Carousels lead LinkedIn format rankings on engagement rate in 2026: Buffer measures a 21.77% median for carousels against 7.35% for video and 3.18% for text, and Socialinsider shows 7.00%, 6.00% and 4.50%. Studies that score posts against each creator's own median show a much smaller gap, with video weakest on personal profiles.
That is the headline, and it hides the useful part. Format studies disagree because they measure different things on different accounts. This article compares nine studies side by side, reconciles their conflicts, and ends with a labelled analysis of what a B2B founder should post. It is a companion to the guide on the best time to post on LinkedIn for B2B and builds on the LinkedIn statistics for 2026.
#How do carousels, text and video compare in the main studies?
The table puts each study's numbers in its own units. Do not compare across columns. Compare the order within each row.
| Study | Sample and period | Metric | Carousel / document | Video | Text | Image |
|---|---|---|---|---|---|---|
| Buffer | Not stated, published 19 Mar 2026 | Median engagement rate | 21.77% | 7.35% | 3.18% | 6.52% |
| Socialinsider | 2,450,610 posts, 26,114 pages, 2025 values | Engagement rate on impressions | 7.00% | 6.00% | 4.50% | 5.30% |
| Flux analysis | 7,913 posts, 63 profiles, Jun 2025 to Jun 2026 | Multiple of each profile's median | 1.05x | 1.12x | 0.77x | 1.13x |
| Postiv | 62,820 posts, 1,340 creators, Jan 2025 to Sep 2026 | Median lift on creator median | 0.96 | 0.92 | 0.96 | 1.03 |
| AuthoredUp | 3M+ posts, Mar 2025 to Feb 2026 | Reach multiplier, personal profiles | Higher than video (no value) | 0.86x | Higher than video (no value) | Higher than video (no value) |
| Metricool | 673,658 posts, Jan to Feb 2025 vs 2026 | Engagement rate, Company Pages | 49.52% | Not stated | Not stated | 5.77% |
| The SEO Works | 114 posts from one Company Page, 2025 | Engagement rate incl. swipes | 48.3% | 8.5% | Not tested | Not tested |
The ranking of carousel over video holds in the top three rows by rate, and flips for text and image in the next two. That split, between rate-based studies and creator-normalised studies, is the main reconciliation task in this article.
#Which studies are primary data, and how reliable are they?
Nine underlying studies sit behind roughly 17 pages. Sorting them matters because several articles repeat the same Socialinsider and Metricool figures.
Large first-party datasets:
- Socialinsider analysed organic posts from 26,114 business pages, with a reported 2,450,610 posts in its methodology table, although the introduction says 1.3 million and the date range runs past the publication date of 16 March 2026. Social Media Today, reporting on the same report, gives 1.3 million posts, 16,645 pages and January 2024 to December 2025. Treat the exact sample as unresolved.
- Metricool studied 673,658 posts from 63,108 accounts, comparing January and February 2025 with the same months in 2026. It is a vendor press release that does not define "engagement."
- AuthoredUp reports 3M+ personal-profile posts from March 2025 to February 2026, with a video length subset of 36,946 posts.
- Buffer published a 2026 State of Social Media Engagement report covering millions of posts across platforms with no LinkedIn sample size, reach figures or date range given.
Smaller or independent:
- The 7,913-post analysis of 63 profiles, normalised to each profile's median, with impressions unavailable.
- Postiv, 62,820 public posts from 1,340 creators with a median audience of about 52,700 followers, post type classified by AI, no impressions.
- The SEO Works, 114 organic posts (76 carousels, 38 videos) from one Page with over 50,000 followers, 2025.
Platform statements and press: LinkedIn's own video statements, reported by Social Media Today and Digiday.
Secondary: Pollen, Vulse, ChartLocal, the French summary of Metricool and a general carousel roundup that mixes platforms. They contribute cross-checks, not new data.
#What does Buffer's format benchmark show?
Buffer's table is the most quoted and the least documented. Its median engagement rates on LinkedIn:
- Carousel (document or PDF): 21.77%.
- Video: 7.35%.
- Image: 6.52%.
- Link: 3.81%.
- Text: 3.18%.
Buffer states the carousel lead as 196% more than video, 234% more than images and 585% more than text-only posts. Checking its arithmetic: 21.77 divided by 7.35 is 2.96, which is 196% more; divided by 6.52 is 3.34, or 234% more; divided by 3.18 is 6.85, or 585% more. The figures are internally consistent. Video over image is 12.7%, which Buffer rounds to 13%.
