LinkedIn video posts: the strongest format signal in our data

Native LinkedIn video posts made up 26.3% of the top-performing decile in our analysis of 12,988 posts, against 10.1% of the bottom half. That is a 2.6x over-index, and unlike most format claims on LinkedIn it comes from a sample that can carry it: 1,613 video posts from 46 different creators. Video was also the only format in our data that appeared more often among the best posts than the worst by a wide margin. Images, link shares, and text posts all sat closer to even.

What counts as a native LinkedIn video post?

A native video post is a video file uploaded directly to LinkedIn through the composer. It plays inside the feed, it autoplays muted as the reader scrolls past, and LinkedIn reports its own view and completion metrics for it. That is the format this article is about.

Three things that are not native video posts, and get counted as something else entirely:

• A YouTube or Vimeo link. LinkedIn treats this as a link share with a preview card. Some of them play inline, but the post is still classified and distributed as a link.
• A video attached to an article. The article is the object; the video is decoration inside it.
• A repost of someone else's video. Useful for them. Does very little for you.

The distinction matters because our data separates the two clearly. Native video is the best-performing category in our format table. Shared links are the worst. Posting the same video as a YouTube link instead of a file upload moves it from one column to the other, which is the single easiest format mistake to fix. We covered the wider version of that argument in our analysis of external links and LinkedIn reach.

What did 12,988 posts say about LinkedIn video?

We scored 12,988 English LinkedIn posts by engagement rate, defined as reactions plus four times comments, divided by the author's follower count. Then we compared the top decile (1,298 posts) against the bottom half (6,494 posts). Here is the full format table, with the cohort sample size for each so you can judge which rows to trust.

Read the sample size column before the percentage column. The document and poll rows show bigger multiples than video, and both are built on fewer than 25 posts across the two groups combined. We treat them as suggestive at best, which is exactly how we frame them in the piece on LinkedIn carousel posts and the piece on LinkedIn polls. The video row is different. It is 1,613 posts from 46 accounts, and the gap is wide enough that it does not depend on a handful of outliers.

The absolute numbers say the same thing from a different angle. Across all 1,613 video posts, the median earned 181 reactions and 27 comments, at a rate of 0.60 engagement per 1,000 followers. The cohort median was 122 reactions, 12 comments, and 0.40 per 1,000. Image posts, the most common format by volume, had a median rate of 0.33. Link shares were also 0.33.

The comment gap is the one worth staring at. Video posts pulled more than twice the median comment count of the cohort. Comments are the scarcer signal on LinkedIn: across our whole dataset, the top-decile median was 72 comments against 5 in the bottom half, a far wider separation than reactions (235 against 80). If video reliably pulls comments, it is pulling the thing that separates good posts from average ones.

Why does native video over-index?

We can measure the pattern, not the mechanism. But four explanations are consistent with what we see, and they are not mutually exclusive.

Autoplay buys attention that text has to earn. A video starts moving as it enters the frame. A text post has to win the same attention with its first line. That is a structural head start, and it is why the first two lines of a text postcarry so much weight and a video's first frame carries the equivalent.

Watch time is engagement the reader does not have to choose. Someone can watch forty seconds of your video without clicking anything. That dwell time is a signal available to the ranking system that a scrolled-past text post never generates.

Video invites reply rather than reaction. Our video posts open with a question 16.6% of the time, against 11.4% across the whole cohort, and contain a question mark 23.5% of the time against 15.9%. Creators writing video captions ask more questions, and questions produce comments.

Selection, again. Nobody films, edits, and uploads a video casually. The format filters out low-effort posts before we ever measure it. This is the honest caveat on every format finding, ours included: part of what we are measuring is effort, not the format itself. The difference with video is that the sample is large enough that effort cannot be the whole story, since 1,613 posts from 46 creators includes plenty of mediocre video.

One more caveat specific to our dataset. It captures a period before LinkedIn built its dedicated vertical video feed, so if anything the platform has pushed video harder since these posts were published. That cuts in favour of the finding, but it also means our exact multiple is a snapshot of a moment, not a constant.

Does our finding agree with larger studies?

Partly, and the disagreement is instructive. Socialinsider's LinkedIn benchmarks, built from 1.3 million posts across 16,645 business pages between January 2024 and December 2025, rank video third by engagement rate at 6.00%, behind native documents (7.00%) and multi-image posts (6.45%), and ahead of images (5.30%), text (4.50%), polls (4.20%), and link posts (3.25%).

