Most LinkedIn advice is someone's opinion dressed as a rule. We wanted numbers, so we took a public dataset of 34,012 real LinkedIn posts, filtered it to 12,988 posts from 65 creators, ranked every one by engagement rate, and compared the top 10% against the bottom half. The top posts open in first person about twice as often, talk directly to the reader, ask more real questions, and stay native instead of linking away. Three rules that get repeated constantly, including “keep your hook short” and “always lead with a statistic”, showed no measurable effect at all.
The short version
- Top posts open in first person 19.6% of the time vs 10.1% for the bottom half.
- Nearly half of top posts say “you” in the visible hook (47.5% vs 32.7%).
- Native video is 2.6x more common in top posts; shared-link posts skew to the bottom.
- Hashtag stuffing (4+) appears in 41.9% of weak posts vs 25.9% of top posts.
- Hook length is a myth: the median is 206 characters in both groups.
- Comments are the real separator: 72 at the top-decile median vs 5 at the bottom.
How we ran the study
The raw material is a public Kaggle dataset of 34,012 posts from LinkedIn influencer accounts, captured with reaction counts, comment counts, media type, hashtags, and author follower counts. We filtered to English posts from authors with at least 1,000 followers and at least 25 reactions, which left 12,988 posts from 65 creators. Then we scored each post:
engagement rate = (reactions + 4 × comments) ÷ author followers
Comments get four times the weight of reactions for three reasons. They are much scarcer, so they discriminate better between good and great. They take real effort, so they signal genuine interest rather than a reflex tap. And they push a post into the commenter's own network, which is how a post escapes the audience you already have. Dividing by followers lets a 3,000-follower account be compared fairly against a 300,000-follower one.
We then split the ranked list into a top decile of 1,298 posts and a bottom half of 6,494 posts, and compared them on everything measurable in the visible text and the post metadata.
One quirk of the data that turned out to be a feature
The dataset captures each post's text up to LinkedIn's “see more” fold, roughly the first 210 characters. That is a limitation if you want to study full post bodies. It is a gift if you want to study what makes people stop scrolling, because the pre-fold fragment is exactly what a reader judges before deciding to engage. Every text finding below describes the part of the post that actually gets judged.
What this study cannot tell you
Being straight about the limits matters more than the findings themselves, because most pages in this niche cite hard-sounding numbers with no methodology at all.
- It is correlational. These are patterns that travel with high engagement, not proven causes of it.
- The cohort skews established. 65 creators, all above 1,000 followers. Findings are directional for a brand-new account.
- Engagement counts are lifetime totalsat the time of collection, and post ages vary, which adds noise to any individual post's score.
- Author effects are real. Some accounts simply have more engaged audiences. Where a finding could be explained by a handful of accounts, we say so.
What separates the top 10% of LinkedIn posts?
Here is the full comparison. Every row is the percentage of posts in that group showing the pattern.
Hook patterns: top decile vs bottom half (% of posts)
Opens in first person (I, my, we)
Says 'you' anywhere in the hook
Contains a question
Opens with a question word
Contains an emoji
Contains a number
Finding 1: the first 210 characters are the whole game
LinkedIn truncates every post in the feed at roughly 210 characters on desktop, and less on mobile. Readers decide from that fragment alone whether to tap. Every single pattern that separates top posts from weak ones in this study lives inside that fragment. Nothing below the fold can rescue a hook that failed, because most people never see it. This is why we wrote a separate guide on how to start a LinkedIn post and a deeper one on writing LinkedIn hooks.
Finding 2: top posts open with a person, not a topic
19.6% of top-decile posts start with I, my, or we, against 10.1% of the bottom half. That near-doubling is the cleanest split in the study. Weak posts open with observations about an industry. Strong posts open with something that happened to a specific human being.
Yesterday, I joined PayPal as Vice President and Venmo's Chief Technology Officer. I'm truly humbled by everyone that extended their congratulations. …see more
People stop for people. A topic sentence like “Leadership is about trust” is true, general, and has nobody in it. “I stopped reviewing my team's work for a month” has a person, a decision, and an implied consequence.
