LinkedIn Engagement Benchmarks: What 100,000 Real Posts Actually Show
Almost every published LinkedIn benchmark reports an average. Averages on LinkedIn are pulled upward by a small number of viral posts. We measured 100,459 posts from 1,054 professionals and report the medians instead. The typical post earns a 2.29% engagement rate and 10 engagements.
Almost every LinkedIn engagement benchmark you can find online reports an average. Averages are the wrong statistic for LinkedIn.
A small number of posts on LinkedIn go very large. Those posts pull the average up and away from what a normal post achieves, so a creator who measures themselves against a published average is measuring themselves against a number that almost nobody hits.
We run ContentIn, an AI writing tool for LinkedIn. Our customers connect their LinkedIn accounts, which means we hold performance data for their posts, including the posts they published long before they ever met us. That gives us a corpus of 100,572 posts from 1,055 professionals, on real personal profiles, with performance collected from LinkedIn's own analytics.
This page reports the medians from that corpus. Where the median and the average differ, we show you both.
The benchmark that matters most: a typical LinkedIn post earns a 2.29% engagement rate
Across 93,696 posts with impression data, from 1,054 accounts:
| Statistic | Engagement rate |
|---|---|
| Median (the typical post) | 2.29% (95% CI 2.27-2.30) |
| Mean (the average post) | 3.62% |
| 90th percentile (a strong post) | 7.65% |
The mean is 1.58x the median. That gap is the entire story. Half of all posts land below 2.29%, but the arithmetic average sits at 3.62% because the top of the distribution is very long.
For comparison, two of the most widely cited third-party benchmarks report 5.20% across business pages (Social Insider, 1.3 million posts) and approximately 3.85% platform-wide (ConnectSafely, Q1 2026). Both are averages. Neither is wrong. They are simply answering a different question than the one most creators are actually asking, which is "what should a normal post of mine do?"
The two other top-line numbers from the same corpus:
| Metric | Median | Mean |
|---|---|---|
| Engagements per post (reactions + comments + reposts) | 10 | 69 |
| Impressions per post | 412 | not reported |
n = 100,459 posts, 1,054 accounts, published 2023 to August 2026.
If your posts are landing near 10 engagements and a 2.3% engagement rate, you are at the middle of the distribution for a real personal profile. That is a materially different message than "you should be hitting 5%."
Benchmark 2: format
Format is the largest single lever in our data, and the gap is wider than most published benchmarks suggest.
| Format | Posts | Accounts | Share | Median engagements | Median engagement rate | 95% CI |
|---|---|---|---|---|---|---|
| Text only | 41,804 | 993 | 41.6% | 5 | 1.54% | 1.52-1.56 |
| Image | 44,261 | 948 | 44.0% | 18 | 2.94% | 2.90-2.97 |
| Video | 9,237 | 561 | 9.2% | 16 | 2.78% | 2.73-2.85 |
| Document | 5,270 | 489 | 5.2% | 16 | 2.67% | 2.54-2.77 |
Two things in this table are worth sitting with.
Any media beats no media, by roughly 3x. Image, video and document posts all cluster around a 2.7-2.9% median engagement rate against 1.54% for text only. In raw engagement counts, the median media post earns 16 to 18 engagements against 5 for text.
We do not reproduce the industry's "documents are number one" finding. Image, video and document are statistically indistinguishable from each other in our data; their confidence intervals overlap. Published studies that rank documents clearly first are mostly measuring company pages, which is a different population with different posting behaviour. On personal profiles, the meaningful decision is media or no media, not which media.
This result holds in both halves of the corpus. Split by publication year, posts from 2023-2024 and posts from 2025-2026 both show the same ordering and the same rough magnitude, so this is not an artifact of one era of the LinkedIn algorithm.
The same finding, measured a harder way
A format table like the one above has an obvious weakness. Accounts with large, engaged audiences may simply post more images, in which case the table is measuring account quality and calling it a format effect.
So we ran the comparison a second way: within each account. For every account with at least three text-only posts and three image posts, we compared that person's image posts against their own text posts. Account size, audience quality and industry all cancel out, because each person is compared only against themselves.
| Comparison | Accounts | Median difference in engagements | 95% CI | Share of accounts where the format wins | Median difference in rate |
|---|---|---|---|---|---|
| Image vs their own text-only posts | 741 | +7 | 6 to 8 | 81.6% | +0.77pp |
| Video vs their own text-only posts | 328 | +7.25 | 6 to 9.5 | 79.3% | +0.68pp |
| Document vs their own text-only posts | 281 | +6 | 4 to 7.5 | 79.4% | +0.42pp |
Adding an image beats the same person's own text-only posts for 81.6% of the 741 accounts we could test. The effect survives the control, which is the bar we care about.
One honest limit: this is correlational, not causal. People may attach an image to posts they had already decided were important, and put more effort into them. We can measure that images and strong posts travel together. We cannot prove the image is what caused it.