A different Buffer article, on timing, says carousels get "up to 596% more engagement than text-only posts". That is a different number from 585% on the same product's data. The gap is small, but it shows the headline moves between pages and the underlying cut is not published.
What Buffer does not say matters more: no sample size for LinkedIn, no reach or impressions, no date range, no definition of the engagement formula, no polls. A median rate also depends on who posts carousels. If a minority of strong creators and brands post them, the median reflects selection as much as the format.
#What does Socialinsider's 2.45 million-post dataset show?
Socialinsider is the best-documented format source, though its page has internal inconsistencies. Its 2025 engagement rates by format, defined as reactions plus comments plus shares divided by impressions, with a 5.20% overall average:
| Format | 2024 | 2025 |
|---|---|---|
| Native document (carousel) | 6.10% | 7.00% |
| Multi-image | 6.60% | 6.45% |
| Video | 5.60% | 6.00% |
| Image | 4.85% | 5.30% |
| Text | 4.00% | 4.50% |
| Poll | 4.40% | 4.20% |
| Link | 3.30% | 3.25% |
Three observations. Native documents rose from 6.10% to 7.00% (about 14.8% by my arithmetic, while the page text says 14%). Multi-image posts slightly declined and are the second best format. Video rose from 5.60% to 6.00%, which conflicts with other studies that show video falling, a conflict addressed in the next sections.
Socialinsider also reports average impressions per post by follower band. For pages of 10,000 to 50,000 followers: multi-image 1,850, image 1,450, video 1,420, poll 1,255, native document 1,150, text 845 and link 340. For pages of 100,000 to 1 million followers: poll 9,797, text 4,000, multi-image 3,867, video 2,410, image 1,790, native document 1,180 and link 2,765. The reach table reverses the rate table's order. Documents rank first on rate and last or near last on reach in the largest band.
Posting frequency explains part of it. Socialinsider reports average posts per month in 2025 of 2 native documents, 4 videos, 7 images and 1 text, so the sample for documents is thin. Social Media Today flags the same issue: document posts were less common in the sample, which could skew results. The Socialinsider methodology also says accounts under 1,000 or over 100,000 followers were excluded, yet its tables show a 1 to 5K band and a 100K to 1M band.
#What did Metricool find about carousels, polls and video?
Metricool's third annual study is the only large dataset that separates Company Pages from personal profiles on format. From the Netinfluencer summary and the press release:
- Company Pages: carousels earned 49.52% engagement and 1,234.18 impressions per post. Images earned 5.77% and 952.90 impressions. Carousels therefore reached about 1.3 times as many people as images (my calculation) with an engagement rate about 8.6 times higher, a gap that reflects how Metricool defines the rate rather than a literal 8.6x difference in audience response.
- Polls (Company Pages): 3,418.57 impressions per post, nearly three times any other format, though polls are under 1% of posts.
- Personal profiles: video is the most used format at 40.11% of posts, yet it underperforms carousels, images and multi-image content. The highest engagement format for personal profiles was multi-image at 3.71%.
- Frequency vs results: the study's own conclusion is that "adoption has outpaced performance" for video. A French summary reports that carousels and multi-image posts generate up to 11 times more interactions than standard images, and that images were posted six times more often than carousels.
- Visible metrics: likes fell 13%, comments 17% and shares 10% year over year, while clicks rose 5%. Overall engagement rose nearly 14%, a combination the study attributes to "invisible interactions."
The caveats are material. The release does not define how engagement was calculated, and the carousel engagement rate of 49.52% is not comparable to Socialinsider's 7.00%. Different formulas are the likeliest explanation, and Metricool does not publish its own.
#Why did AuthoredUp find video weakest on personal profiles?
AuthoredUp's analysis is the most direct test of video for individuals. In its personal-profile data:
- Video has a 0.86x reach multiplier against the profile's median across formats, described as the only major format below the profile average.
- Median video reach per post fell 36% year over year and median video engagement fell 26%.
- The page does not give numeric values for text, image or carousel, only that these formats "still deliver higher reach."
For Company Pages, the picture reverses. The top 5% of company pages saw a 1.72x video reach multiplier against their baseline, while all company pages saw 1.06x. Top pages reach 62% more viewers with video than the average page.
Video length also matters. From 36,946 personal-profile videos, the reach multiplier against the median video was:
| Length | Reach | Share of videos | Engagement |
|---|---|---|---|
| Over 3 minutes | 1.21x | 11% | 1.17x |
| 90 seconds to 3 minutes | 1.07x | 15% | 1.09x |
| 60 to 90 seconds | 0.97x | 17% | 1.00x |
| 0 to 30 seconds | 0.96x | 27% | 0.91x |
| 30 to 60 seconds | 0.95x | 30% | 0.96x |
The shortest clips are the most common and the lowest reaching. AuthoredUp attributes this to LinkedIn weighting dwell time, which it presents as explanation rather than a tested result. It also reports captions of 20 or more sentences at 1.13x median video reach and captions under 5 sentences at 0.73x.