Finding 3: the reader wants to be in the post
47.5% of top posts use “you” or “your” somewhere in the visible hook, against 32.7% of weak posts. Combine findings 2 and 3 and you get the shape of a strong opener: my story, your problem. The author supplies the specific experience. The reader supplies the stakes.
Dear recruiters, I know you have an extremely difficult job. The hiring market is white hot and getting worse (better, for candidates) by the hour. We all get pinged repeatedly, daily. I mostly don't mind. …see more
Finding 4: comments are the metric that actually separates winners
This is the most under-appreciated number in the study. Look at the raw medians:
| Group | Median reactions | Median comments |
|---|---|---|
| Top decile | 235 | 72 |
| Whole cohort | 122 | 12 |
| Bottom half | 80 | 5 |
Reactions roughly triple between the bottom half and the top decile (80 to 235). Comments go up more than fourteenfold (5 to 72). Whatever a top post is doing, it is doing it to the comment box. Questions in the hook appear in 19.0% of top posts vs 14.2% of weak ones, and the effect is stronger still among posts whose engagement skews hardest toward comments. If you want comments, end an open loop the reader can close. More on this in the comment strategy playbook and how to end a post.
Finding 5: native video wins, shared links lose
Media format mix (% of posts in each group)
Native video
Image
Shared article or link
Text only
Two things here surprise people. First, native video is the single strongest format signal in the data, appearing in 26.3% of top posts against 10.1% of weak ones. It is also the format most professionals refuse to use, which is precisely why the advantage is still available. See the video breakdown.
Second, images skew weak: 40.3% of bottom-half posts carry one against 30.7% of top posts. That cuts against the universal advice to always attach a graphic. Our read, which the data cannot prove, is that a stock image is the default dressing on a low-effort post, so the image is a symptom rather than a cause. We unpack that in text-only vs image posts.
Posts built around a shared external link appear 1.5 times more often in the bottom half. LinkedIn's incentives are not subtle: content that keeps people on the platform gets distribution. The full picture, including the studies that disagree with each other, is in do external links kill your reach.
Finding 6: document posts over-index, but the sample is tiny
Document posts, the format behind LinkedIn carousels, make up 1.2% of top-decile posts and 0.1% of the bottom half. That is a twelvefold over-index and the largest ratio in the format data. It is also based on only 60 document posts in the entire cohort, which is nowhere near enough to conclude anything. We are reporting it because it is suggestive and because most pages would quietly round it up into a headline. Treat it as a hypothesis worth testing on your own account, not a finding. More in the carousel guide.
Finding 7: a few hashtags are fine, a wall of them is a warning sign
78.0% of top posts carry at least one hashtag, against 59.0% of weak posts, which reflects an era when tags still aided discovery. But four or more hashtags flips the signal hard: 41.9% of bottom-half posts vs 25.9% of top posts. Hashtag walls travel with weak content, most likely because they are what people do instead of writing a better opening line. LinkedIn has since removed hashtag following and de-emphasized tags in search, which makes the case for using them weaker still. Full analysis in do hashtags still work on LinkedIn.
Finding 8: emoji in the hook over-index ninefold
24.3% of top posts use at least one emoji in the visible hook. Weak posts: 2.7%. That is the largest ratio in the entire study, and also the one to treat most carefully. Emoji use travels with a persona, the kind of creator who writes casual, personal, first-person posts, so some of this effect is finding 2 wearing a different hat. The defensible read: an emoji will not hurt you, and the stiff corporate register that avoids them entirely correlates with the bottom half.
Finding 9: the patterns hold in most topics, and clearly fail in one
This is the section we would most like to have written differently, and the one worth reading most carefully.
Comparing top-decile posts against the whole cohort has an obvious flaw: subject matter is not held constant. Leadership posts and job-search posts are different animals, and a comparison across all of them partly measures which subjects happen to attract engagement. So we ran a stricter cut. Inside each topic, we compared that topic's own top decile against that topic's own bottom half. Same subject, same kind of author, only the performance differs. This removes the biggest confounder in engagement data, and it is the strongest thing in this study.