Benchmark 3: engagement rate by follower count
Raw engagement counts rise with audience size. Engagement rate does not follow them up.
| Followers | Posts | Accounts | Median engagements | Median engagement rate |
|---|---|---|---|---|
| Under 1,000 | 7,710 | 197 | 5 | 2.17% |
| 1,000 to 2,499 | 11,723 | 245 | 7 | 2.19% |
| 2,500 to 4,999 | 8,933 | 160 | 9 | 2.42% |
| 5,000 to 9,999 | 10,090 | 128 | 10 | 2.24% |
| 10,000+ | 15,381 | 97 | 13 | 1.77% |
Median engagements climb cleanly from 5 to 13 as an audience grows from under 1,000 to over 10,000 followers. A bigger audience does deliver more engagement.
The rate tells a different story. Between the under-1,000 band and the 5,000-9,999 band it sits in a narrow range of 2.17% to 2.42% and does not move in any consistent direction. The only unambiguous movement is the drop at 10,000+, where the median rate falls to 1.77%, below every smaller band.
We are deliberately not calling this a smooth decline, because our data does not show one. The 2,500-4,999 band outperforms both its neighbours, and we cannot explain why. What the data does support is narrower and still useful: growing past 10,000 followers reliably buys you more total engagement and a lower percentage.
The practical implication is about how you set targets. A 2.5% engagement rate at 800 followers and a 1.8% rate at 15,000 followers are both normal. Comparing your rate against someone with a very different audience size is comparing two different games.
Follower counts here are the account's snapshot for the month each post went out. Only accounts with follower history are included, which is why the account counts are lower than the corpus total.
Benchmark 4: post length
| Plain-text length | Posts | Accounts | Median engagements | Median engagement rate |
|---|---|---|---|---|
| Under 500 characters | 32,381 | 952 | 5 | 1.79% |
| 500 to 899 | 20,238 | 894 | 13 | 2.34% |
| 900 to 1,299 | 18,696 | 891 | 17 | 2.61% |
| 1,300 to 1,699 | 12,646 | 816 | 16 | 2.68% |
| 1,700 to 2,199 | 8,392 | 693 | 15 | 2.64% |
| 2,200+ | 6,568 | 583 | 13 | 2.77% |
Engagement counts peak in the 900 to 1,299 character band at a median of 17, then decline gently. Engagement rate does not peak there; it keeps climbing all the way to 2.77% at 2,200 characters and above.
Those two facts are compatible. Long posts appear to reach fewer people but hold the people they do reach. If you are optimising for total engagement, the 900 to 1,700 character range is where the mass of good outcomes sits. If you are optimising for depth of response from a smaller, more committed audience, length is not penalised.
Two caveats on this table. Length is measured on plain text after stripping formatting, not on raw characters including markup. And this cut is not controlled for format: the under-500 bucket contains a disproportionate number of link-style and reshare-style posts, which drag it down for reasons that have nothing to do with brevity.
What we could not measure, and are not going to guess
Benchmark pages tend to answer every question asked of them. We tested several claims that are standard on pages like this one and could not support them. Rather than fill the gap with an estimate, here is what we dropped and why.
Best time of day to post. A third of the posts published through ContentIn land in a single three-hour UTC window, because that is our own scheduler's default. Any "best hour" figure drawn from this corpus would be measuring our product's settings rather than creator behaviour, so we are not publishing one.
Optimal posting frequency. We tested it. Only three posts-per-week buckets cleared our minimum sample size, the highest of them was "2 to 3 posts per week", and their confidence intervals overlapped almost entirely. We could not detect an effect, and we could not even populate a "3 to 5 posts per week" bucket well enough to comment on the most commonly repeated advice on the internet. No number here.
Company pages versus personal profiles. Our corpus is personal profiles almost entirely; it contains fewer than ten company-page accounts. Publishing a "company page benchmark" from a sample that small would describe a handful of identifiable companies rather than a population, which fails both our statistical bar and our privacy bar.
Opening line and hook patterns. The raw cut looked strong: posts with a first line of 80 characters or fewer showed a median of 12 engagements against 7 for posts opening above 140 characters. Then we ran it within-account, and the effect disappeared completely (43.1% of 635 accounts, median difference of zero). It was an account-mix confound, not a hook effect. We are reporting it here because a lot of published hook advice rests on exactly the uncontrolled version of this comparison.
Saves, sends, profile views and followers-gained per post. These columns exist in our database and have been empty since July 2026, when our browser extension lost distribution. We are not going to report figures from a partial pre-July window and present them as current.
One contrarian result: links in the post body
The near-universal advice is to keep links out of the post and put them in the first comment. We could not find the penalty.
