I made a two-part comedy video about me arguing with an AI agent. It was good. I mean that as plainly as I can say it about my own work, because the data backs it up: the people who watched part two stayed for 2 minutes 27 seconds of a 4 minute 36 second video. That is a 53% watch-through on a cold feed, which is the kind of retention you do not get by accident. The people who got in, stayed.
It reached 15 people. Fifty saw it at all.
That gap is the whole story, and it is the most useful thing I have learned about LinkedIn video in a while. When a genuinely good video dies, almost every time it died at the distribution gate, not the quality gate. The machine never showed it to anyone, so its retention had nothing to retain. I went and figured out exactly where it died, and this is the reconstruction.
The incident: a 53% watch-through that reached almost nobody
Here are the real numbers off LinkedIn's own post summary, from an account with 815 followers.
Episode one ran 3:56. It pulled 309 impressions, 64 video views, six reactions, four comments, and an average watch time of 56 seconds. The longer one, episode two, ran 4:36 and pulled 50 impressions, nine video views, two reactions, no comments, and that 2:27 average watch. Across both posts there were zero reposts, zero sends, and basically zero saves, where episode two squeaked out a single one. Neither post drove a profile view or a follower.
Sit with the contrast. Watch-through is the only honest measure of whether a video is working, and by that measure the better video reached one-sixth of what the weaker one did. Retention went up while reach fell off a cliff. That is the fingerprint of a distribution failure, and once you see it you cannot unsee it.
“When a genuinely good video dies, almost every time it died at the distribution gate, not the quality gate.
”
Timeline: where the reach actually leaked
Walking it in order, here is where each viewer was lost.
First leak is the muted scroll. On episode one, 64 of 309 people who had it in front of them counted as a view. That is a 21% view rate, which means roughly four out of five people scrolled past before it registered at all. The second episode was 18%. LinkedIn autoplays muted, so the first frame is the thumbnail, and a weak first two seconds gets skipped before the joke ever starts.
After that comes the cold-start cap. Both episodes got shown to a small slice of the network, the platform watched the early signals, and the signals were thin. There were no early comments worth the name and almost nothing in the way of saves or shares, so neither post earned expansion past the first ring. The newer one had it worse because it went up a few hours after another post of mine, and stacking posts close together splits the test pool and throttles whatever lands second.
Then the silent multipliers never fired. Saves, sends, and reposts are the heavy signals now, and I had essentially none of them. A like is cheap, but a save says "I will need this again" and a send says "this is so my coworker." Without any of those, the post had no fuel to escape the people who already follow me.
Last leak is the one that bugs me a little. People watched, some stayed for half the video, and not one of them got curious enough to click my name and find out who built this. No profile views on either post. The skit entertained, but it never said "go see who made this."
Root cause: the golden-hour cold-start gate
The video did not fail. It failed a test it was never built to pass.
When you post, LinkedIn shows it to a small sample of your network and watches the first 60 to 90 minutes. This is the golden hour, the cold-start test. Strong early signals, especially dwell time, real comments, saves, and shares, earn the post expansion out to second and third-degree reach. Weak early signals cap it where it sits. Episode two went into that test with a small follower base, a muted weak first frame, and another recent post eating its sample. It got capped at 50 and never recovered, no matter how well the nine people who actually watched it behaved afterward.
The 2026 feed makes this stricter, not looser. It runs on an interest graph now, a recommendation model that matches a post's topic to each viewer's professional interests instead of just pushing it outward by raw engagement volume. The practical effect is that engagement from people whose interests line up with your topic carries real weight, and engagement from random off-topic accounts carries almost none. Reach is gated on the right people reacting in the right window, not on a vanity count.
The demo below lets you run that golden-hour test yourself. Move the early-signal sliders and watch where the cold-start gate either expands the post or caps it. It is the clearest way I have found to feel why a 53% watch-through can still die at 50 impressions.
This also explains the part that looks like a paradox. Dwell time and watch-through are real ranking inputs, so my retention should have helped. It could not help, because retention is a second-stage signal. You only get to spend it on the audience the cold-start gate already let in. I was winning the retention fight inside a room of 50 people the platform had locked the door on.