#What do the creator-normalised studies show?
Two independent studies score each post against its own author's median, which removes audience size and posting habits from the comparison.
The Flux analysis (7,913 posts, 55 profiles in the normalised cut, June 2025 to June 2026) reports image 1.13x, video 1.12x, document/carousel 1.05x and text-only 0.77x. It also reports posts under 200 characters at 0.67x and posts of 500 to 2,000 characters at roughly 1.09x to 1.19x. Engagement there is likes plus comments plus shares, with no impressions.
Postiv's study (62,820 posts) reports median lift of 1.03 for image (63.5% of posts), 0.96 for text (14.9%), 0.96 for carousel (3.9%) and 0.92 for video (17.7%). The comments per like ratio is highest for carousels at 0.447, then text at 0.373, image at 0.334 and video at 0.271. Longer text and image posts did slightly better: lift of 0.98 under 300 characters, 1.00 for 300 to 799, 1.03 for 800 to 1,299, 1.02 for 1,300 to 1,999 and 1.05 at 2,000 or more.
The two disagree. Flux puts text far behind (0.77x), and Postiv puts text at 0.96, nearly level with carousels. Both agree that video is not the winner for individual creators, and neither finds carousels leading. The Flux sample is small and Postiv's is a library of creators with a median audience around 52,700, so neither represents a founder with a few thousand followers.
#What the data disagrees on, and how to reconcile it
1. Do carousels win? Rate-based studies (Buffer, Socialinsider, Metricool, The SEO Works) say yes. Creator-normalised studies (Flux, Postiv) say no. Reconciliation: carousel engagement rates are inflated by mechanics. The SEO Works notes its engagement formula favors carousels because each swipe counts as an interaction, while passive video viewing counts for nothing. Its own figures show carousels at 48.3% against video at 8.5%, from 114 posts. Postiv's comments-per-like ratio of 0.447 for carousels is the one place a conversation signal supports them.
2. Is video up or down? Socialinsider has video up from 5.60% to 6.00%. Metricool says video performance slipped. AuthoredUp says median video reach fell 36% and engagement 26% for personal profiles. These can all be true: rates on impressions can rise while reach per post falls, and Company Pages differ from profiles. Socialinsider's sample is business pages; AuthoredUp's is personal.
3. The number 36%. It appears with three meanings. AuthoredUp: median video reach per post fell 36%. LinkedIn's own figure, reported by Social Media Today: video watch time rose 36% in 2024. Vulse's summary says video views fell 36% year over year, while AuthoredUp also quotes LinkedIn as saying viewership rose 36% and uploads rose 45%. It is possible that LinkedIn's total video consumption rose while per-post reach fell, because more videos compete for the same attention. That is my reading, not any study's finding.
4. Short or long video? AuthoredUp finds over 3 minutes best and 30 to 60 seconds weakest on reach. The SEO Works found videos under 2 minutes averaged a 9.4% rate against 7.0% for videos over 4 minutes. Different metric (reach multiplier against a rate that includes clicks), different account type, 36,946 videos against a few dozen.
5. Where does text rank? Lowest in Buffer (3.18%) and Flux (0.77x), fourth in Socialinsider (4.50%), second in comments per like in Postiv (0.373). Pollen, summarising Socialinsider, shows the same ordering as the source. Text is the format most dependent on the writer, and format-level medians hide that.
6. Are polls good or dead? Socialinsider has polls at 4.20% rate and the highest impressions for pages of 100K to 1M followers (9,797). Metricool has polls at 3,418.57 impressions on Company Pages. But polls are under 1% of posts, and Socialinsider's quarterly figures show poll engagement dropping from 4.50% to 3.60%. Reach from polls does not translate into rate.
7. Sample sizes, dates and definitions. Socialinsider's date range extends past its publication date; Buffer gives no LinkedIn sample; Metricool compares two-month windows; Flux and Postiv exclude impressions. Treat each as a snapshot, not a verdict.
#Does video get distribution advantages that the engagement studies miss?
LinkedIn's own statements about video are the strongest argument for it and the least checkable. Social Media Today reported LinkedIn figures from April 2025: video watch time up 36% year over year in 2024 and video generating 1.4 times more engagement than other formats. It also lists the specs: files up to 5 GB, up to 15 minutes, aspect ratios from 1:2.4 to 2.4:1. The article notes that you cannot post directly to the immersive video feed, which instead features content based on the quality and relevance of existing videos.
