How to analyze your LinkedIn post performance

Analyze your LinkedIn post performance by recording four numbers and five inputs for every post, converting the numbers into a rate per 1,000 followers so posts from different months are comparable, and then comparing groups of posts against the median of your own work rather than against your best one. Almost every mistake in LinkedIn analytics comes from one of two places: comparing raw counts across a period when your follower count changed, or drawing a conclusion from a single post that went unusually well.

What should you record for every LinkedIn post?

Two categories, and most people only record one of them. Outputs are what the post earned. Inputs are what the post was. Only the second category can ever explain the first, and it is the one people skip because LinkedIn does not fill it in for you.

Outputs, taken from your analytics after roughly a week. Impressions, reactions, comments, reposts, and any follower or profile activity your dashboard attributes to the post. Record unique commenters rather than the raw comment count if you reply to people, because your own replies inflate the number and you will otherwise conclude that chatty posts perform better than quiet ones when you were the one being chatty. Impressions and reach are different measures and get confused constantly, which we untangle in LinkedIn impressions versus reach.

Inputs, taken from the post itself at the moment you publish.Format (text, image, video, document, link share), topic, opening style, the actual first line, character count, hashtag count, day of week, time of day, and whether you replied to comments in the first hour. Five minutes a week of bookkeeping. Without it your analysis can only ever produce statements like “March was better than February”, which is not a finding you can act on.

There is also a third category that has no column and matters more than either: who engaged, what they said, and whether anyone messaged you afterwards. A post with 30 reactions that produced two conversations with people you want as clients outperformed a post with 300 reactions from strangers. Keep a notes column and use it.

How do you compare LinkedIn posts fairly?

Divide by followers. That is most of the answer, and it is the step almost nobody takes.

Suppose a post in January earned 90 reactions and a post in June earned 120. The June post looks 33% better. If your audience grew from 2,000 to 3,200 followers in between, the January post reached 45 reactions per 1,000 followers and the June post reached 37.5. The January post was better, and the raw counts said the opposite. Every month you grow, raw counts drift upward on their own, which means an account that is slowly getting worse at writing can look like it is improving for a year.

The measure we use throughout our own analysis is reactions plus four times comments, divided by followers, reported per 1,000 followers. Two things about that formula deserve stating plainly.

First, the weight on comments is a deliberate choice, not a measurement. We picked it because comments are far scarcer than reactions in the data. Across our cohort of 12,988 posts, top-decile posts carry a median of 72 comments against 5 in the bottom half. The equivalent gap in reactions is much narrower, 235 against 80. A comment costs a reader real effort and a reaction costs one tap, so weighting them equally would let a post win on the cheap signal alone. Four is a round number that reflects that gap. If you prefer three or five, use it. What matters is that you use the same weight every time.

Second, dividing by followers is imperfect, because LinkedIn does not show your post to every follower and the fraction it shows varies. It is still much better than not dividing at all. If your analytics gives you impressions, you can also compute engagement per 1,000 impressions, which removes the distribution effect and answers a narrower question: given that people saw it, did they respond. Both are worth having. The follower-based rate tells you how far your work is travelling. The impression-based rate tells you how good the post was for the people who saw it.

You can run a single post through our LinkedIn engagement rate calculator to see where it lands against the percentiles below without setting up a sheet first.

What do 12,988 posts say about where your numbers should sit?

Our cohort gives four reference points that are more useful than the single average figure most benchmark pages publish, because a distribution tells you where you sit and an average does not.

Look at the distance between those rows. The 90th percentile is roughly fifteen times the median, not fifty percent above it. That shape is the single most important fact about LinkedIn analytics, and it drives everything else in this article. When a distribution has that much spread, averages lie, small differences are invisible, and one post can dominate a quarter.

Two caveats before you compare yourself to those numbers. Every account in the cohort already had at least 1,000 followers and every post had at least 25 reactions to be included, so these percentiles describe established creators and flatter a small account. And the counts are lifetime-cumulative at the time the data was collected, so older posts had longer to accumulate. Fuller comparisons by account size are in the engagement benchmarks guide.

Why should you use the median rather than the average?

Because one post can move an average and cannot move a median.

Take ten posts that earned between 40 and 90 reactions, and one that earned 1,200. The average of all eleven is around 165, a number that describes none of them. The median is somewhere in the sixties, which describes the post you are most likely to publish next. When you set a target off the average, you set a target you will miss ten times out of eleven, and then you conclude your writing has got worse.