What I changed: the legitimate version of "get people to engage early"
The obvious thought, the one a guy slid into my DMs offering, is to manufacture the early signals. Buy the engagement, or join a pod where everyone likes everyone's posts on cue. I looked hard at whether that works, because the underlying instinct is correct: early on-topic engagement in the golden hour is the single biggest lever there is. The bot version of that instinct is the problem.
The honest comparison sits in one table.
| Lever | Buying engagement / off-topic pod | A real golden-hour circle |
|---|---|---|
| Signal LinkedIn sees | Velocity spike from unrelated accounts, generic repeated comments | On-topic comments from real peers with genuine ties to you |
| Distribution weight | Near zero; off-topic engagement is discounted on the interest graph | High; matched-interest comments are the heaviest early signal |
| Risk | Silent reach throttling, escalating to restriction or shadowban | None; this is the behavior the platform is built to reward |
| Cost | Monthly pod fee or per-engagement spend, for a number that does not move reach | Free; it is five real people you already know in your niche |
The legitimate version of the exact same mechanic is a small, real, on-topic circle. Five genuine peers in your niche, the AI builders and creative-tech operators and DTC operators who actually care about what you posted, who see it early and leave a real comment in the first hour. That is not a pod, it is a handful of people who would have commented anyway, nudged to do it while the golden-hour timer is running. Five real peers commenting in the first hour beats 200 bought likes every single time, because the platform is measuring whether the right people stopped to engage, and these are the right people.
The white-hat moves around that are simple. Post in your audience's golden hour and reply to every comment inside the first 60 minutes, because fast author replies lift visibility. Write the post copy to provoke a comment instead of a cryptic inside joke. Drop the first comment yourself with a question or a "tag the engineer who does this." Put any external link in the first comment, never the post body, because outbound links route people off-platform and cost reach.
What I would do differently: the craft fixes that earn the gate
The cold-start gate is the diagnosis. Craft is the treatment, and none of it requires changing my actual production format, which is just me talking to Claude for half an hour and cutting it down. My unfair advantage there is speed, so the move is to ship more short ones, not fewer long ones. This is the same instinct behind keeping a weekly cadence I can actually sustain: volume you can hold beats a heroic post every two weeks.
Length is the biggest single lever. A four-and-a-half-minute video on a cold feed is fighting physics, because 63% of LinkedIn clips run 15 to 60 seconds and comedy clusters around 58 to 64. So the default cut goes to 45 to 75 seconds. Export full vertical 9:16 at 1080 by 1920, because vertical is the only format eligible for both the main feed and the dedicated vertical video tab, and square and landscape get deprioritized on mobile.
Then the open. Cold open mid-conflict, first frame is the AI already panicking, no intro and no title card. Put a stakes line on that first frame as on-screen text, since it doubles as the muted thumbnail. Make the joke survive mute by rendering the AI's replies as distinct on-screen text, because around 80% of LinkedIn video is watched with the sound off, and a back-and-forth that only works with audio is invisible to four out of five viewers. Burn in clean captions, which lift watch time meaningfully over raw auto-captions.
The fix for my zero-profile-views problem is the ending. End every skit on one real, dryly delivered takeaway. The bit is the bait and the lesson is the hook, because a concrete reusable takeaway is what earns the save, and a takeaway that proves I actually know the subject is what makes someone click my name. This is the same reason grounding AI-assisted content in something real works: the substance is what travels, and an entertaining shell with nothing underneath gets watched and forgotten.
Timing closes it out. Post Tuesday through Thursday, 10am to noon in your audience's time zone, the same slot every week, with Wednesday the strongest single day. Space posts at least a day apart, because episode two cratered partly from going up a few hours after another post and splitting its own test pool.
There is a bigger pattern under all of this that I keep relearning. The work being good is necessary and nowhere near sufficient, which is the uncomfortable lesson sitting underneath the case for an author brand over pure scale. Distribution is its own craft, and on LinkedIn in 2026 that craft is mostly about the first 90 minutes. It is the same separation I run in my own work when I let a second set of eyes catch what I cannot see in my own draft: making the thing and getting the thing seen are two different jobs, even when the team is one person. Treating distribution as a craft to learn, the way I learned the editing and the writing, is also just part of holding all the jobs at once when you are the only person on the file.
