Why LinkedIn carousels get 3x higher dwell time than single images

The line "LinkedIn carousels get 3x higher dwell time than single images" is a search, and it is a claim. LinkedIn has not published that multiplier. Dwell is time on a post. A document post can hold someone through several pages. A single image can be understood in a glance. Those are different clocks. They are also scored inside their own format. A PDF that simply lasts longer is not a published key that triples distribution.

Boykli builds the document side of this choice: a 2 to 8 slide PDF, square or portrait, plus the caption. A single photograph is a separate create. This page is the measurement behind the 3× line: which number is an engagement rate, which number is a second-count nobody can export, and how to test the two objects on one personal profile.

How impressions are assigned is the LinkedIn algorithm guide. How a photo grid differs from a PDF is multi-image vs carousel. Video, text, and document impressions sit in video vs carousel vs text.

What the 3× line is counting

Dwell time is how long a member stays on a post. Fengyu Zhang, Mohit Kothari, and Birjodh Singh Tiwana (1 October 2024) separate it from likes and comments, because a large share of feed use is reading. A view shorter than a threshold they call Tskip is a skip. A single "long dwell" cutoff failed their tests. 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. The model they shipped asks whether this member will dwell longer than a percentile of similar posts, grouped by content type, with the percentile recomputed daily.

On 12 March 2026, Hristo Danchev described the retrieval prompt. Format is one of the fields in that prompt, next to the post text, the author, and engagement counts. Ranking then predicts skip, click, long dwell, like, comment, and share for one member. Format is an input. It is not a public bonus of "carousel × 3."

Creator analytics do not show dwell. Combined post analytics show impressions (an estimated display count), members reached, reactions, comments, and reposts. Watch time is a video meter. There is no page-seconds export for a PDF and no glance-seconds export for a JPEG. Anyone quoting "15–20 seconds vs 8–10 seconds," or "30–60 seconds vs 5–10 seconds," is using a chart LinkedIn did not publish. Fifteen over eight is about 1.9×. Thirty over ten is 3×. Sixty over ten is 6×. The same blogs rotate those bands and call all of them the algorithm. Treat the band as a story about swiping. Record the fields your post screen actually shows.

The number that does exist in a large public sample is an engagement rate, and it is not 3×.

The company-page rates, beside the dwell story

Socialinsider's LinkedIn benchmarks (page checked 27 October 2026) publish engagement rate by impressions: total of reactions, comments, and shares, divided by impressions, averaged by format. The yearly table:

Seven divided by 5.30 is about 1.3×, for 2025, on company pages, on reactions plus comments plus shares. It is not a dwell ratio. The same page's quarterly note for Q3 2026 lists native documents at 6.40% and images at 4.70%, about 1.4×. Multi-image sits at 6.40% in that quarter, level with documents. The platform average they state for the yearly view is 5.20%.

The page's own sample language has moved. The opening cites 1.3 million business posts. The methodology block describes 2,450,610 organic posts from pages between 1,000 and 1 million followers, and a note that the yearly "2026" framing was 2025 data at first publication, with quarterly updates after that. Social Media Today's 2 April 2026 write-up of the report used 1.3 million posts from 16,645 pages, January 2024 through December 2025, and flagged the skew: documents are rarer than images, so the average can move because of who bothers to upload a PDF. Socialinsider's frequency table matches that rarity. In 2025, pages averaged about 2 native documents a month and about 7 image posts a month.

Impressions on the same report do not hand the reach trophy to documents once a page is past the smallest bracket. Average impressions per post, by follower band:

On pages under 10K followers, the document's average impression count beats the single image. From 10K up, multi-image leads, and the single image beats the document in the 10–50K and 100K–1M bands. Average likes follow the same split. In the 10–50K band the report lists 30 likes on a native document, 65 on an image, and 65 on a multi-image post. A higher engagement rate on fewer impressions is a different outcome from "3× the attention, then 3× the distribution."

These rows are company pages. A personal profile is a different account type. Use the table to retire a fake multiplier. Use your own pair of posts to choose a format.

