How to run a LinkedIn content audit

A LinkedIn content audit is a review of your last 30 to 90 days of posts, scored on one comparable measure, grouped by format, topic and opening style, and finished with a written decision about what you will stop publishing. The last part is the point. Most audits produce a pleasant hour of rereading your own work, a vague sense that video did well, and no change to next month. The framework below forces every group of posts into one of four buckets, and one of those buckets is stop.

What is a LinkedIn content audit actually for?

Three jobs, and only the third one is difficult.

The first is bookkeeping. Get your posts into one place with their numbers attached, so that you are reasoning about forty posts rather than about the three you remember. Memory is heavily biased toward your best post and your most recent one, which are usually not the same post and are almost never representative.

The second is pattern-finding. Group the posts by things you controlled and see whether the groups differ. Format, topic, opening style, day. This is where most audits stop, and it is genuinely useful, but a pattern you notice and do not act on has cost you an afternoon for nothing.

The third is subtraction. Deciding to stop writing a category of post is harder than deciding to write more of something, because every post type you publish has a reason attached and the reason usually still sounds good. The audit exists to put a number next to the reason. If your industry commentary posts have sat below your median for three months and produced no conversations, the reason for writing them has been tested and it lost.

Subtraction is also what makes room. Nobody has time to add three post types to their week. Almost everyone has time to swap one out.

How far back should a LinkedIn content audit go?

Between 30 and 90 days, with 90 as the default for most accounts and 30 only if you post daily.

The lower bound is about sample size. At three posts a week, 30 days gives you 12 or 13 posts. Split those by format and you have four in one group and three in another, which is not enough to distinguish anything from luck. Ninety days at the same cadence gives you around forty, which supports two or three cuts before you are fitting patterns to noise.

The upper bound is about comparability. Over six months your follower count changed, your writing changed, your subject matter probably drifted, and LinkedIn changed things it did not announce. A post from March is not a fair comparison for a post from September even after you normalise for followers. If you want to look further back, do it as a separate exercise: read the old posts for what they say, not for what they scored.

One exception. If you are auditing after a deliberate change in direction, start the audit window at the change. Mixing eight weeks of your old approach with four weeks of the new one produces an average that describes neither.

What goes in the audit sheet?

One row per post. The columns split into what you chose, what happened, and what you thought. People fill in the middle group and skip the other two, which is exactly backwards, because the middle group is the only one LinkedIn already stores for you.

The last column does more work than people expect. Some posts perform and cost you three hours. Some perform and are not the sort of thing you want to be known for. A number cannot tell you either of those things and you will forget both within a fortnight.

If you keep a per-post log continuously, the audit stops being an archaeology project. The full version of that log and the reasoning behind each column is in how to analyze your LinkedIn post performance.

How do you score posts fairly across three months?

Divide by followers, use the median, and separate reactions from comments. Three rules, and each one prevents a specific mistake.

Divide by followers. If your audience grew 20% over the quarter, your recent posts get a 20% head start on raw counts alone. Normalising per 1,000 followers removes it. Our cohort measure is reactions plus four times comments, divided by followers, reported per 1,000. The weight on comments is a judgement call rather than a measurement, chosen because comments are far scarcer: across 12,988 posts, top-decile posts show a median of 72 comments against 5 in the bottom half, while reactions differ by much less, 235 against 80.

Use your median as the baseline. Not your best post, and not the average. LinkedIn engagement is skewed enough that one unusual post pulls an average away from anything typical. Your median is the post you are most likely to publish next, which makes it the only useful comparison point.

Keep reactions and comments visible separately. A post type that earns reactions and no discussion is doing something different from one that earns discussion and few reactions, and collapsing them into one score hides which is which.

For an outside reference point, our cohort-wide percentiles are a median of 0.40 engagements per 1,000 followers, 1.27 at the 75th percentile, and 5.95 at the top-decile cut, with a median post earning 122 reactions and 12 comments. Compare against those with care: every account in that cohort already had 1,000 followers and every post had at least 25 reactions, so it describes established creators and flatters a smaller account. More on that in the engagement benchmarks guide, and you can score a single post through the engagement rate calculator.