The measurement detail matters for B2B. LinkedIn defines a view as at least two continuous seconds with the video at least 50% on screen, and treats videos under two minutes as short form, informally. Digiday's reporting from February 2025 found five creators with average watch times of roughly 15 seconds, and one creator's same video averaging 5.3 seconds on TikTok, 20 on Instagram and 22 on LinkedIn. LinkedIn introduced average watch time to creators the day before that article.
Two implications. A view counts at two seconds, so impressions-based video numbers look flattering next to the 15-second average watch time. And LinkedIn's platform statements are about aggregate viewing, which cannot tell you whether your next video will reach your buyers.
#Shruti's analysis: what the nine studies mean for a B2B founder
This section is my synthesis. It uses only the figures quoted above and adds no new data.
Finding 1: The format ranking depends on whether your audience is a page or a person. Every rate-based winner for carousels comes from business pages (Socialinsider, Metricool Company Pages, The SEO Works). Every study with personal profiles (Metricool personal, AuthoredUp, Flux, Postiv) places video at or below average and carousels in the middle. A founder posts from a personal profile, so the personal-profile evidence is the relevant set.
Finding 2: Images and longer text posts are the most consistent performers for individuals. Image sits at 1.13x in Flux and 1.03 in Postiv, the highest in both. Text is polarised, 0.77x in one and 0.96 in the other, but Postiv shows text posts at 2,000 characters or more with a lift of 1.05, and Flux shows posts of 500 to 2,000 characters at 1.09x to 1.19x. The weak text result in Flux is concentrated in very short posts (under 200 characters at 0.67x). My reading is that "text" is not the problem; thin text is.
Finding 3: Carousels win on conversation, not on reach. Postiv's 0.447 comments per like is the highest of any format, and Socialinsider's reach table puts documents low on impressions in large pages. For a founder whose goal is replies from buyers rather than impressions, that is the right trade, and it suggests the carousel is a reference asset, not a reach engine. The sample is small: only 2,457 carousel posts in Postiv, 3.9% of the data.
Finding 4: Video is a selective tool, not a default. AuthoredUp's 0.86x for personal profiles and Metricool's note that video underperforms are consistent. The one pocket where video shows an advantage is length, with over three minutes at 1.21x reach and 1.17x engagement. A founder with a distinct spoken point of view and the time to produce a real piece has a case. A founder repurposing a 30-second clip does not, on this data.
Finding 5: Format is secondary to what is in the post. Postiv's lift range across formats is 0.92 to 1.03, a spread of 11 points around creator median. Caption length in AuthoredUp's video data moved reach from 0.73x to 1.13x, a larger spread than between some formats. I would not choose a format by its benchmark and then write to fit it. I would choose the format that suits the idea.
A working mix for a B2B founder (my recommendation, not a study result):
- Default: text or image posts of roughly 800 to 2,000 characters, because that is where both normalised studies are strongest.
- Use a carousel when the post is a framework, checklist or comparison the buyer will want to keep, and when you want comments. Do not use one for opinions that read well as text.
- Use video when the point needs your voice, and either go past 90 seconds or accept that the short clip is a brand play, not a reach play.
- Use polls rarely, and only to learn something you will act on, since reach does not convert to rate.
- Use links in the first comment or not at all on a personal profile. Metricool found links cut personal profile impressions by 27% and interactions by 20%, while helping Company Pages by 51% and 41% (via Netinfluencer). The first-comment placement is my practice, not a tested result.
For the evidence on whether founder content converts, see founder LinkedIn content and B2B pipeline data. For production questions, see LinkedIn ghostwriting for founders, and for the effect of AI drafting on reach, the AI-written posts reach data. Timing is covered in the best time to post guide.
#A four week format test for your profile
The method is mine.
- Choose three formats: text, image and one of carousel or video.
- Post each format three times across four weeks, on varied Tuesday to Thursday slots so day does not confound format.
- Score every post against your own median engagement, the lift method used by the two normalised studies.
- Track comments separately from reactions, since comments per like was Postiv's most distinguishing format metric.
- Judge on repeat, not on one outlier. Metricool reports that half of lifetime impressions arrive within 48 hours, so read results after two days and again after a week.
