All In
FR EN
Back to blog
Profil & Publications 18 min read

LinkedIn hooks that stop the scroll: four openings and why they work

| By Patrick de Carvalho

Contents

A LinkedIn hook is the text a reader sees before the "see more" link, roughly 140 characters in the mobile app, not 210, which is the desktop figure most hook guides quote without saying which device they mean. I have posted here since April 2004, the platform's first year, and written first lines that died in that gap. What follows is four openings, each with the mechanism behind it and where it stops working. I drafted fifteen originally and kept the four with the clearest mechanism, dropping the one every guide recommends because a 1.2 million-post dataset says it underperforms.

In short: LinkedIn truncates by rendered lines, not characters, about 140 in the mobile app and 210 on desktop. Write for 140. Question openers earn 19 median likes against 29 for non-questions across 1,179,958 posts (MagicPost, June 2026), and the "open loop" everyone credits to the Zeigarnik effect rests on a different effect: a 2025 meta-analysis of 59 publications put the Zeigarnik memory advantage near zero.

What actually gets cut, and where the line falls in English

The truncation point is a line count, not characters. John Espirian documented the break points in January 2020: desktop allows five rendered lines on a text-only post, three with an image, document, or link preview, mobile allows two, everything after sits behind "see more." The budget shifts with what you type: AuthoredUp's 2026 reference puts the practical cutoff at approximately 140 characters on mobile, 210 on desktop, and two consecutive line breaks end the snippet early, so a hook stacked as three fragments can truncate at 60 characters.

Write for 140 and your hook survives on both devices; write to 210 and about a third disappears on the phone, where most readers are. English also has a small edge: it averages about 4.5 letters per word against roughly 4.84 for French, close to one extra word for free inside the same 140 characters.

The attention figure I could not verify

Here is where I break with almost every article on this subject, including my own French draft: an effective attention window of 7.8 seconds, measured across 2.1 million adult internet users, credited to a joint Pew Research Center and Digital Wellness Foundation report. I went looking: it is not on pewresearch.org, and the actual Digital Wellness Lab, at Boston Children's Hospital, has published nothing of the kind. The only trace leads back to a statistics aggregation site, so I dropped it, along with the eight-second attention span claim (shorter than a goldfish's), traced by BBC journalist Simon Maybin in March 2017 to a site called Statistic Brain, which produced no research when asked; Microsoft quietly pulled the 2015 marketing report that had carried the figure everywhere.

What does exist is Gloria Mark's work at UC Irvine: average attention on any single screen is now 47 seconds, down from about two and a half minutes in 2004, though that counts screen-switching and says nothing about whether a post gets read. Nobody has published a credible measurement of how long a member looks at one feed item before expanding it. LinkedIn holds that data and has never released it, and every guide with a precise number is repeating something it did not check, the same failure I took apart in the All In dossier on the AI slop button.

The mechanism most hook guides name is the wrong one

Open any article about LinkedIn hooks and you will find "open loop" followed by a confident reference to the Zeigarnik effect, the 1927 finding that interrupted tasks are remembered better than completed ones, and the theoretical spine of the hook industry. It does not hold up. In July 2025, Romain Ghibellini and Beat Meier published a meta-analysis in Humanities and Social Sciences Communications covering 59 publications, 38 on the Zeigarnik effect: the weighted recall ratio between interrupted and completed tasks came to 0.99, meaning interrupted tasks are remembered about as well as finished ones (dz = 0.15 across the eight publications reporting an effect size, small enough to be noise). The same meta-analysis found something that matters more for anyone writing a first line: the Ovsiankina effect, whether people go back and finish an interrupted task, came in at a 67.00% resumption rate across 21 publications, well above chance.

"The Ovsiankina effect represents a general tendency, whereas the Zeigarnik effect lacks universal validity." Ghibellini and Meier, Humanities and Social Sciences Communications, July 2025

Your reader won't remember your unfinished story better, but will feel a pull to finish it. Memory isn't the lever, resumption is, giving a hook one job: leave a task visibly open, and make the reader believe it closes with a single click.

The second mechanism, Loewenstein's information gap theory (Psychological Bulletin, 1994), says curiosity arises when attention lands on a gap between what you know and want to know, scaling with how close the answer feels. That is why "I learned something surprising last week" produces nothing while "our churn dropped 40% the quarter we stopped doing quarterly business reviews" produces a click: the first names no gap, the second puts the answer one paragraph away.