Every row below meets our reporting rule: at least 8 distinct authors in both the top and bottom group, so the figure describes a pattern rather than a few prolific accounts.
| Topic (posts / authors) | First person, top vs bottom | “You” in hook, top vs bottom | Native video, top vs bottom |
|---|---|---|---|
| Finance (2,795 / 51) | 12.9% vs 2.1% | 46.2% vs 14.9% | 21.9% vs 3.1% |
| Leadership (2,117 / 53) | 13.3% vs 9.3% | 52.6% vs 30.4% | 36.5% vs 7.9% |
| Marketing (1,470 / 49) | 23.1% vs 14.0% | 53.7% vs 38.6% | 41.5% vs 11.6% |
| Customer experience (1,463 / 52) | 24.0% vs 10.7% | 47.9% vs 28.1% | 24.7% vs 9.0% |
| Career growth (1,389 / 50) | 15.9% vs 13.7% | 64.5% vs 34.4% | 31.9% vs 9.1% |
| Education and learning (969 / 56) | 29.2% vs 11.8% | 36.5% vs 24.5% | 31.3% vs 9.7% |
| Startups and founders (833 / 50) | 39.8% vs 11.0% | 25.3% vs 22.1% | 32.5% vs 13.2% |
| Technology (1,628 / 49) | 20.4% vs 21.9% | 28.4% vs 26.9% | 16.0% vs 12.0% |
Seven of those eight rows tell the same story, and the video column is the most consistent thing in the entire dataset: in marketing, the best posts use native video 41.5% of the time against 11.6% for the weakest, and leadership runs 36.5% against 7.9%. Finance is the sharpest case for writing in the first person, at 12.9% against 2.1%, roughly a sixfold difference within a subject where almost nobody writes personally at all.
Then there is technology.First-person openers appear in 20.4% of that topic's top decile and 21.9% of its bottom half, which points the wrong way. Second person is flat (28.4% against 26.9%). Video is the weakest separation of any topic we measured (16.0% against 12.0%). Whatever distinguishes a strong technology post from a weak one, it is not the thing that distinguishes them everywhere else.
Productivity contains a second reversal worth naming: “you” appears in 46.2% of that topic's top-decile hooks against 58.9% of its bottom half (1,300 posts, 46 authors). Direct address, which helps almost everywhere else, is associated with the weaker half of productivity content. Our read, and it is a guess rather than a finding, is that productivity is the topic most saturated with advice-giving, and that second-person phrasing there reads as one more listicle.
We are reporting the exceptions because a study that finds the same effect in every subgroup usually has a method problem, not a discovery. If the patterns really were universal, that would be evidence we were measuring something about our filtering rather than about writing. The technology reversal is the most useful single line in this study for anyone posting about technology: copy the norms of your subject, not the norms of LinkedIn. Each topic's own numbers, hook-style breakdown and real examples are on the post examples hub.
Does the pattern survive holding the author constant?
Topic is one confounder. The author is the bigger one, since a handful of prolific accounts can dominate any cut of a dataset like this. So we ran the same within-segment comparison by author role: each role's own top decile against its own bottom half.
The founder and CEO bucket shows the widest gap we measured anywhere. First-person openers appear in 19.6% of that group's top-decile posts and 5.2% of its bottom half, close to a fourfold difference. For a group whose weakest posts are mostly company announcements written in the corporate third person, the instruction is unusually clear: the difference between a founder's best and worst posts is largely whether a person is visible in them. Role breakdowns are on the founders and CEOs page and its siblings.
One role bucket is deliberately missing. Our recruiters segment contains 61 posts from just 2 authors, which is not a pattern, it is two people. We do not publish it, and we would treat any recruiter-specific engagement figure elsewhere with the same suspicion unless it states its author count. The hiring topic (543 posts, 46 authors) is the honest substitute.