What do 12,988 posts suggest you should look for?

Use our cohort figures as a list of hypotheses to check against your own archive, not as verdicts. Your audience is not our cohort, and the whole point of auditing your own posts is that general findings do not settle specific accounts.

Three things to take from that table into your own audit. Video and text carry roughly double the median comment counts of images and link shares. Images are the most published format in the cohort and the most over-represented in the bottom half. And link shares are the weakest on every measure, which matters because they are also the easiest post to produce on a busy week, so they accumulate quietly in an archive. The fuller argument is in what the data says about external links and text-only against image posts.

One more hypothesis worth carrying in: four or more hashtags appear in 41.9% of bottom-half posts against 25.9% of the top decile. If your sheet shows most of your posts carrying a block of six hashtags, that is a cheap thing to change and test.

A caution before you act on any of it. Format is a hypothesis, not a verdict about your writing. This image post from our dataset earned far more discussion than the image-format median of 10 comments would predict:

How do you audit your openings and not just your formats?

Format is the easiest cut and the opening style is the most useful one, because it is the variable you can change on every post you write from now on without buying a camera.

Tag every row in your sheet with one of five opening styles: a story or first-person incident, a question, a flat declaration, a number-led list, or direct address to the reader. Then compute the median rate for each group, as you did with format. Two cohort-level gaps are worth checking against your own numbers. First-person openers appear in 19.6% of top-decile posts and 10.1% of the bottom half. The word “you” appears somewhere in the hook of 47.5% of top-decile posts against 32.7% of bottom-half posts.

The within-topic cuts are sharper still, because they hold subject matter constant instead of comparing a topic against everything else. Among leadership posts, the best ones use the word “you” in 52.6% of hooks (n=211 posts from 14 authors) against 30.4% in that same topic's weakest posts (n=1,059 from 33 authors). Among career growth posts the split is wider: 64.5% in that topic's own top decile (n=138 from 19 authors) against 34.4% in its own bottom half (n=695 from 33 authors). If either subject is one of your pillars, that is a specific thing to look for in your sheet: how often does your opening line address the reader at all?

Read the first lines in your bottom ten posts together. The most common finding in this exercise is not that they used the wrong style. It is that they were about you, addressed to nobody in particular, and could have opened any of forty posts. The taxonomy and the numbers behind each style are in what 12,988 posts say about LinkedIn hooks.

How do you tell a real pattern from a coincidence in your own archive?

Four checks, in the order that catches the most errors soonest.

1. Drop the best post from each group and recompute. This single check kills most false conclusions. If video beats text only because one video went unusually well, you have learned about that video.
2. Split the period in half and check both halves. A pattern that holds in the first six weeks and the second six weeks is worth acting on. One that appears in only one half is usually about something that changed, not about the variable.
3. Count the group sizes before you believe the medians. Three posts in a group is an anecdote regardless of what the median says. Note the count next to every median so you cannot forget.
4. Say it out loud.If the finding is “my video posts get about double the comments”, that is a difference an archive of forty posts can support. If it is “posts published at 8:15 do 9% better”, it is not, and no amount of spreadsheet formatting will make it so.

The reason the bar is this high is worth internalising. Even across 12,988 posts, some variables do not separate the best from the worst 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. If a sample that size cannot see a difference, forty posts cannot see a smaller one. The same logic drives what is and is not testable in how to A/B test LinkedIn posts.

How do you decide what to stop, fix, keep, or scale?

Every group in your audit goes in exactly one bucket. Forcing a choice is the mechanism. Groups allowed to stay unclassified survive by default, which is how a link-share habit lasts two years.

Two rules about the stop bucket. It has to contain something, or the audit did not happen. And whatever goes in it needs a written reason, because in three months you will have forgotten the evidence and only remember that the post type felt useful.

What should you usually stop writing?

Four categories account for most of what ends up in the stop bucket. None of them are bad posts in principle. They are the posts people publish when they need to publish something.

