The mistakes founders make on LinkedIn

The mistakes founders make on LinkedIn are mostly measurable, and the most common one takes ten seconds to stop. Across 5,619 posts from 21 founders and CEOs, four or more hashtags sit on 39.4% of the role's own weakest posts and on 13.2% of its strongest. The rest of the list runs the same way: the corporate “we” in place of a person doing something (5.2% of the worst posts open in first person against 19.6% of the best), the image attached by reflex, the feed built entirely from announcements, and reactions treated as the score when comments are the scarce thing. One habit founders worry about constantly, hook length, is not a mistake at all. The medians are one character apart.

How did we decide which founder habits count as mistakes?

By ranking founder posts against other founder posts. The role bucket holds 5,619 posts from 21 founders and CEOs, drawn from a filtered cohort of 12,988 English posts by 65 creators. We split that role against itself: its own top decile is 561 posts from 17 authors, its own bottom half is 2,810 posts from 12 authors. A habit earns a place on this list only when it is meaningfully more common in the role's own bottom half than in its own top decile. That is a stricter test than comparing founders against the rest of LinkedIn, because subject matter stays constant on both sides.

The distance between those two groups is not cosmetic. The median post in the role's top decile ran an engagement rate of 5.66 per 1,000 followers. The median post in its bottom half ran 0.08. Everything below is a habit that shows up on one side of that line more than the other.

Two rules govern the rest. Everything here is correlational: a habit can be more common among posts that did badly without having caused it, and the top group leans on 17 authors while the bottom leans on 12. And where we have no measurement, we label the item as judgment. Three of the eleven are exactly that.

Mistake 1: why do four or more hashtags sit on 39.4% of the worst founder posts?

Within the founders and CEOs role, four or more hashtags appear on 39.4% of the bottom half (2,810 posts, 12 authors) and 13.2% of the top decile (561 posts, 17 authors). Three to one. Cohort-wide the same habit splits 41.9% against 25.9%, so founders are worse at this than the average creator, not better.

The more useful figure is the one for any tag at all. Founders' top-decile posts carry at least one hashtag 49.2% of the time; their bottom-half posts, 58.3%. Even moderate tagging is doing nothing for this role, which reverses the cohort-wide picture, where at least one tag shows up on 78.0% of top-decile posts against 59.0% of bottom-half ones. Hashtags help posts that need to be found by topic, and a founder post is found by name.

The repair. Cap it at three, put them on their own line at the end, and drop them entirely on any post about your own company or your own decision. If a draft feels like it needs eight tags to reach anyone, the tags are not the problem you are solving. The full distribution across every segment is in the hashtag analysis.

Mistake 2: are you posting the link instead of the thought?

Here is where we have to be careful, because the founder numbers do not say what the popular advice says. Within the founders and CEOs role, shared articles account for 34.8% of the top decile and 31.7% of the bottom half. That is flat, and if anything it tilts the wrong way. Founders in this dataset share links at roughly the same rate whether they are winning or losing.

Cohort-wide the pattern is different. Across all 12,988 posts, shared article or link posts make up 32.7% of the bottom half and 22.1% of the top decile, an over-index of about 1.5x at the bottom. Inside the startups topic (833 posts, 50 authors) the gap is wider still: 41.0% of that topic's own bottom half against 25.3% of its own top decile. So the penalty is real in two of our three cuts and absent in the one that describes founders specifically. We are not going to smooth that over. Anyone who tells you links are uniformly fatal has not looked at a role-level split.

What survives all three cuts is narrower and more useful. A link post that consists of a link plus “great read” contains no information the reader could not get from the headline. That is not a distribution problem. It is an authorship problem.

The repair. Write the thought first, at full length, and treat the link as a footnote to it. If the post would be worth reading with the URL deleted, keep the URL. If it would collapse, you were forwarding, not writing. Our reconciliation of the contradictory published studies on this sits in the external links breakdown.