What did not matter: three myths the data killed
- Hook length. Median visible hook: 206 characters for top posts, 205 for the bottom half. Word counts nearly identical at 36 and 34. Writing a longer or shorter opener changes nothing by itself. What the characters do is everything. See how long a LinkedIn post should be.
- Numbers in the hook.32.1% of top posts contain a digit in the hook. So do 32.8% of weak posts. “Always lead with a stat” has no support here. A number helps only when it carries a story.
- Opening with a question word. 12.4% vs 10.6%, a difference small enough to ignore. Note the contrast with finding 4: a question somewhere in the hook is worth more than a question as the very first word. Real curiosity beats the interrogative formula.
How do you compare?
Benchmarks are useless without your follower count attached, so here are the cohort percentiles, expressed per 1,000 followers using the weighted formula above.
| Percentile | Engagements per 1,000 followers | What it means |
|---|---|---|
| 50th (median) | 0.40 | A typical post from an established creator |
| 75th | 1.27 | Clearly above average for this cohort |
| 90th | 5.95 | Top-decile. Roughly 15x the median |
Note how steep that curve is. The jump from the median to the 75th percentile is about 3x, and from there to the 90th is another 4.7x. LinkedIn engagement is not normally distributed, which is worth remembering before you judge a single post. Work out your own number with the engagement benchmarks guide.
How does this compare to the other published LinkedIn studies?
Ours is not the biggest, and it is worth saying so before anyone else does. Buffer has analysed more than 4.8 million posts. Metricool's link study covered 673,658. Against those, 12,988 posts from 65 creators is small.
Sample size is not the only axis, though, and it is the one the niche competes on because it is the easiest to print in a headline. Three things matter more for whether a finding is usable.
| Question to ask a study | Why it matters | Our answer |
|---|---|---|
| Is the denominator stated? | Engagement without follower normalisation measures account size | Yes: (reactions + 4 x comments) per 1,000 followers |
| Is the author held constant? | Otherwise you measure who posts, not what works | Yes, within-topic and within-role cuts |
| Is the author count published per figure? | A big n from few accounts describes those accounts | Yes, and we suppress segments under 8 authors |
| Are the exceptions reported? | Universal findings usually signal a method problem | Yes, see the technology and productivity reversals |
A study of 4.8 million posts that reports raw average engagement without a follower denominator will tell you, with enormous statistical confidence, that large accounts get more reactions. That is a real number and a useless one. We would rather have 12,988 posts with the confounders removed than a million with them left in, and we would rather publish the two topics where our own pattern breaks than claim it holds everywhere.
What this study cannot tell you
Every limitation below is a reason to trust some other part of this page less. We would rather list them than have a reader discover them.
- It is correlational. Nothing here is an experiment. We did not assign anyone to write in the first person. Posts that open with a person may perform better because of the opener, or because writers who open that way differ in other ways we cannot see. Treat every finding as a description of what winning posts look like, not a guarantee of what will make a post win.
- It describes the visible hook only.The source text is truncated at LinkedIn's “see more” fold, so we can measure roughly the first 210 characters and nothing after. Any claim we made about full post bodies would be invented. This is why the study is silent on post length, structure and calls to action.
- The cohort skews to established creators. Authors need 1,000+ followers and posts need 25+ reactions to enter. That is 65 accounts. Nothing here describes what happens to an account with 200 followers, and we would not extrapolate.
- Engagement counts are lifetime-cumulative at scrape time. Post ages vary, so an older post has had longer to accumulate. We normalise by follower count, not by age.
- Reactions and comments are not the goal. They are what we could measure. Nobody is paid in reactions, and the relationship between engagement and pipeline is weak enough that we wrote a separate guide about it.
And the one that matters most for currency:this data was collected before LinkedIn rebuilt feed retrieval and ranking in March 2026. It describes what earned engagement under the previous distribution system. We expect the hook and format findings to age reasonably, because they describe how readers respond rather than how the machinery works, and the ranking half of any such system is trained to predict human response. But anything here that is really a claim about distribution mechanics should be read as historical. What changed, from LinkedIn's own engineering post rather than from the summaries circulating, is in what actually changed in the 2026 algorithm.