The practical rules that follow:

• Report the median for your own baseline. This is what you compare each new post against.
• Report the average only next to the median, never instead of it. The gap between them is itself informative. A big gap means your results are driven by a few posts, which tells you your strategy is closer to buying lottery tickets than you might like.
• Count your top-decile posts as events, not as performance. Note what they were about and what happened around them. Do not average them into anything.

How do you avoid being fooled by one viral post?

This is the most common analytical failure on LinkedIn, and it has a predictable shape. A post goes unusually well. You look for the cause. You find one, because you can always find one after the fact. You rewrite your approach around a sample of one. Three months later the numbers are worse and you cannot work out why.

The uncomfortable part is that the real cause is often something you cannot repeat. Someone with a large audience reshared it. It touched a news event that week. It landed in a moment when a lot of your industry happened to be scrolling. None of those are properties of the writing, and all of them look like properties of the writing when you only have one post to look at.

Three defences, in order of how much they help:

1. The drop-one test. Whatever comparison you are making, remove the single best post from each side and compute it again. If image posts beat text posts only because of one image post, you have learned something about that post and nothing about images. This one check catches the majority of false conclusions.
2. Record what happened, not just what it scored.A column noting “reshared by someone with 200k followers” costs you five seconds and saves you a quarter of misdirected effort.
3. Wait for the second one. A pattern you cannot reproduce is not a pattern. If a format worked once, publish three more in that format before you rebuild your calendar around it.

The two posts below are both from our dataset and both did well, and they illustrate why a single headline number hides what actually happened.

If you tracked only reactions, the first post is the clear winner and you would write more list posts. If you tracked comments, the second one produced far more conversation per reader, which is the thing that puts a post in front of new audiences and the thing that occasionally turns into a client. Neither number is wrong. Recording only one of them is.

How many posts do you need before a pattern is real?

The honest answer is that there is no single number, and anyone who gives you one has skipped the part that determines it. How many posts you need depends on how much your own posts vary. If your posts land between 40 and 90 reactions almost every time, a consistent 30% difference between two groups will show up quickly. If they swing between 20 and 900, the same 30% difference is invisible for a very long time, because the swinging is larger than the effect.

You can see the shape of this problem in our own data, which is a useful calibration because the cohort is enormous by individual standards. Across 12,988 posts, some differences are unmistakable: emoji appear in 24.3% of top-decile hooks and 2.7% of bottom-half hooks, and native video is 26.3% of the top decile against 10.1% of the bottom half. Others simply do not separate at all: the median hook is 206 characters in the top decile and 205 in the bottom half, and a number appears in 32.1% of top hooks against 32.8% of bottom ones.

Sit with that second group for a moment. Twelve thousand posts, and hook length still does not distinguish the best from the worst. If a dataset that size cannot separate two groups on a variable, forty posts from one account certainly cannot separate anything smaller. That is not a reason to stop measuring. It is a reason to only trust differences that are large, repeated, and survive the drop-one test.

Practical guidance that does not require inventing a threshold:

• Compare groups, never pairs. Two posts tell you nothing. Ten on each side, spread across several weeks, is the point at which the exercise starts being worth the time.
• Spread each group across the same period. If all your text posts are from spring and all your video posts are from autumn, you have measured the season, your audience growth, and possibly a change in your own writing.
• Look for differences you could describe to someone without a spreadsheet. “My video posts get roughly double the comments” is the kind of gap a small sample can support. “Posting at 8:15 beats 8:45” is not.
• Require it to hold twice. A pattern that appears this quarter and again next quarter is worth acting on. One that appears once is a hypothesis.

The related question of whether you can run a genuine experiment on LinkedIn, and what to do instead, is covered in how to A/B test LinkedIn posts. The short version is that you cannot, and the substitute is more disciplined than most people expect.

What does a LinkedIn post tracking sheet look like?

One row per post. Fifteen columns, of which you fill in nine at publish time and six a week later. Anything more elaborate than this gets abandoned by week three.

The two columns people leave blank are followers at publish and anomaly, and they are the two that make the whole sheet trustworthy. Without the first, none of your history is comparable. Without the second, your best month will mislead you for a year.

Which patterns should you look for first?

