LinkedIn algorithm: video vs carousel vs text impressions

The search "LinkedIn algorithm video vs carousel vs text impressions" is usually a request for a ranking: which format LinkedIn will show more. LinkedIn has not published that ranking. It has published three facts that decide how you should read the numbers you already have. An impression is a display on screen, and combined post analytics call the count an estimate. Retrieval includes the format of the post in the item prompt. Long dwell is scored against similar posts of that format, not against a universal second-count.

A 40-second native video that people skip at second four is a skip on video. A 140-word text post that people finish in twelve seconds can still be a long dwell on text. A six-page PDF that people click through because the point is on page five is a long session on a document. Those three clocks are not interchangeable. Treat a vendor chart that crowns "video 5× text" as someone else's sample, mixed years, mixed Pages and profiles, and usually a different formula in the denominator.

Boykli writes text posts and designs a 2–8 slide PDF for a document post. It does not upload a native video file. A YouTube URL becomes a teaser and, if you want slides, a separate PDF. Native video still lives in LinkedIn's own composer. The system that assigns displays, independent of which object you chose, is the LinkedIn algorithm guide. This page is the format choice in front of that system.

Logged-in visitors can browse Radar and Algorithm with no plan. Writing, carousels, Voice Agent, publishing, and scheduling need a paid seat. BYOK is $29 a month. Managed is $88 a month. Prices are the public list on boykli.com as of 26 October 2026.

Three objects, not three versions of a post

Text is a commentary post. LinkedIn Help caps it at 3,000 characters and points you at an article when you need more. The first two lines are what a skimmer sees before "see more." There is no player and no page counter. Dwell is reading. A skip is leaving those two lines.

Carousel, in the feed people mean, is a document post: one PDF, PPT, or Word file that LinkedIn paginates. Upload and share documents sets the ceiling at 100 MB and 300 pages. You add a title, then a description. Members who can see the post can download it as a PDF. After publish you can edit the description. You cannot replace the file. One page is on screen. The first page is the card. How square 1080×1080 and portrait 1080×1350 sit in that card is LinkedIn carousel dimensions. A multi-image photo grid is a different object, covered in multi-image vs PDF carousel.

Native video is a file you upload in LinkedIn's composer. Video sharing troubleshooting lists 3 seconds to 15 minutes on a member post (2 seconds from the mobile app), 75 KB to 5 GB, 256×144 to 4,096×2,304, aspect 1:2.4 to 2.4:1, 10 to 60 fps, and types including MP4, MOV, AVI, WebM, MKV, and WMV. It notes that a ProRes file may upload and then fail to process. Video specifications for Pages and Career Pages caps Page video at 10 minutes and says AVI, QuickTime, and MOV are no longer supported there. MP4 is the export that both pages accept. One video per post. Autoplay in the feed is muted until the member unmutes. A YouTube link pasted into the composer is a link card, not this object.

Boykli's publish call sends text, an optional image, or a PDF document to the personal profile you connect. It does not send the video file. If the experiment needs native video, record and upload it on LinkedIn, then use Boykli for the text and document legs of the same topic.

Specs and what an impression is counting

An impression is still a display of that object. It is not a completed read, a completed swipe, or a completed watch. Members reached is the distinct people. One person can produce several impressions. In-network versus out-of-network is the split you can audit. None of those three numbers is "video wins."

Why a longer clock is not a higher impression tier

Fengyu Zhang, Mohit Kothari, and Birjodh Singh Tiwana (1 October 2024) define dwell as time on a post, separate from likes and comments, because a large share of feed use is reading without a reaction. A view shorter than a skip threshold they call Tskip is a negative. A single "long dwell" cutoff failed. Their illustration is 30 seconds: that duration is a different event on an image than on a video, and the median for a format moves over months. The model they shipped asks whether this member will dwell longer than a percentile of similar posts, grouped by content type, creator type, and distribution method, with the percentile recomputed daily.

On 12 March 2026, Hristo Danchev described the retrieval prompt for a post. The text includes format, author name, headline, company, industry, engagement counts, article metadata, and the post text. Format is an input, not a bonus multiplier published as "video +40%." Ranking then predicts skip, click, long dwell, like, comment, and share for one member at a time. The same day's product note from Tim Jurka said the feed would show less engagement bait and less video that has nothing to do with the text, used to chase distribution.

So the practical reading is:

• A video that holds people because they are watching the thing the caption promised can produce long dwell for video.
• A PDF that takes 90 seconds because the type is tiny and the point is on slide seven can produce a long session that still looks like a skip on the first page.
• A text post that states the rule in 160 words can finish in well under 30 seconds and still sit above the text median.
• A stock clip under a caption about a pricing change is the mismatched-video case Jurka named.

The bait-versus-value test for the caption, independent of format, is engagement bait vs deep value.

What the roundups mix

Sponsored video in Campaign Manager is a different viewability meter. Do not add ad plays to organic impressions and call the sum reach.

Which format for which job

Pick the object that matches how the idea is consumed. Then measure that object against itself.

Text when the point is a decision, a number you measured, a constraint, or a sentence you can defend on a call. The reader should be able to leave with the rule after the opening. Short, medium, and long modes on Create posts sit inside the 3,000-character cap. A hook pack that could sit on any author's post is a skip prediction with nothing to contradict it.

