The feed learns you two don’t match.
The second phone is the uncomfortable half. A post shown to someone who does nothing is used as a hard negative: the feed learns that this post and this reader don’t match. Reaching the wrong people isn’t free.
03 · The limits
What the dwell finding does not prove
People use “LinkedIn rewards dwell time” as a reason to write longer posts, tease the ending and chop text into one-line paragraphs. Nothing LinkedIn published backs that up. Three things get in the way.
Nobody knows the cut-off
A long dwell is a yes or a no. LinkedIn has never said where the line is.
It's guessed per reader
Dwell is guessed for each reader on each post, not a score you build up over time.
The clock starts at half a post
Timing begins when at least half your post is on screen. That detail comes from a 2020 engineering post.
One more. LinkedIn never says how much a long dwell is worth next to a comment, and nothing published links a post’s length to how long people dwell on it. “Dwell matters” is backed up. “Dwell beats comments” is not.
Knowing how the feed works is not the same as knowing what to post.
04 · The first lever
Reply in your own comments
Two habits survived our checks. This is the stronger one. Buffer, a social media tool, looked at 2 million posts from more than 220,000 accounts, comparing each account's posts with its own usual results, adjusted for follower count and subject.
Engagement lift when the author replies in the comments
Buffer · 2M posts · 220,000+ accounts · each account compared with itself · to Dec 2025
LinkedIn is second only to Threads. The boost is roughly three to four times what the same habit is worth on X (+8%) or Facebook (+9.5%). The 83% figure is a different measure: how many profiles it works for, not how much.
Comparing each account with itself answers the usual complaint about social media numbers: that the accounts doing the thing were just bigger to begin with.
The catch Buffer points out itself. Posts that are already doing well get more comments, and people reply to those. So read 30% as the most the effect could be, a pattern rather than a proven cause. It is still the cheapest item on any campaign checklist.
05 · The second lever
Tag a person. A hashtag does less.
Tagging a person means naming them in the post so they're linked and notified. Usera and Durham studied 991 posts. Their paper came out in Business and Professional Communication Quarterly on 16 May 2025, and it's the only peer-reviewed source here.
Each person tagged is worth about 15% more reactions, the one-click responses under a post. Each hashtag is worth about 6%.
That gap isn’t the end of the story. Both numbers are per tag, and hashtags stack.
One person tagged
per tag · 1.15
1.15
about 15% more
Three hashtags
per hashtag · 1.06
1.19
three multiplied
The paper doesn’t check whether the tag number and the hashtag number are truly different from each other, so treat the order as likely, not proven.
Two more limits. Every post came from one US city area, and the top 1% of posts by reactions were taken out before measuring, so this is good evidence about the ordinary post and no evidence about the biggest ones.
Both posts below name the person the author worked with, exactly the kind of tag the study measured.
Arfa Farheen
arfa-farheen · 60,794 followers · 25 Apr 2025
Most people write LinkedIn posts the wrong way. That's why they never blow up.
So Norbert Bezzina and I decided to fix that. We broke down the exact formula.…
Leonard Rodman
leonardrodman · 59,550 followers · 30 Apr 2024
There are 1,000,000+ remote jobs! by my friend Adam Biddlecombe
Here are 9 websites to find your next role today. 👇
1. FlexJobs Hand-screened, scam-free remote job listings.…
06 · What to do
Four simple rules
- 1
Write to be read, not just clicked.
A long read scores like a comment, even when the likes are few.
- 2
Reply in your own comments. Every time.
About 30% more engagement, and it costs no ad money.
- 3
Tag the person the post is about.
Not a wall of strangers, and hashtags stack if you add them.
- 4
Aim at the right people, not the most people.
A small, well-matched audience beats a big, wrong one.
And one thing not to do: quote the dwell finding as a reason to write longer. Nobody has measured whether longer posts get more reading time.
Next in the series: engagement rate shrinks by 4.6 times as accounts grow, and what that does to pricing.
Sources · show
- LinkedIn Engineering, Engineering the next generation of LinkedIn's feed, 12 March 2026. Passive and active tasks, the shared Mixture-of-Experts model, the member history of more than a thousand interactions, and impressions without engagement as hard negatives.
- LinkedIn Corp., arXiv 2510.14223, 16 October 2025. “Professional Interactors” and the training label, a single OR across long dwell, react, comment and repost. Neither the long-dwell threshold nor its weight against a comment is published.
- Dangi et al., LinkedIn Engineering, Understanding dwell time to improve LinkedIn feed ranking, 12 May 2020. The half-visible detail only.
- Buffer, State of Social Media Engagement 2026. Reply sub-analysis: 2M posts, 220,000+ accounts, to December 2025, each account compared with itself, adjusted for follower count and subject. Direction of cause not established. The 83% figure is a separate measure from the 30% lift.
- Usera & Durham, “What Predicts Engagement on LinkedIn?”, Business and Professional Communication Quarterly, 16 May 2025, n=991. Negative binomial regression; effects per tag and per hashtag, not tested against each other. Single US metropolitan area; top 1% of posts by reactions removed as outliers.
- Post cards from the our production database, queried 2 September 2026; the records do not store whether a named person was tagged. Full report: What actually goes viral on LinkedIn, linklabs.wiki, September 2026.
- Photographs: Pexels, free licence. The two posts and the video's cover frame are public LinkedIn posts, linked above.