Start with the comparisons where our cohort shows large gaps, because large gaps are the ones a small sample has any chance of confirming. Treat every one of these as a hypothesis to check against your own numbers rather than as a verdict.

1. Format. Group your posts by format and compare medians. In our cohort native video is 26.3% of the top decile and 10.1% of the bottom half, while shared article and link posts run 22.1% against 32.7%. If your own numbers disagree, believe your own numbers, because your audience is not our cohort.
2. Opening style.First-person openers appear in 19.6% of top-decile posts and 10.1% of the bottom half. The word “you” appears somewhere in 47.5% of top hooks and 32.7% of bottom ones. Both are large enough gaps to look for in your own archive.
3. Topic. Group by content pillar. This is usually the most actionable cut for a personal account, because it tells you what your specific audience turns up for, and audiences differ far more by subject than by tactic.
4. Comment ratio. Comments divided by reactions, per post. Posts with an unusually high ratio said something people needed to answer. Those are the posts worth understanding, and the ratio is stable enough to compare across months without any normalisation.
5. Day and time, last. It is the variable with the smallest effect and the largest amount of confident advice attached to it. The published research is summarised in the best time to post on LinkedIn, and the honest reading is that it is a second-order lever.

What can LinkedIn analytics not tell you?

Four things, and each one has a workaround that costs less than it sounds.

• Why a post travelled. You see the number, not the mechanism. The workaround is to check the reposts and the notable names in the comments while the post is still recent, and write it down.
• Who read without engaging. The largest group of readers leaves no trace at all, and they include most of the people who will eventually contact you. This is why a post with quiet numbers and one good inbound message beats a loud post with none.
• What would have happened otherwise. There is no control group for an organic post. Every comparison you make is between two different moments, not between two versions of the same moment.
• Whether any of it produced revenue. No dashboard closes that loop. Ask new inbound conversations where they found you and keep a tally. It is a rough instrument and it is still better than every alternative.

How do you turn the analysis into a decision?

The failure at the end of the process is subtler than the ones at the start. People build the sheet, run the comparison, find something real, and then change nothing, because the finding was interesting rather than actionable.

Force a decision by writing the review as three sentences: one thing to do more of, one thing to stop, and one thing to test next quarter. Constrain it to three. A review that produces nine changes produces zero, because next quarter you will not know which change caused what.

Then leave the change in place long enough to measure. Most people change their approach every three weeks, which guarantees they will never have enough posts in any one configuration to tell whether it worked. A quarter is the shortest sensible commitment. If you want the longer version of this process, including the review of what to stop writing entirely, see how to run a LinkedIn content audit.

How do you run this analysis yourself in an hour?

1. Open a spreadsheet with the fifteen columns above. Fill in your last thirty posts from your LinkedIn activity feed and analytics. This is the boring hour and it is the only hour that matters.
2. Add a computed column for rate per 1,000 followers. Sort by it descending and read the top five and bottom five posts side by side, as text. Half the insight in this exercise comes from rereading them rather than from the arithmetic.
3. Compute the median rate across all thirty. That is your baseline from now on.
4. Group by format and compute the median of each group. Then drop the best post from each group and recompute. Note which conclusions survived.
5. Repeat for topic and for opening style. Stop there. Three cuts is enough for thirty posts, and a fourth cut on that sample size is fitting patterns to noise.
6. Write the three sentences: more of, stop, test. Put a reminder in the calendar for ninety days from now.

The usual caveats apply to every figure quoted above. Our cohort is 12,988 English posts from 65 creators, all with at least 1,000 followers and at least 25 reactions per post, so it describes established accounts rather than new ones. Every relationship is correlational rather than causal. Engagement counts are lifetime-cumulative at the time the data was collected and post ages vary. And the text findings describe the visible hook above the fold rather than complete post bodies, because that is what the source captured. The full methodology is in our study of 34,000 LinkedIn posts.

The short version of LinkedIn post performance analysis

Analyze your LinkedIn post performance by recording inputs alongside outputs, converting every post to a rate per 1,000 followers, comparing groups against your own median, and running the drop-one test before you believe any difference you find. Trust large, repeated gaps. Ignore small ones, because even 12,988 posts cannot separate hook lengths that differ by a single character. Then write down one thing to do more of and one thing to stop, and leave it alone for a quarter.