• Link shares with a sentence of agreement on top.The cheapest post to produce and the weakest performer in our cohort on every measure. If the article is genuinely worth your audience's time, write the argument yourself and mention where you read it.
• Milestone announcements with no transferable content. A certificate, an anniversary, an award. Fine occasionally. As a category they teach a reader nothing and give them no reason to return.
• Generic industry commentary. Observations about your field that any of four hundred people could have written. These are the hardest to cut because they feel professional. Check the conversations column: they almost never produce one.
• Posts written because it was Tuesday. You will recognise these in the sheet by the blank first-line column, because you cannot remember how they opened.

What replaces them is usually narrower rather than more elaborate. Specific stories from your own work, opinions you would defend in a meeting, and answers to questions clients keep asking. Our data supports the first of those: first-person openers appear in 19.6% of top-decile posts and 10.1% of the bottom half, and among founders and CEOs the gap inside their own posts is wider still, 19.6% of that group's top decile (n=561 posts from 17 authors) against 5.2% of its own bottom half (n=2,810 from 12 authors).

What if you do not have enough posts to audit?

If you have published fewer than about twenty posts in the window, skip the arithmetic and do the reading version instead. Print the posts. Read them in order. Answer four questions in writing:

1. Which three would you send to a prospective client, and why those three?
2. Which three would you delete, and what do they have in common?
3. What subject appears most often, and is it the subject you want to be known for?
4. Reading them together, what do they say you believe?

That exercise is more useful than a median computed over fifteen posts, and it is the right audit for anyone in their first few months. The numbers become informative once you have enough of them to have a median worth comparing against. Until then, the honest instrument is your own judgement and the comments people actually left.

How often should you run a LinkedIn content audit?

Quarterly, with a continuous lightweight log in between.

Monthly audits fail for a specific reason rather than a general one. A month gives you twelve or thirteen posts, split across three formats and four topics, which leaves groups of three. Conclusions drawn from groups of three change every month, and reacting to them means you never leave any approach in place long enough to find out whether it works. The appearance of responsiveness, with the effect of paralysis.

Quarterly gives you forty posts, groups of ten, and enough time between audits that the changes you made after the last one have had a chance to show up. It also matches the pace at which anything on LinkedIn actually changes for an individual account.

Between audits, spend two minutes per post filling in the log. That is the difference between a ninety-minute audit and a lost Saturday. The cadence question more broadly is covered in building a LinkedIn content calendar, and the strategy the audit feeds back into is in the LinkedIn content strategy guide.

How do you run a LinkedIn content audit in ninety minutes?

1. Twenty minutes. Pull your last 90 days of posts into the sheet from your activity feed and analytics. Fill the numbers in first, then go back for the input columns.
2. Ten minutes. Add the computed rate per 1,000 followers. Sort descending. Compute your median. Write it at the top of the sheet where you will see it.
3. Fifteen minutes. Read the top five and the bottom five posts as text, not as numbers. Note what the top five have in common in one sentence. Do the same for the bottom five.
4. Fifteen minutes. 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. Fifteen minutes. Repeat for topic, then for opening style. Stop at three cuts. A fourth cut on forty posts is decoration.
6. Ten minutes. Put every group in one of the four buckets. Every group, no abstentions.
7. Five minutes. Write three sentences: what you will do more of, what you will stop, and what you will change about one thing you are keeping. Put a reminder in the calendar for ninety days.

The usual caveats apply to every cohort figure quoted here. Our analysis covers 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, which is why the format table above is a list of hypotheses rather than instructions. Engagement counts are lifetime-cumulative at the time the data was collected and post ages vary. The text findings describe the visible hook above LinkedIn's fold rather than complete post bodies. The full methodology is in our study of 34,000 LinkedIn posts.

The short version of a LinkedIn content audit

Run a LinkedIn content audit over 90 days, score every post per 1,000 followers, compare groups against your own median, and check every conclusion by dropping the best post from each group. Then put every group into stop, fix, keep or scale, and make sure the stop bucket is not empty. Do it quarterly rather than monthly, keep a two-minute log per post in between, and finish with three written sentences rather than a general resolution to post better.