Your carousel flopped: three gates, and the number that tells you which one failed

A flat carousel failed at one of three gates, not at design

Three different failures produce the same disappointing screen. The post was never served to many feeds. It was served and nobody opened it. It was opened and abandoned on slide two. Those three have separate causes and separate fixes, and the reason most carousel advice does not help is that it assumes the third one and starts talking about design.

The distinction matters commercially, because a deck costs hours and a text post costs minutes. Redesigning a carousel that was never served is paying the expensive fix for the cheap problem, and it teaches you nothing, since the second deck gets served differently too and you learn only that the numbers moved.

The good news is that your own analytics can separate the three, if you stop reading them as one number. Impressions answer the first question and nothing else. A click answers the second. The third question is the one the platform refuses to answer, and there is a way around that which most people never try.

The Carousel Autopsy: three gates and the number that clears each

Run the gates in order and stop at the first one that fails, because everything downstream of a failed gate is untested rather than broken. A deck that was never served has no design verdict attached to it at all, however long you spent on the typography.

The autopsy takes about four minutes on a post and it is worth doing on the good ones too. A deck that performed well for the wrong reason will be copied, and copying a cover that worked because the post happened to be served widely is how a format quietly stops working over a quarter.

Gate one: build your own baseline, because no published benchmark fits

Use the median of your last 10 posts, not the mean, and not somebody else's number. There is no published engagement benchmark for document posts that is worth quoting at you, and the figures that do exist for other formats describe an average account rather than yours. Your median is the only comparison that carries information about your audience.

Here is why the median rather than the mean, with real arithmetic on invented but ordinary numbers. Say your last 10 posts came in at 1,100, 2,400, 900, 1,800, 6,700, 1,300, 2,000, 1,500, 3,100 and 1,200 impressions. The mean is 2,200. The median is 1,650. One unusually good post has dragged the mean a third above the middle of your actual range, so a carousel at 1,700 impressions looks like a failure against the mean and is entirely ordinary against the median.

Reading a carousel against your own median of 1,650 impressions.

The 60% line is a working convention rather than a measured constant. I am proposing it so that you stop redesigning decks that were never shown, and you should move it if your own range is wide. In the example above the posts run from 900 to 6,700, which is a spread of more than seven times, and with variance that wide a single post below the line is weak evidence. Three consecutive posts below it is a pattern.

One more caution about small samples. Ten posts is a small window and it mixes formats, which is deliberate, because you want to know whether this account is being served at its usual level rather than whether documents beat text this fortnight. If you want the format comparison as well, keep a second median for documents only, and expect it to take a quarter before it means anything.

Gate two: the click is the only number that proves the cover worked

An impression on a carousel proves the post occupied space in a feed. It does not prove that anybody looked at the cover, and it certainly does not prove the deck was opened, because the cover sits in the feed as an image and a reader can scroll past it without any interaction. The click is where the cover slide is actually judged.

What each number in post analytics can and cannot prove about a carousel.

The pattern to look for at this gate is impressions at or above your median with a click count well under what your last few decks produced. That combination is the signature of a cover that failed, and it is the one case where redesigning the first slide is the correct and sufficient response.

The opposite pattern is more common and more misread. Low impressions with a respectable click rate means the cover worked on the small number of people who saw it, and the post was starved rather than rejected. Redesigning that deck removes the one thing that was working.

Gate three: LinkedIn will not show slide dropoff, so instrument for it

There is no per-slide dropoff report for document posts, which means the question everybody wants answered, namely where readers left the deck, has no metric attached to it. The way around this is to build the signal into the deck yourself, so that a comment or a reply becomes evidence of depth rather than evidence of politeness.

This is a quarterly instrument rather than a per-post one, and it is worth being honest about the sample size. With single-figure comment counts you are reading noise. What it is genuinely good for is settling an argument that otherwise runs forever inside a team, which is whether the decks are too long or the covers are too vague. Those two beliefs produce opposite redesigns and both feel obvious to whoever holds them.

Saves do some of this work without any instrumentation. A save is a deliberate act taken by somebody who expects to want the thing again, which is why a deck with strong saves and weak comments is usually a deck that delivered to a quiet audience rather than one that failed. Track saves per deck across a quarter and the number becomes more useful than any single post's reactions.