PDF carousel when the point is a sequence: definition, then steps, then close. One idea per page. Six to eight pages for a feed swipe. Boykli stops the form at 8. A 40-page leave-behind is a file you upload yourself. The first page has to carry the post, because that is the card. The caption above the file is still a post. Do not hide the conclusion until page seven and call the extra seconds a ranking win. How the deck is built from a topic, an idea, or a file is the LinkedIn carousel generator.

Native video when motion is the evidence: a product walk-through, a whiteboard, a screen recording of the bug, a face saying the sentence you would not type the same way. Put the claim in the first two seconds and in the caption. Add captions on the file; autoplay is muted. Keep the clip inside the length you would watch on a commute, not the 15-minute ceiling. If the video is already on YouTube and the LinkedIn job is a teaser plus a swipe deck, that is two Boykli passes, not an upload of the YouTube file.

Do not film a talking head that restates a caption the reader already finished. That is the mismatched-video shape with extra production.

A three-week format test on one profile

Do this on a personal profile. Boykli publishes only to the personal profile you connect with OAuth. A company Page uses Page analytics and a 10-minute video cap. This product does not queue to it.

Use one topic you would sign: a change you made, a number you measured, a sequence a reader can copy. Write the claim in one sentence and keep that sentence stable.

1. Week 1, text. Draft the claim as a commentary post. Opening line is the specific fact, not a mood. Publish or queue it at the clock you already use. Schedule is posting-day toggles, one time of day, one timezone. Do not change the hour for this test.
2. The next day. Open LinkedIn's discovery block. Write down impressions, members reached, in-network percent, out-of-network percent. Note comments that add a fact. Ignore a like-to-follower ratio.
3. Week 2, PDF. Same claim, as a 6-slide deck. Page 1 is the claim. Pages 2–5 are the steps. Page 6 is the close the text post already contained. Square or portrait, type in the 220px safe zone. Caption is four short lines, not a second copy of every slide. Queue it on a different weekday, same clock.
4. The next day. Same four discovery numbers. If LinkedIn shows document-specific figures on that post, record those too. Do not invent a swipe count LinkedIn did not print.
5. Week 3, native video. Same claim, as a file you upload in LinkedIn's composer. First two seconds say the claim on screen, with captions. Length is the walk-through, not the archive. Commentary matches the text post's point. Same clock, a different weekday.
6. The next day. Same four discovery numbers, plus whatever video meter that post actually shows (view count, watch time, if present). Treat autoplay displays as displays.
7. Compare rows, not trophies. Put the three posts in a table with format, impressions, members reached, in-network share, out-of-network share, and one qualitative note ("people asked about step 3," "nobody finished the clip"). A PDF with a pretty reaction rate on 400 impressions is not a win over text that reached 3,000 members. A video with a high impression count and a 4-second average watch is a skip on video.

Hold topic and clock still. If you change the idea, the hour, and the format in the same week, the delta is unreadable. After the three rows exist, pick the format that produced the comments you would want on a sales call, then repeat that format on the next three ideas. The Algorithm page will not import the table. LinkedIn remains the meter.

A fourth object, a YouTube link card, is a different test. Run it only if the job is to send people off LinkedIn. Put the point in the commentary so the on-feed dwell is not an empty preview.

FAQ

Does LinkedIn's algorithm prefer video, carousels, or text in 2026?

It retrieves and ranks each post for one member, with format inside the item prompt and long dwell scored against similar posts of that format. It has not published a preference order or a multiplier. Jurka's March 2026 note limits bait and mismatched video. The June and September notes limit generic and recycled text. A specific post in any of the three formats can still travel. A thin post in a fashionable format can stay in-network.

Why do my videos show more impressions than my text posts?

Video autoplays. A display of a muted player is still an impression. Members reached, out-of-network share, and the watch meter on that post (when LinkedIn shows one) tell you whether people stayed. Compare those, at least a day later, against a text post on the same topic. A higher impression count with a four-second watch is not the same outcome as a finished read.

Do PDF carousels get more reach because people swipe?

Swiping keeps the member on the document. Combined analytics still count displays, members reached, and the network split. LinkedIn has not published a page-turn bonus that unlocks a wider impression tier. Design page 1 as the only page a skimmer sees. Hide the point on page six and you bought time, not a published ranking feature.

Can I schedule native video with Boykli?

No. The queue sends text, an optional image, or a PDF document. Native video is an upload in LinkedIn's composer. You can still draft the commentary in Create posts, copy it, and attach the file there. A YouTube URL in Boykli becomes a teaser post and, separately, a PDF if you paste an outline on the carousel page.

Is a 15-minute video better for dwell than a 90-second one?

The 15-minute figure is a member-post ceiling, not a dwell target. Long dwell is relative to similar videos. A 15-minute file that loses people at second six is a skip. A 90-second walk-through that matches the caption can sit above the video median. Pages are capped at 10 minutes on LinkedIn's Page specification.

Will posting all three formats in one day manufacture impressions?

Each post is a separate retrieval item. Repeat generic posts are what the June and September 2026 notes describe as less likely to travel past the immediate network. Three formats of the same empty lesson are still empty. One claim, three weeks, one clock, is a test. Three thin posts on Tuesday is volume.