How do you run this analysis on your own account?
Our numbers describe 65 creators who are not you. Your own account is a better dataset for your own decisions, and the method transfers. It takes about an hour.
- Export your last 50 posts with reactions, comments and the date. Your post analytics export or a manual sheet both work.
- Compute a rate, not a raw count. Use the same formula we did: (reactions + 4 x comments) divided by followers. Weighting comments is a judgement call, and we explain ours above. Raw reaction counts mostly measure how long ago you posted.
- Split at your own median. Do not use our thresholds. You want your top 10 posts against your bottom 25, which holds your voice, subject and audience constant in exactly the way our within-topic cut does.
- Tag the hooks by hand.First person or not. “You” present or not. Question or not. Video, image, link or text. Fifty posts takes twenty minutes and no tooling.
- Only believe gaps that are large.With 50 posts, a difference of a few percentage points is noise. A 3x difference is worth acting on. This is the step people skip, and it is why most personal “what works for me” conclusions are wrong.
If your findings contradict ours, trust yours. Our cohort is 65 creators with large followings; you are one person with a specific audience. The value of this study is the method and the direction, not the decimal places. Our content audit guide covers the fuller version, and the engagement rate calculator does step two for you.
The playbook, compressed
- Write the first 210 characters as if they are the entire post. They are.
- Open in first person with a specific moment, decision, or mistake.
- Put the reader in it. “You” belongs in the hook.
- Want comments? Ask a question you genuinely want answered, and leave the loop open.
- Go native. Video is the most under-used advantage in the data.
- Links go in the comments if they must exist at all.
- Zero to three hashtags. Never four or more.
- Stop padding the hook and stop forcing a statistic into line one.
- Judge yourself on comments per 1,000 followers, not on reactions.
Why we ran this study
If you want the practical version rather than the research, start with how to write a hook, then how to build a content strategy around it. If you want to see the raw material, the examples hub has real high-performing hooks across all 20 topics with their engagement attached.
Frequently asked questions
What is a good engagement rate on LinkedIn?
In our cohort of 12,988 posts, the median post earned 0.40 engagements per 1,000 followers using a comment-weighted formula. The 75th percentile is 1.27 and the 90th percentile is 5.95. In raw counts the median post drew 122 reactions and 12 comments. If you are clearing roughly 1.3 engagements per 1,000 followers you are in the top quarter of this cohort.
What kind of LinkedIn post gets the most engagement?
A native post that opens in first person with a specific unresolved moment, addresses the reader directly, and asks a question worth answering. Native video over-indexes about 2.6 times in top-decile posts, while posts built around a shared external link over-index about 1.5 times in the bottom half. Format helps, but the opening line does most of the work.
Does posting time matter more than the content?
No. Timing shifts who sees a post in its first hour, but every content pattern in this study separated strong posts from weak ones regardless of when they went out. Fix the first 210 characters before optimizing the clock, because a well-timed post with a weak hook still fails the initial test audience.
Do these findings apply to small LinkedIn accounts?
Treat them as directional below 1,000 followers, since that is where our cohort starts. The underlying mechanics still apply: LinkedIn shows every post to a small test audience first, and the same hooks that earn engagement from 100,000 followers earn it from 500. Engagement rate is measured per follower, which makes small and large accounts comparable.
Is this study causation or correlation?
Correlation. We can say top posts open in first person about twice as often as weak ones. We cannot prove the opener caused the engagement. The patterns hold across 65 creators and 20 content topics, which is what you would expect if the patterns themselves do real work, but confounders like audience quality and posting history are not controlled for.
How did you calculate engagement rate?
Engagement rate is (reactions + 4 x comments) divided by the author's follower count, reported per 1,000 followers. Comments carry four times the weight of reactions because they are far scarcer, take real effort, and push a post into the commenter's own network. Weighting by followers lets a 3,000-follower account be compared against a 300,000-follower one.