Mistake 3: why does the corporate “we” sink a founder post?

This is the widest gap in the role and the one worth fixing first. Within the founders and CEOs bucket, 19.6% of the top decile opens in first person against 5.2% of the bottom half. Nearly four to one. Inside the startups topic the same split runs 39.8% against 11.0%, across 83 top posts from 14 authors and 417 bottom posts from 31 authors. Across the whole role, including everything in the middle, only 10.9% of founder posts open in first person at all.

One honest complication. Our first-person flag counts I, my and we, so the corporate “we” is already inside the winning column. The split therefore cannot be a pure pronoun effect. What separates the two groups is whether a human being is doing something in the opening sentence at all, as opposed to a company being positioned, a market being described, or a principle being asserted. “We shipped it and it broke” and “I shipped it and it broke” are both on the right side. “Innovation in logistics has never been more critical” is not.

The repair.Read your first sentence and answer two questions: who did this, and when. If the honest answer is “the company, over the last quarter”, you have written the bottom-half version. Rewrite it so a named person does a specific thing on a specific day, then let the company appear in sentence four.

Mistake 4 and Mistake 8: do announcements and conference photos add up to a feed?

They do not, and the reason is scheduling rather than quality. Announcements work. We can show you that they work, which is why this section is more complicated than the usual advice.

Both are enormous relative to the role, where the median post draws 132 reactions and 12 comments at an engagement rate of 0.19 per 1,000 followers. Take the reach when the event arrives. The mistake is treating it as a content plan, because the event supplied the reach and the event is not repeatable. A founder who has had one post like the second one above spends the next six months trying to reproduce it, and what they produce is the milestone post, the anniversary post, and this:

Five comments is not a failure of distribution. It is the correct yield for a post that reports attendance. The median bottom-half founder post earns 6 comments, so this one is performing exactly at the floor despite appearing in an exemplar set. The conference photo has the same structure: you were somewhere, it was good, here is a badge. Nobody has anything to add.

The repair. Announcements stay, at whatever rate reality supplies them, and you spend the attention they earn on something with an argument in it within the following week. Recaps get converted: write the one thing the speaker said that you think is wrong, or the one line you have quoted twice since. A recap with a disagreement in it is a post. A recap without one is a receipt.

Mistake 5: is the image you attached actually helping?

Probably not, and this is the second-widest format gap in the role. Image posts account for 44.7% of the founders' bottom half and 32.8% of its top decile. Text with no media at all runs the other way, 14.5% at the bottom and 26.6% at the top. The cohort-wide split points the same direction: images are 40.3% of bottom-half posts against 30.7% of top-decile ones.

The mechanism we would guess at is not that images are penalised. It is that attaching one is what people do when they suspect the words are thin. A quote card, a dashboard screenshot with the numbers blurred, a stock photo of a team meeting: each adds nothing a reader did not have from the first line.

Now the part we cannot resolve. Native video runs at 5.7% in the founders' own top decile and 9.1% in its own bottom half. That is backwards from every other cut we have. Cohort-wide, video is 26.3% of top-decile posts against 10.1% of bottom-half ones, an over-index of about 2.6x, and inside the startups topic it is 32.5% against 13.2%. So video looks strong everywhere except in the role bucket that this article is about. We do not know why. It might be the kind of video these 21 authors shoot, it might be an artefact of which accounts landed where, or it might be that founder credibility travels fine in text. Correlational data cannot separate those, and we are not going to tell you to stop filming on the strength of a contradiction.

The repair. Default to text. Add media when the media is the content: a photo of the thing that broke, a chart with real axis labels, thirty seconds of you saying something you would not put in writing. Never add media to a finished post because it looks bare.

Mistake 6: does your hook need to be shorter?