Four openings, the mechanism behind each, and where each one fails

I originally drafted fifteen; some collapsed into repeats of the same mechanism wearing different clothes (the list hook, the process reveal, and "here is what I wish I'd known" are cost-benefit arithmetic or the admission with different furniture). Eight survived that first pass; below are the four with the clearest mechanism and real data where it exists.

1. The number that should not be true

Mechanism: prediction error. Your reader carries an internal model of a normal business number; a violation demands an explanation behind the link.

  • "Sixty-eight percent of our pipeline last quarter came from a channel we shut down in 2024."

Where it fails: round numbers and borrowed numbers. "About 70%" reads as a guess, and "Studies show 73% of B2B buyers" hands the surprise off. Data: across 278,996 posts with a number in the first line, MagicPost recorded 35 median likes against 26 without.

2. The admission

Mechanism: costly signaling. A statement that damages your own position is expensive to fake, so readers price it as credible, the oldest trust mechanism there is.

  • "We lost our largest account in March. I found out from their receptionist."

Where it fails: the humblebrag, which costs nothing and signals nothing ("My biggest flaw is caring too much about my team"), and repetition: the third confession in a month reads as a content format, not candor.

3. The one-sentence scene

Mechanism: Ovsiankina resumption. An interrupted narrative creates a task the reader wants to close, a pull the meta-analysis puts at roughly two thirds of people.

  • "Tuesday, 4:40 p.m., our biggest customer's CFO called my cell phone. He had never called my cell phone."

Where it fails: invention. Readers recognize the genre instantly ("yesterday a young man approached me outside a Starbucks"). Time stamps and a pointless detail separate a scene you lived from one you assembled.

4. The belief reversal

Mechanism: Loewenstein's information gap, aimed at a belief the reader already holds: naming it makes the gap personal, their knowledge on the line.

  • "'Post daily or the algorithm forgets you.' I posted eleven times in the first quarter and it was my best quarter here."

Where it fails: the straw man. If nobody actually holds the belief you are demolishing, the hook reads as posturing. Related: an "Unpopular opinion:" prefix attached to a popular opinion is now common enough to work as a warning label.

The hook I dropped, and the numbers that made me drop it

The question opener: every guide recommends it, my French draft had it at number fifteen, and the data says it is the weakest opening on the platform.

MagicPost analyzed 1,179,958 LinkedIn posts over twelve months, pulled in June 2026, excluding reshares and deleted posts: question first lines earn 19 median likes against 29, a 34% gap, across 143,789 question posts and 1,036,169 others. The standard defense, that questions trade likes for comments, doesn't hold either: median comments came in at 4 against 6, and the penalty held across three follower bands (under 5,000, 5,000-50,000, above 50,000), ruling out that only small accounts open with questions.

Caveat: median likes is a weak proxy for distribution, and MagicPost sells a LinkedIn tool it competes in. Even so, a question hands the reader an easy exit: the loop closes in their head, and the pull the Ovsiankina effect would have created stops existing. A statement leaves it open.

The same reasoning cut the rest of the original fifteen: the dated-trend opener ("In 2026, the algorithm rewards...") names no mechanism, the list hook and process reveal are cost-benefit arithmetic with a different first word, and "here is what I wish I had known" is the admission, softened.

Which opening to use, and when

Opening Mechanism Best used for Breaks when Evidence status
The number that should not be true Prediction error Proprietary data, results posts The number is rounded or borrowed MagicPost: 35 vs 26 median likes for number-led first lines
The admission Costly signaling Lessons learned, strategy reversals It costs you nothing to say No LinkedIn-specific data
The one-sentence scene Ovsiankina resumption Client stories, turning points The scene is invented Meta-analysis: 67% resumption rate, 21 publications
The belief reversal Information gap Positioning, contrarian analysis The belief is a straw man Loewenstein 1994
The question None that survives contact Nothing, on this platform Always MagicPost: 19 vs 29 median likes, 34% penalty

One rule matters more than the table: match the mechanism to what you have. Lead with a number if you have one; if all you have is an opinion, the belief reversal is the one that won't expose you, provided the belief you are naming is one people actually hold.

The three openings I will not write

Manipulative hooks work, which is precisely the problem: they produce clicks and a cost that doesn't show up in the analytics for months. Kaushal and Vemuri measured it in IEEE Transactions on Technology and Society (2021): clickbait headlines significantly reduced perceived credibility, and the reader quietly downgrades you after clicking. On LinkedIn there is a second, mechanical cost: dwell time has been a ranking objective since May 2020, and a hook that overpromises produces an expand followed by an immediate exit, the exact signal the ranking model was built to punish.