Why a document can still hold someone longer

The mechanism is ordinary, and it does not require a secret 3× weight.

A single image is one file. Share photos on LinkedIn caps that upload at 5 MB in the photo article, with a recommended width of 1080 pixels and an aspect window from 3:1 to 4:5. The whole picture is the post. A reader who gets the chart, the face, or the product in two seconds has finished the media. If those two seconds sit under Tskip for images, the ranker records a skip even when the picture was clear. If the picture is the kind of thing people stop on, the same two-to-eight seconds can sit above the image median. Long dwell for an image is "longer than similar images," not "as long as a six-page PDF."

A carousel, in the feed people mean in 2026, is a document post. Upload and share documents allows one PPT, PPTX, DOC, DOCX, or PDF, up to 100 MB and 300 pages. LinkedIn shows a page at a time, in file order, with the document title. Moving from page 2 to page 3 keeps the member on that post. Anyone who can see the post can download the PDF. After publish you can edit the description. You cannot replace the file. The pixel size is your choice. Square 1080×1080 and portrait 1080×1350 are the canvases that stay readable. The measurement is LinkedIn carousel dimensions.

That extra time is real when each page adds a fact the previous page did not contain. It is padding when slide 4 restates slide 2 in a smaller font. Zhang's team rejected a universal second-count precisely because a format that is slow by construction would win a fixed cutoff. Inside documents, a six-page PDF that people finish can sit above the document median. A twelve-page PDF that loses them on the cover is a skip among documents. The cover is the first screen. Tiny type on a 16:9 deck, cropped to a short strip on a phone, produces a long file and a short visit.

A swipe is not a public engagement event unless the analytics screen you are looking at shows it and you choose to count it. The standard rate above does not include swipes. Nine page turns are not nine likes.

When the single image is the better object

Use the image when one frame is the argument. A shipping screenshot with the number in the picture. A room, a face, a whiteboard that does not need a second page. A chart someone can read without a legend on slide 2. The photo help article puts the emphasis on that frame, and there is no page counter telling a skimmer the post is "8 pages" before they have seen the point.

Use the PDF when page 3 is nonsense without page 2. A definition, then the steps, then the close. A before-and-after that only makes sense in order. A checklist someone will download. Socialinsider quotes practitioners on that save behavior: a document works when it is a framework a reader can keep. The download is a product feature of the document post. It is not a dwell score.

Do not export a slide deck as one flattened JPEG to chase the image format, and do not slice a single chart into eight slides to chase the 3× line. The first throws away order. The second manufactures time. LinkedIn's March 2026 feed note from Tim Jurka said the feed would show less engagement bait and less inauthentic distribution. A carousel whose only job is to slow the thumb is the document version of that bait. The caption test, separate from format, is engagement bait vs a post that already contains the point.

A paired test on one profile

Run this when a format debate is about to set the next month. One claim. Two posts. Same profile. Different days.

1. Write the claim in one sentence you would say out loud. "We cut the Friday deploy from 40 minutes to 11, and the rollback path is the script in the repo" is a claim. "Productivity tips" is a label.
2. If a later frame is useless without the earlier one, build the PDF. If one picture carries it, build the image. Do not make both just to feed a spreadsheet when the job is obvious.
3. For the PDF: one idea per page, cover states the result, square or 1080×1350, five pages unless the argument needs six to eight. On Boykli, the carousel page takes a topic up to 2,000 characters and returns the PDF, a document title, and a caption. Read every slide. A number the model invented is your post once LinkedIn has the file, and the file cannot be swapped. How a topic becomes that PDF is LinkedIn carousel generator.
4. For the image: one frame, at least 1080 wide, inside 3:1 to 4:5, under the photo article's 5 MB line. Put the number on the picture. Add alt text. Assume you cannot crop it after publish.
5. Put the useful point in both captions. A reader who never swipes, and a reader who never expands the image, should still leave with the result. Preview the first lines on a phone. That check is LinkedIn post preview.
6. Post on different days. A second post inside 24 hours is a weaker comparison, for the reason in the algorithm guide. The personal-profile queue for a PDF you already saved is how to schedule a PDF on LinkedIn. Boykli does not turn a single generated image into a document post.
7. The next day, write down impressions, members reached, in-network percent, out-of-network percent, reactions, comments, and reposts. Engagement rate for the test is (reactions + comments + reposts) / impressions. You will not see dwell. Use two proxies you can actually read: members reached, and whether comments quote a specific page or the picture. "Slide 4 is the part we got wrong too" is evidence someone stayed. "Great post!" is not.
8. Keep the format that moved members reached for that kind of claim, and that produced comments about the substance. One pair is a sample of two. Three pairs on the same profile beat any downloaded 3×.