#Methodology
I searched for 2025 and 2026 LinkedIn format studies on 30 October 2026 and read each page directly where the fetch tool allowed. I kept studies that published a sample, a date or a method, and sorted them into first-party large datasets, independent small samples, platform statements and secondary summaries. Nine underlying studies are compared (Buffer, Socialinsider, Metricool, AuthoredUp, Flux, Postiv, The SEO Works, LinkedIn statements via Digiday and Social Media Today, and the Pollen compilation), through 17 pages and articles. Date coverage runs from January 2024 to September 2026.
I did not average across studies, because the metrics are different: median engagement rate, engagement on impressions, a multiple of each profile's median, a reach multiplier, and a rate that includes carousel swipes. Ratios such as 2.96x for Buffer's carousel over video are my arithmetic from the cited tables.
Limits:
- Buffer's LinkedIn sample, date range and formula are undisclosed.
- Socialinsider's page contradicts itself on sample size (1.3 million versus 2,450,610), date range and follower exclusions.
- Metricool is a press release and defines no formula; its figures come via two summaries as well as the release.
- Vulse, Pollen and ChartLocal are secondary and cannot be checked against the originals. A Stacked Marketer article referenced by ChartLocal could not be opened.
- The Moneris roundup mixes platforms and is used only as a caution about cross-platform carousel statistics.
- No study isolates B2B audiences. B2B relevance is inferred from account type.
- The recommendations are mine and are not tested against client data in this article.
#Sources
- Buffer, "Best Content Format on Social Platforms in 2026: 45M+ Posts Analyzed", 19 Mar 2026. https://buffer.com/resources/data-best-content-format-social-media/
- Buffer, "The Best Time to Post on LinkedIn in 2026: 4.8 Million Posts Analyzed", 9 Sep 2026. https://buffer.com/resources/best-time-to-post-on-linkedin/
- Socialinsider, "2026 LinkedIn Benchmarks", 16 Mar 2026. https://www.socialinsider.io/social-media-benchmarks/linkedin?abtest=true
- Social Media Today, "Report shows document posts on LinkedIn see more engagement", 2 Apr 2026. https://www.socialmediatoday.com/news/report-shows-document-posts-on-linkedin-see-more-engagement/816551/
- Metricool, "Metricool's 2026 LinkedIn Study Reveals the Rise of Invisible Interactions as Video Performance Slips", 14 Apr 2026. https://metricool.com/press-release-linkedin-study-2026/
- Netinfluencer, "LinkedIn engagement shifts toward clicks as visible metrics decline, study finds", Apr 2026. https://www.netinfluencer.com/linkedin-engagement-shifts-toward-clicks-as-visible-metrics-decline-study-finds/
- J'ai un pote dans la, summary of Metricool's LinkedIn study, 29 Apr 2026. https://jai-un-pote-dans-la.com/etude-metricool-engagement-linkedin/
- AuthoredUp, "LinkedIn video posts", published 17 Apr 2026, updated 11 Aug 2026. https://authoredup.com/blog/linkedin-video-posts
- The SEO Works, "Data-led analysis of LinkedIn video vs carousels", modified 20 Mar 2026. https://www.seoworks.co.uk/downloads/data-led-analysis-of-linkedin-video-vs-carousels/
- Trevor Fox, "The best day to post on LinkedIn is different for everyone", 18 Jun 2026. https://trevorfox.com/2026/06/the-best-day-to-post-on-linkedin-is-different-for-everyone/
- Postiv, "LinkedIn post study 2026", 22 Oct 2026. https://postiv.ai/blog/linkedin-post-study-2026
- Social Media Today, "Looking to post video to LinkedIn, and tap into rising video engagement in the app?", 28 Apr 2025. https://www.socialmediatoday.com/news/linkedin-video-posting-tips-specs-2025/746552/
- Digiday, "Why LinkedIn is spotlighting the average watch time metric to support its video push", 5 Feb 2025. https://digiday.com/media/why-linkedin-is-spotlighting-the-average-watch-time-metric-to-support-its-video-push
- Pollen, "LinkedIn Engagement in 2026: What the Data Actually Shows", 1 Jul 2026. https://www.justpollen.com/blog/linkedin-engagement-study-2026
- Vulse, "LinkedIn content benchmarks 2026: which post format drives most engagement", 6 Apr 2026, updated 12 Oct 2026. https://vulse.co/blog/linkedin-content-benchmarks-2026-which-post-format-drives-most-engagement
- ChartLocal, "Texts Reap Benefits Over Videos", 6 Apr 2026. https://chartlocal.com/2026/04/06/texts-reap-benefits-over-videos/
- Moneris, "Why you need carousels for your social posts", 9 Jun 2026. https://www.moneris.com/en/blog/posts/growth-strategy/why-you-need-carousels-for-your-social-posts