No, and this is the item on the list that is not a mistake. Founders spend real effort trimming opening lines, and the data gives them nothing back. Within the founders and CEOs role, the median visible hook measures 206 characters in the top decile and 205 in the bottom half. One character. Across the whole 12,988-post cohort the medians are the same two numbers, with word counts of 36 and 34. At every cut we have looked at, length separates nothing.

The related superstition is that a number in the hook lifts a post. Cohort-wide it splits 32.1% against 32.8%, which is noise. Within the founders role there is a mild positive signal, 38.5% against 33.2%, so founder-specific numbers may be doing modest work. That is a much smaller effect than the pronoun or the hashtag gap, and it is worth roughly one of the minutes you were about to spend counting characters.

The repair. Stop editing for length and start editing for position. LinkedIn truncates the preview at roughly 210 characters on desktop and sooner on mobile, so the question is not how long your opening is but whether the interesting clause lands before the cut. If the surprise sits at character 240, nobody sees it. The mechanics of what to put in that window are in the hooks study.

Mistake 7 and Mistake 9: are you chasing reactions when comments are the scarce signal?

Reactions are the number founders quote to each other and the number that moves least. In the founders and CEOs role, the median top-decile post drew 94 reactions and the median bottom-half post drew 89. That is a difference of five. Comments went 13 against 6. The gap between a founder post that worked and one that did not is almost entirely a comment gap.

Cohort-wide the same asymmetry is louder. The top decile median is 235 reactions and 72 comments; the bottom half median is 80 reactions and 5 comments. Reactions triple, comments go up roughly fourteenfold. Inside the startups topic, median comments run 70 against 7. A reaction costs a reader nothing and means about that much. A comment costs them a sentence they have to sign their name to.

Which brings us to the close, and to the honesty rule. We cannot measure post endings. Our source text is truncated at LinkedIn's see-more fold, so every text finding in this article describes the visible hook and nothing else. What follows is judgment, not measurement: “Thoughts?” and “Agree?” are requests for a favour rather than questions, because they ask the reader to generate both the topic and the answer. The founder posts pulling large comment counts tend to contain a specific disagreement or a specific gap in the writer's own knowledge, which gives a reader a slot to fill.

The repair.Close with a question only your reader can answer, about their situation rather than about your post. “How many demos does it take you to close one?” beats “thoughts?” because there is a real answer and it is short enough to type on a phone. Then answer every reply for an hour. The comment mechanics piece covers the first-hour behaviour and what a decent reply rate looks like. What does not work is buying the signal back through a pod, which produces comments from people who did not read the post.

Mistake 10: what happens when you go quiet for six weeks and then post a fundraise?

Mark this one as judgment. Our dataset is a snapshot of posts, not a time series per account, and engagement counts are lifetime totals at the moment of collection with post ages varying, so we cannot measure what a six-week gap does to the post that follows it. Anyone quoting you a precise decay figure for silence is quoting something they did not measure either.

What we can say is structural. The bursty pattern is the founder default: nothing, nothing, nothing, then a 400-word funding announcement, then nothing again. The audience that would carry that post is the audience you spend the quiet weeks not building. And the announcement posts that perform in our set belong to people who are visible the rest of the time, so the reach you are counting on has a base under it that the burst pattern does not create.

There is a second cost that has nothing to do with the algorithm. Writing rusts. The first post after six weeks is the worst one you will write that quarter, and you have chosen to spend it on the announcement that mattered most.

The repair. Set a floor, not a target. One post a week that you would send to a customer, published whether or not anything exciting happened. A floor survives a bad month in a way a five-a-week commitment does not, and it means the fundraise lands on an audience that has heard from you recently. The posting frequency data covers what cadence does and does not buy.

Mistake 11: should you outsource the writing entirely?

Also judgment, and we should be fair about it, because we sell software that writes posts. We have no measurement of who wrote what in this dataset. Nothing in our data distinguishes a founder's own draft from a ghostwriter's or a model's.