Comparing each account's posts containing a URL against that same account's posts without one, across 513 accounts, the median difference was +0.5 engagements in favour of the no-link posts, with a confidence interval of 0 to 1, and a rate difference of +0.09pp. Posts without a link came out ahead for 50.5% of accounts, which is a coin flip.
This is a null result, and null results deserve caution. We cannot distinguish between link types, so a link to a personal newsletter and a link to an unrelated commercial page are pooled together here. What we can say is that in 513 accounts' worth of paired comparisons, the link penalty was not large enough for us to detect it. If you have been routing every link into the first comment and finding it awkward, our data does not support the tax you think you are paying.
How we measured this
The figures on this page come from ContentIn's own database, not from third-party estimates or scraped public data.
The corpus. 100,572 LinkedIn posts published by 1,055 professionals, on personal profiles, with performance collected from LinkedIn's post analytics. 99.8% of these posts were published in 2023 or later, and the analysis window is 2023 to August 2026. These are our customers' own LinkedIn histories, imported when they joined, which is why the corpus reaches back well before they became customers. Every figure on this page names the number of posts and accounts behind it.
We report medians, not averages. A handful of viral posts pulls an average far above what a typical post achieves. In our data the average engagement rate is 58% higher than the median. Medians describe the post you are actually about to publish; averages describe the corpus. Where a mean is informative, we show it next to the median rather than instead of it.
Definitions. Engagements are reactions plus comments plus reposts. Engagement rate is engagements divided by impressions, calculated only on posts where impressions are greater than zero. Some imported posts carry engagement counts but no impression figure, so the sample for rate-based claims is smaller than the sample for count-based claims, and both are stated separately.
Every metric is a running maximum across all snapshots of a post, not its most recent reading. This matters more than it sounds. LinkedIn's analytics occasionally return zero for a post that previously reported real numbers, and 32% of the posts in this corpus have an all-zero latest reading despite a non-zero maximum. An analysis that naively joined on the most recent snapshot would report roughly a third of this corpus as having zero engagement. Taking the maximum also neutralises a period in early 2026 when our own fetcher wrote a zero on API errors.
Minimum sample sizes. No bucket appears on this page with fewer than 25 posts, and nothing account-shaped appears with fewer than 25 distinct accounts. Buckets that failed either test were dropped, not merged into a neighbour.
Within-account comparisons. Where we compare formats or writing choices, we compare them within the same account, meaning the same person's image posts against their own text posts. Differences in audience size and audience quality cancel out rather than masquerading as a format effect. We report the share of accounts where the effect holds alongside the median size of the effect, because a large effect present in half of accounts and a small effect present in four fifths of them are different findings.
One selection caveat. Where we cite figures from posts published through ContentIn (this page does so once, in the robustness note below), those posts are a tracked subset that over-represents media: 75.2% of tracked posts carry an image, video or document against 68.2% of all posts published through the product. Posts that attracted analytics coverage skew toward media, and any comparison drawn from that subset carries that skew.
Robustness note. The 2.29% median engagement rate above is measured on our customers' own LinkedIn histories. The separate population of posts published through ContentIn gives a near-identical median of 2.34% on 10,876 posts from 387 accounts, subject to the media-skew caveat above. Two differently-collected populations landing within 0.05pp of each other is the main reason we are comfortable anchoring on this number.
Aggregation. Every figure is aggregated. No individual account, post, or person is identified anywhere on this page, and no customer data leaves this analysis in any other form.
What this means if you are trying to improve
The benchmarks above suggest a short list, in rough order of expected effect:
- Attach media. It is the largest and most robust effect we can measure, it holds within-account, and it holds for roughly four accounts in five.
- Write to length. The 900 to 1,700 character range carries the highest engagement counts in our corpus.
- Reset your target. If you have been benchmarking against a 5% engagement rate, you have been comparing yourself against an average of a long-tailed distribution. 2.3% is the middle.
- Stop optimising the things we could not detect. Posting hour, exact weekly cadence, opening-line length and first-comment link placement all failed to survive a controlled test in our data. That does not prove they are worthless. It does mean the evidence commonly cited for them is weaker than it is presented as being.
Related reading: how long a LinkedIn post keeps earning engagement, our hour-by-hour measurement of 2,030 tracked posts, and 60+ LinkedIn content statistics for 2026, which sets these first-party figures alongside the major third-party studies.
Where ContentIn fits
ContentIn writes LinkedIn posts in your voice, trained on how you already write, and schedules them. The data on this page is a by-product of running it: our customers connect their LinkedIn accounts, so we can see what actually happened to their posts rather than estimating it.
The practical version of the findings above is built into the product. Posts are drafted at the lengths that perform, media is prompted rather than optional, and your own analytics are measured against the medians on this page rather than against an industry average nobody hits.
Related posts
Create Engaging LinkedIn Content
Use ContentIn's AI Ghostwriter to write posts that resonate with your audience and build your personal brand effortlessly.