Company-page averages above are a map of other people's Pages. They are the wrong denominator for a founder posting from a personal profile with a few thousand connections.

How to spend the extra seconds on purpose

If the job is a sequence, design for a finish, not for a longer file.

• The cover is a result, a number, or the rule. A title that says "A thread" spends the first screen on a label.
• One claim per page. Five finished pages beat twelve padded ones, because the reader can see the end, and because Boykli stops the form at 8 for that reason. Three hundred pages is LinkedIn's ceiling, not a feed length.
• Type that survives a phone. Portrait 1080×1350 gives a short stack of lines room. A 1920×1080 board becomes a strip. The 220px safe zone and the type ramp are in the dimensions guide.
• The last page restates the rule in one line. A "follow for more" closer trains people to leave on the page before it.
• The caption carries the same result, so the post still works for someone who reacts to page 1 and moves on. Hiding the point until slide 7 makes the on-feed read look like a stall.

Brand kit (canvas, accent, logo, handle) is applied on the carousel page, not in Settings. The layouts are a separate choice from the dwell question. A Neo-Brutalist cover and a quiet cover are the same document post once LinkedIn paginates the PDF.

Logged-in visitors can browse Radar and Algorithm with no plan. Writing, carousels, images, 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 27 October 2026.

FAQ

Do LinkedIn carousels really get 3× the dwell of a single image?

LinkedIn has not published dwell seconds by format, and creator analytics do not show dwell. The 3× figure in tool roundups is usually a seconds band (often 30–60 against 5–10) with no primary table under it. Socialinsider's 2025 company-page engagement rates are 7.00% for native documents and 5.30% for single images, about 1.3×, on reactions, comments, and shares divided by impressions. That is a different metric, on Pages, and it is not a distribution multiplier you can install.

Does each swipe count as engagement?

The public engagement rate used in the Socialinsider table is reactions, comments, and shares over impressions. A page turn keeps the member on the document, which is time-on-post, the thing Zhang's team models as dwell. It is not automatically a like, and LinkedIn has not published a swipe-to-impression conversion. Count the fields on the post. Name the formula in the note next to it.

Why would I make a carousel if the multiplier is not 3×?

Because the argument has an order, or because you want a file people can download. Documents led Socialinsider's 2025 engagement-rate table, and on pages under 10K followers they also led average impressions against single images. On larger pages, multi-image posts led impressions. The format follows the job. A sequence that was ignored as one crushed JPEG and read as a PDF is a format result even when no one hands you a 3×.

How many slides earn the time?

LinkedIn allows 300 document pages. A feed swipe that gets finished is a cover plus a handful of claims. Boykli's form allows 2 to 8 slides and defaults to 5. Extra pages that repeat the cover add file length. Inside the document group, long dwell is a percentile of similar documents, so a slow, empty deck is a long file and a skip.

Can I see dwell time anywhere in LinkedIn?

No member analytics screen publishes dwell for an image or a document. Impressions, members reached, and the reaction counts are the audit. Video posts can show watch time. That meter does not transfer to a PDF. Boykli does not record impressions or dwell either. The test above uses the numbers LinkedIn already shows you the next day.

Will a scheduled carousel get less dwell than one I post by hand?

The document is the same file either way. What changes is who sends it and when. A queued PDF still has to be a specific document. Scheduling mechanics, separate from this dwell claim, are scheduled posts and reach. The file still cannot be replaced after it is live, so the read-every-slide step happens before the slot, not after.