Bucket A, never served: what actually moves distribution

If gate one failed, the fixes are about the account rather than the deck, and you should be sceptical of anybody who states them with confidence, including this page. None of the distribution mechanics are auditable from outside the platform, so what follows is a list of variables with a defensible mechanism, not a list of rules.

• Format switching. An audience trained on your text posts may be a different audience from the one that opens documents, so the first few decks after a switch are being served to people who never chose that format from you.
• Recency of the account's activity. A dormant account gives the system less current signal about who wants the content, which is a mechanism rather than a measurement, and it argues for regular cadence over intensity.
• Who engages in the first hour. Engagement from people who normally engage with you carries a different signal from engagement bought in a pod, and the pod version is the one that quietly degrades an account.
• Outbound links in the post. Published tests disagree about whether a link suppresses reach, and any effect that exists is not stable across accounts or across time. Treat it as a variable worth testing on your own account, never as a settled rule.
• Audience composition. Followers accumulated from one popular post are often the wrong readers for everything after it, which shows up as high impressions and almost no clicks.

The discipline that makes any of this testable is changing one variable at a time and giving it three posts. The arithmetic from the earlier example shows why. If your posts already range from 900 to 6,700 impressions, a single deck moving from 1,650 to 2,000 sits comfortably inside the variation you had before you changed anything, so one post can never settle a question.

Iteration speed is the real constraint here, and it is much harsher for some teams than others. Three posts per variable, at one deck a fortnight, is a six week experiment, and for anybody whose posts pass through a review queue the number is worse. That is a planning problem rather than a content problem, and it is the subject of publishing under compliance without going silent.

Buckets B and C: the cover slide, and the second slide where they leave

A cover fails for one of two reasons. It contains the answer, so there is no reason to open the deck, or it cannot be read at the size it actually appears in a feed. Both are fixable in twenty minutes and neither is a question of taste.

The rule for the cover is that it must make a promise it cannot itself fulfil. A cover reading five ways to run a better one to one has already delivered the concept, and the deck is now optional. A cover naming the specific thing that goes wrong in the fourth month of a one to one cadence has made a promise that needs nine slides to keep. Test the legibility separately by shrinking the cover on your screen until it is roughly the size of a postage stamp. If the headline stops being readable, it has already failed for most of the feed, and the fix is fewer words at a larger size rather than a better layout, which is the subject of carousel dimensions and the type size nobody specifies.

Slide two is where opened decks are abandoned, and the cause is almost always the same. Slide two is a contents page, an introduction to the author, or a restatement of the cover in a larger font. All three are promissory notes, and the reader has already paid one swipe on credit. The rule worth adopting is that no slide may be a promise, including the second one, because every swipe is a small decision that the previous slide has to earn.

Work out what an agenda costs you. In a 10 slide deck, an agenda slide and a credentials slide are 20% of the deck spent before any value arrives, and they sit in the two positions with the highest attrition. Move the first real payoff to slide two and the deck is asking for the second swipe with something already in the reader's hand.

Repost, redesign, or stop making carousels

The autopsy result determines the next action, and in two of the five common outcomes the correct action is to change nothing about the deck. Reposting a starved deck is legitimate and underused, provided you change the opening line of the post text and leave the document alone, because that isolates distribution from design.

What to do next, by what the autopsy showed.

There is a fourth option that nobody in this field says out loud, which is that the format may not pay for itself for your audience. Cost it honestly. Assume a deck takes 3 hours from outline to export and a text post takes 40 minutes, and assume 2 decks a month. That is 72 hours a year on carousels against about 16 hours for the same number of text posts, so a deck has to be worth about four and a half times a text post before it breaks even on your time.

Run that comparison against your own median rather than against the feeling that decks look more serious. If three consecutive decks, all of which cleared gate one, land below your text post median on saves and on comments that cite content, the format is not working for this audience and the honest response is to stop. That decision sits alongside the other format question worth taking seriously, which is that a newsletter is a different product to a post rather than a longer version of one.