Three formats I refuse.

The fabricated dialogue. "A CEO said to me yesterday: 'Patrick, we have tried everything.'" If it didn't happen, in those words, on a day I can name, I don't write it.

The tragedy-to-pitch. Opening on a layoff, a diagnosis, or a bereavement, then landing on a service offer. Ask whether you'd say the same sentence to the person the story is about.

"Agree?" Along with "Thoughts?" and every three-word comment solicitation, a request for a favor dressed as engagement that produces the shallow comments the ranking model discounts.

If you need one post to perform, manipulate the opening. If you need a reader still here in three years, do not. Our own rules on sourcing and disclosure are public in the All In editorial disclosure.

Build your own bank Monday morning, with no tool

The four openings above are worth nothing as templates; copied literally, they produce the flattened prose the whole platform is currently arguing about. What follows takes about forty minutes and a text file.

Pull twelve real things from the last ninety days: a number you own, a bad call, a price you changed, a hire you got wrong, a customer sentence you still remember word for word. The useful material starts around number eight.

Run each one through the mechanisms above. Take "we lost our biggest account in March," write it as an admission, then as a one-sentence scene; you'll feel which one it wants to be.

Measure what you can actually see: LinkedIn exposes neither dwell time nor "see more" clicks, only impressions, members reached, and comments. Sort comments into two piles, people repeating your point back and people adding something new; the second predicts outcomes and appears in no dashboard (more on what is measurable).

Test in pairs: two posts, same substance, different openings, eight to ten days apart. Two data points prove nothing, but after twenty posts the pattern shows up.

For context: Socialinsider's 2026 benchmarks (1.3 million posts, 16,645 business pages) put average LinkedIn engagement at 5.20%, up 8% year over year, native documents leading at 7.00%, page-level figures, not a goal for your account.

What I don't know

The honesty of an article like this is judged here rather than in the hook list.

I do not know the click-through rate on the "see more" link, for any format, and neither does anyone outside LinkedIn. I do not know how heavily the first line weighs in the ranking model relative to everything else in the post. LinkedIn described 360Brew, the large language model it built for ranking, in a paper submitted to arXiv on January 27, 2025 by Hamed Firooz and 22 co-authors, and that paper was later removed by arXiv administrators because the submitter did not hold the rights to agree to the license at submission. The most-cited technical source on the current LinkedIn algorithm is a withdrawn preprint. Anyone quoting it with confidence, myself included, should say so.

I do not know whether the question-hook penalty is causal. It could be that questions attract weaker posts rather than weakening posts. The follower-band control reduces that risk without eliminating it.

I have no measurement of how long these mechanisms stay valid. Costly signaling and selective attention are properties of human beings and will outlive the platform. The 140-character mobile cutoff is a product decision that could change next quarter. If it moves, the number in this article expires, and you will be able to work out the new one from the line count. It will be covered here and in the weekly All In newsletter when it does.

The part that has not changed since 2004

I joined this platform in its first year, when the feed did not exist and a member could only list where they had worked. Every ranking change since has rewarded the same thing under a different name: something a specific person could not have gotten anywhere else. The hook is the shortest statement of why the post exists. When people tell me their openings are not working, the opening is almost never the problem: they are trying to write a compelling first line about something they weren't especially interested in saying.

If you have the number, the mistake, or the sentence a customer actually said, the four mechanisms above will carry it. If not, no opening will save the post, and the forty minutes you were about to spend rewriting the first line are better spent finding something worth opening with.

FAQ

How many characters can you write before "see more" on LinkedIn?

Roughly 140 characters in the mobile app and roughly 210 on desktop. The underlying rule is a line count: two rendered lines on mobile, five on desktop for text-only posts, three on desktop when an image, document, or link preview is attached. Consecutive line breaks cut the snippet short. Write to 140 and your opening survives on every device.

Do question hooks work on LinkedIn?

The largest public dataset says no. Across 1,179,958 posts analyzed by MagicPost in June 2026, question first lines earned 19 median likes against 29 for non-questions, a 34% gap, with median comments also lower at 4 against 6. The penalty held across three follower bands. Caveat: median likes is a proxy for distribution rather than a measure of it, and the publisher sells a LinkedIn tool.

Is the Zeigarnik effect a real basis for "open loop" hooks?