The short answer, for anyone searching this
If your LinkedIn video gets no reach despite being good, the cause is almost always the golden-hour cold-start test, not your content quality. LinkedIn shows a new post to a small slice of your network for the first 60 to 90 minutes and only expands reach if early signals are strong, with dwell time, real comments, saves, and shares carrying the most weight. A small follower base, a weak muted first frame, posting too close to your last post, and getting no saves or shares all cap the post at that first ring before retention ever matters. The fix is craft plus a small real circle of on-topic peers who engage in the first hour, not bought engagement, which the platform detects and silently throttles.
FAQ
Why does my LinkedIn video get no reach even though it's good?
Almost always because it failed the golden-hour cold-start test, not because the content was weak. LinkedIn shows a new post to a small sample of your network and watches the first 60 to 90 minutes. If early signals are thin, especially saves, shares, and real comments, the post gets capped at that first ring before its watch-through can do anything. I had a video with 53% watch-through that reached 15 people, which is the cold-start gate, not a quality problem.
What is the LinkedIn golden hour and why does it decide reach?
The golden hour is the first 60 to 90 minutes after you post, when LinkedIn shows the content to a small slice of your network and measures the early signals. Strong dwell time, fast real comments, saves, and shares earn expansion out to second and third-degree reach. Weak signals cap the post where it started. Almost all of a post's reach ceiling is set in that first window.
Do LinkedIn engagement pods or buying engagement actually work?
No, and they carry real risk. LinkedIn detects coordinated engagement by velocity spikes, network proximity, and semantic analysis of repetitive comments, then silently throttles reach while your like count holds. The 2026 interest-graph feed also discounts off-topic engagement to near zero, so cross-industry pod likes do not move distribution. You pay for a number that does not travel and you risk escalating restriction.
What is the legitimate version of an engagement pod?
A small circle of real, on-topic peers in your niche who genuinely see your post early and leave a real comment in the first hour. That is the same mechanic the pod fakes, except the engagement is from people whose interests match your topic, which is exactly what the interest-graph feed rewards. Five real peers commenting in the golden hour beats 200 bought likes.
How long should a LinkedIn video be in 2026?
For cold reach, 45 to 75 seconds. About 63% of LinkedIn clips run 15 to 60 seconds and comedy clusters around 58 to 64. Longer videos only perform with a warm audience that already follows you. If you can make video fast, ship more short ones rather than fewer long ones.
Why do people watch my LinkedIn video but never view my profile or follow me?
Because the video entertained but never gave them a reason to find out who made it. End every video on one real, dryly delivered takeaway that proves you know the subject. The takeaway is what earns the save, and a takeaway that demonstrates expertise is what makes a viewer click your name. I had videos with strong watch-through and zero profile views, and the missing ending was the cause.
Sources and specifics
- The post analytics are my own LinkedIn post summaries, pulled 2026-06-20, from an account with 815 followers. Episode one logged a 3:56 runtime, 309 impressions, 64 views, and a 56s average watch. The longer episode two logged a 4:36 runtime, 50 impressions, 15 members reached, nine views, a 2:27 average watch (~53% watch-through), one save, zero sends, zero reposts, zero profile views, and zero followers gained.
- The golden-hour cold-start mechanic, dwell and watch-through as ranking inputs, and saves/shares as the heavy multipliers: visla.us, gracker.ai, and medium.com/predict.
- The 2026 interest-graph feed weighting topic-matched engagement, and the pod/bot detection by velocity, network proximity, and semantic analysis: linkboost.co, podawaa.com, and connectsafely.ai.
- Length distribution (63% of clips run 15 to 60 seconds), vertical-format reach, caption watch-time lift, and posting windows: opus.pro, teleprompter.com, and socialmediatoday.com on the dedicated vertical video tab.
- Precise percentages from marketing and SEO sources are directional, not figures published by LinkedIn directly. The consensus picture is consistent across sources and matches my own post data.