The nearest proxy is the first-person split, 19.6% against 5.2% within the role, and what it implies about material rather than authorship. The posts that separate carry information only the founder had: a number from inside the business, a decision with a price, a sentence a customer said on a call. A writer who has never sat in your pipeline review cannot invent those, so the failure mode of full outsourcing is correct prose about nothing, which is the exact shape of the bottom half.

The version that works keeps the founder in two places and only two. You supply the raw material, in whatever state, and you approve the draft before it publishes. Everything in between (structure, opening line, trimming, scheduling) is genuinely delegable, and pretending otherwise is how founders end up publishing nothing for a quarter. The tells that make outsourced writing obvious are catalogued in our piece on why AI drafts read as AI.

The repair. A review loop with a deadline. Someone or something drafts, you read it and either kill it or change one sentence to something only you could have written, and it goes out. A ban on help is not the answer, because the alternative most founders actually pick is silence.

What do all eleven mistakes look like in one table?

Every figure below is within the founders and CEOs role unless the row says otherwise, and the role holds 5,619 posts from 21 authors, split into a top decile of 561 posts (17 authors) and a bottom half of 2,810 posts (12 authors).

How do you audit your own last 20 posts?

This takes about twenty minutes and it is more informative than any benchmark, because it tells you which of the eleven you personally do. Open your activity feed, take the last 20 posts, and count.

1. Count the ones with four or more hashtags.If it is more than three of twenty, you are publishing at the rate of the role's bottom half, where the habit sits on 39.4% of posts.
2. Count the first sentences with a human subject. Not a pronoun count. A count of sentences where a person does something. Fewer than five of twenty and you are in the register that runs 5.2% at the bottom of the role.
3. Count the posts that are a link plus a comment. Then reread each one with the URL removed and ask whether anything is left.
4. Count the images that carry information. A chart with labels counts. A quote card does not. Separately, count the posts with no media at all: if that number is under five, you are further from the 26.6% top-decile share than you think.
5. Count occasions. Announcements, recaps, congratulations, anniversaries. If more than half of twenty are occasion posts, your feed depends on events you do not control.
6. Write down comments, not reactions, for each.Then take the median. The whole-cohort median post gets 12 comments and the founder role median is also 12, so that is your line. The role's bottom half sits at 6.
7. Compute one engagement rate. Reactions plus four times comments, divided by followers, times 1,000. The cohort median is 0.40 per 1,000 followers, the 75th percentile is 1.27, and the top-decile cut is 5.95. The founder role as a whole sits at 0.19, so do not panic at a number under one.
8. Read the last line of each post.Count how many end in “thoughts?” or an equivalent. This is the judgment item, so treat it as a prompt rather than a verdict.

The output is a list of two or three habits, not eleven. Fix the widest one first, usually the hashtag block or the opening sentence, because both are free and neither requires new material. What to write once those are fixed is a separate question, and our guide to what a founder should actually post answers it.

What can this analysis not tell you?

All of it is correlational. We can say that four or more hashtags are three times more common in the founders' weakest posts than their strongest, and we cannot say the tags caused it. The two groups also differ in who wrote them, 17 authors at the top against 12 at the bottom, so author effects and habit effects are mixed together in every figure on this page. The cohort skews toward established creators, all with at least 1,000 followers across 65 accounts, so treat everything as directional if you are starting from two hundred connections. Engagement counts are lifetime totals at the time of collection and post ages vary, which flatters older posts. Every text finding describes the visible hook above the fold, not the body, because that is what the source captures. And three items on the list (the close, the silence, the ghostwriter) are judgment rather than measurement, which we would rather label than dress up. The method is in the 34,000-post engagement study, and the rest of the role-level cut sits on the founders and CEOs page. The mistakes founders make on LinkedIn are cheap to fix in the order given here, and expensive to keep.

Posting in a different role?

The patterns move with the role, so each of these is measured against that group's own strongest and weakest posts rather than the cohort average: founders who are just starting, consultants chasing clients, executives who will not write, B2B SaaS founders.