So which is it: is video first, or third? Both, depending on what you measure. Three differences explain most of the gap:

• Population. Theirs is company pages. Ours is personal profiles of established creators. Pages and people get treated differently by the feed and by readers.
• Denominator. Their engagement rate is per impression. Ours is per follower. A format that reaches fewer people but engages them harder scores well on theirs and poorly on ours, and the reverse.
• Question asked.They report the average engagement rate of each format. We report how often each format appears among the best and worst posts. Those answer different questions, and the second is closer to “what should I post next.”

What both agree on: video is near the top, and link posts are at the bottom. When two samples with different populations and different math agree on the ordering of the extremes, that ordering is probably real.

How long should a LinkedIn video be?

Thirty to ninety seconds covers most of what belongs on LinkedIn. The technical ceiling for native in-feed video is fifteen minutes, per LinkedIn's own video sharing documentation, but the ceiling is not the target. Third-party guides commonly cite ten minutes, which is out of date.

Length is a budget you spend against attention:

• Under 30 seconds. One idea, one claim. Good for reactions to news, single tips, and anything that would otherwise be a two-line text post.
• 30 to 90 seconds. The workhorse. Enough for a setup, a point, and an example. This is where most video posts should live.
• 90 seconds to 3 minutes. A demo, a walkthrough, a proper story. Only if the content genuinely fills it.
• Over 3 minutes. Rarely worth it in the feed. If you have ten minutes of material, that is a LinkedIn newsletter or a video hosted somewhere people arrive on purpose.

The most common mistake is not making videos too long in minutes. It is spending the first fifteen seconds on preamble. “Hi everyone, hope you're having a great week, I wanted to jump on here and talk about something” is fifteen seconds of nothing, and in the feed fifteen seconds is the whole budget.

Do LinkedIn videos need captions?

Yes, and this is the one point in this article where there is no trade-off to weigh. Video autoplays muted in the LinkedIn feed. Hootsuite, citing LinkedIn's own figures in its guide to LinkedIn video, reports that 85% of video is watched without sound. An uncaptioned video is a silent film to the large majority of people who see it.

The accessibility case is more important than the reach case and gets mentioned less. A video without captions excludes deaf and hard-of-hearing viewers completely. It also excludes anyone in an open-plan office, on a train, next to a sleeping child, or reading in a second language, all of whom rely on text to follow along. Publishing without captions is a decision to be unavailable to those people. Frame it that way and the five minutes it costs stop feeling optional.

Two ways to do it, and you should do both:

1. Burned-in captions generated by your editing tool. These always display, look intentional, and survive being downloaded or reshared.
2. An uploaded SRT file. LinkedIn lets you attach a caption file to a native video. This is what screen readers and caption toggles use, and it is machine readable, which burned-in text is not.

Whatever tool generates them, read the captions before publishing. Automatic transcription is good until it meets a product name, an acronym, or an accent, and then it is confidently wrong in a way that makes you look careless.

Talking head or b-roll: what actually performs?

Our dataset records that a post was video. It does not record what was in the frame, so anyone claiming a precise split between talking-head and produced video on LinkedIn is guessing. What we can offer is the pattern we see in the video posts that did best in our cohort, plus what the format constraints imply.

The video posts near the top of our data are overwhelmingly personal and direct. They read like someone talking to you, not like a brand film. Here is one of them:

The practical guidance most people need is not talking head versus b-roll. It is this: the video has to be watchable at arm's length on a phone with the sound off. That single constraint rules out most of what makes corporate video bad.

The first three seconds carry the same job as the first line of a text post. Open on the claim, the question, or the mid-scene moment. “Most people get this wrong” with your face already talking beats a title card, an intro animation, and a greeting. The structural logic is identical to how you open a written post.

How do you write the caption on a LinkedIn video post?

The caption is not a summary of the video. It is the reason someone stops scrolling long enough for autoplay to do its work.

Some patterns from the 1,613 video posts in our cohort, which is a large enough sample to be worth something. Video captions in our data:

• Open with a question 16.6% of the time, against 11.4% across the whole cohort. Video creators ask more questions, and questions correlate with comments.
• Say yousomewhere in the visible hook 43.0% of the time, against 36.4% cohort-wide. Cohort-wide, 47.5% of top-decile posts contain “you” against 32.7% of the bottom half.
• Carry an emoji 14.8% of the time against 6.1% cohort-wide. Cohort-wide the emoji gap between top and bottom posts is stark (24.3% against 2.7%), though emoji are more likely a marker of a certain creator style than a cause of anything.
• Use four or more hashtags 37.8% of the time. That is above the top-decile rate of 25.9%, and hashtag stuffing is one of the few habits that reliably points downward in our data, which we unpack in the hashtag analysis.