Not in the form usually cited. A 2025 meta-analysis by Ghibellini and Meier covering 59 publications found a weighted recall ratio of 0.99 for interrupted versus completed tasks, meaning no memory advantage. The related Ovsiankina effect, measuring the urge to resume an interrupted task, held at a 67% resumption rate. Open loops create a pull to come back and finish. The memory claim is the part that failed to replicate.

How long is the average attention span on a social feed?

Nobody has published a credible figure for LinkedIn specifically. The widely quoted eight-second attention span was traced by the BBC in 2017 to a statistics website with no research behind it, and Microsoft removed the report that popularized it. Gloria Mark's research at UC Irvine measures 47 seconds of attention on any single screen before switching, which counts screen-switching and says nothing about whether a post gets read.

Should the hook be one line or two?

One, when the material allows it. Across 1,179,958 posts, first lines of one to five words earned 30 median likes and six to ten words earned 29, against 25 for eleven to fifteen words. Those differences are small. Length matters far less than whether the line leaves a specific question open that the post then answers.

Does a strong hook alone increase reach?

No. Dwell time is a documented ranking objective at LinkedIn, measured from the moment at least half of your post is visible on screen. A hook that produces an expand followed by an immediate exit sends a worse signal than a modest opening followed by a genuine read. The opening buys you the expand; the body decides what happens next.

How do I avoid sounding like every other LinkedIn post?

Lead with material only you have: an exact number from your own operations, a dated event with a place attached, a customer sentence you can quote verbatim. The mechanisms here are shapes to pour your own material into. Applied to generic material they produce generic posts, the failure mode of every hook template list, this one included.

Sources

  1. AuthoredUp, LinkedIn character limits and post length data, 2026: approximately 140 characters mobile and 210 desktop before "see more"; analysis of 372,126 posts, September 2025 to February 2026.
  2. John Espirian, "LinkedIn 'see more' break points", January 13, 2020: the cutoff is governed by rendered lines rather than characters.
  3. MagicPost, LinkedIn hooks analysis, June 2026: 1,179,958 posts over twelve months, reshares and deleted posts excluded, medians, results controlled across three follower bands. Question openers 19 median likes against 29; number-led openers 35 against 26.
  4. Ghibellini, R. and Meier, B., "Interruption, recall and resumption: a meta-analysis of the Zeigarnik and Ovsiankina effects", Humanities and Social Sciences Communications, vol. 12, article 962, July 1, 2025: 59 publications; Zeigarnik weighted recall ratio 0.99, dz = 0.15; Ovsiankina resumption rate 67.00%.
  5. Loewenstein, G., "The Psychology of Curiosity: A Review and Reinterpretation", Psychological Bulletin, vol. 116, no. 1, pp. 75-98, 1994: information gap theory of curiosity.
  6. LinkedIn Engineering, Siddharth Dangi, Johnson Jia, Manas Somaiya and Ying Xuan, "Understanding dwell time to improve LinkedIn feed ranking", May 12, 2020: definitions of "on the feed" and "after the click" dwell time, and the ranking gains from adding dwell features.
  7. Kaushal, V. and Vemuri, K., "Clickbait: Trust and Credibility of Digital News", IEEE Transactions on Technology and Society, vol. 2, no. 3, pp. 146-154, 2021: clickbait headlines significantly reduce the perceived credibility of the items they introduce.
  8. Simon Maybin, "Busting the attention span myth", BBC News, March 10, 2017: the eight-second figure traced to a statistics aggregation site with no underlying research.
  9. Gloria Mark, University of California, Irvine, "Attention Span" research: average attention on any single screen down to 47 seconds from about 2.5 minutes in 2004.
  10. Socialinsider, LinkedIn organic benchmarks 2026: 1.3 million posts across 16,645 business pages, January 2024 to December 2025; 5.20% average engagement rate, up 8% year over year; native documents at 7.00%.
  11. Firooz, H. et al., "360Brew: A Decoder-only Foundation Model for Personalized Ranking and Recommendation", arXiv 2501.16450, submitted January 27, 2025: 150-billion-parameter model covering more than 30 predictive tasks. Removed by arXiv administrators over a licensing issue.
  12. Pew Research Center: searched for the "7.8 second attention window" study attributed to a Pew and Digital Wellness Foundation collaboration. No such publication found.

All In: the first line is the shortest version of why the post exists

Openings age fast and mechanisms do not. What holds attention on a feed in 2026 is what held it in 2010: something specific enough that only one person could have written it.

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.