A caption structure that works: one line naming what the video shows and who it is for, one line of context or credibility, one line inviting a specific response. Then one to three hashtags at the end, not eight. And write the close deliberately, because a vague “thoughts?” gets vague replies. See how to end a LinkedIn post for the version that gets answers.

How do you start if you hate being on camera?

Most people who avoid video are avoiding one specific thing: their own face and voice played back to them. The format does not require it. Here is a ladder, easiest first.

1. Screen recording with voiceover. Walk through a spreadsheet, a dashboard, a document, a piece of code. Your face is never in frame. This is the single highest-value video most professionals can make, because you are showing something only you have.
2. Whiteboard or paper. Point a camera at a page and draw the idea while you explain it. Hands, not faces.
3. Slides with voiceover. Six slides, ninety seconds, one point per slide. Effectively a carousel with narration.
4. Off-centre talking head. You in frame, but doing something: walking, in front of the thing you are talking about, mid-task. Much easier than sitting straight-on into a lens.
5. Straight-to-camera.The version everyone imagines when they think “video,” and the last rung, not the first.

Practical rules that lower the cost of starting. Film in one take and accept the stumbles, because polished delivery is not what people respond to. Use a phone, not a camera setup. Put the phone at eye height and stand near a window. Do not script it word for word; write three bullet points and talk. And record the same forty seconds twice rather than editing one take for an hour.

The realistic ramp is one video a week alongside your normal text posts, for six weeks. That is enough to get past the awkward phase and enough to compare against your own baseline. Slotting it into a fixed weekly slot works better than deciding each week, because the decision itself is where most people quit.

When is video the wrong format?

A 2.6x over-index is not an instruction to film everything. Video is expensive to make and expensive to fix, and there are cases where it actively costs you.

Anything dense or numeric. Nobody can hold a table in their head from a video. If your point is a set of figures, a comparison, or a definition people will want to re-read, put it in text or slides where the reader controls the pace.

Anything time-sensitive. A reaction to news is worth more within the hour than it is polished. By the time you have filmed, captioned, and uploaded, the conversation has moved. Post the text version now and the video tomorrow if it still matters.

Anything you have not thought through. Writing forces you to notice that an argument does not hold. Video lets you speak past the gap with confident delivery, and readers can tell. A muddled idea is more obviously muddled on camera, not less.

Anything that would break your cadence. Consistency outperforms production value on LinkedIn. If a weekly video means you post once a week instead of four times, the format is costing you more reach than it earns. Work out the frequency you can actually sustain first, then decide how elaborate each post gets.

And a note on repurposing: chopping a webinar or a podcast into feed clips rarely works without a rewrite. Those recordings were made for an audience that chose to be there. The feed did not choose. A clip that starts mid-sentence in a conversation nobody heard the beginning of is the most common bad video on LinkedIn.

How should you measure LinkedIn video performance?

LinkedIn reports video-specific metrics that no other format gives you, and most people ignore them in favour of the impression count.

• Views versus impressions. A view is counted after a short autoplay threshold, so views are much closer to real attention than impressions. Know which one your dashboard is showing you before drawing conclusions, a distinction we untangle in impressions versus reach.
• Completion rate. The most honest number LinkedIn gives you. If it collapses at the twelve-second mark, your opening is the problem, not your topic.
• Comments per view.Our cohort's video median was 27 comments against 12 for all posts. If your videos pull reactions but not comments, the caption is not asking for anything.
• Follower growth on video days. Video is unusually good at converting a viewer into a follower, because they have now heard you think.

For a baseline to compare against, our whole-cohort benchmarks are the median 0.40 engagement per 1,000 followers, 1.27 at the 75th percentile, and 5.95 at the top-decile threshold. The full methodology, cohort rules, and caveats live in our study of 34,000 LinkedIn posts.

What should you do with all this?

LinkedIn video posts carry the strongest format signal in our data: 26.3% of the top decile against 10.1% of the bottom half, across 1,613 videos from 46 creators. That is not a guarantee, and it is not causal. Established creators who make video are different from those who do not, and some of the gap is effort rather than format.

The action it supports is modest and cheap. Upload video files natively instead of linking to them. Keep them between thirty and ninety seconds. Caption every one of them, for reach and because people who cannot hear are part of your audience. Spend your effort on the first three seconds and the caption above the player. Then run one video a week for six weeks and compare completion rate and comments against your own baseline, rather than against anyone's benchmark, including ours.