The nine hook types in 918 top founder LinkedIn posts

Sonar keeps a shelf of high-engagement public LinkedIn posts so that every account we work with can see what is working outside its own niche. Each post on that shelf carries a label naming the kind of opening line it uses. There are nine labels. We had never looked at them in aggregate, so we did, and this is what 918 labelled posts say about which opening lines earn the most.

The short version: the best-performing hook type has about three and a half times the median engagement of the worst, the spread inside any single hook type is between five and eighteen times, and once you account for whose account a post was published on, eight of the nine hook types sit within a few percent of each other. Only one is clearly ahead.

What we counted

The shelf is built by a weekly job that reads public LinkedIn posts from a fixed roster of 30 public profiles that publish about building companies and audiences, keeps the ones above an engagement floor of 200 reactions, comments and reposts combined, and asks a small classifier to name each one's hook type from a fixed list of nine. The list lives in the product, not in this post, and the classifier sees the first 400 characters of the post.

• The sample. 918 posts from 15 accounts, published between 31 December 2025 and 19 September 2026, captured between 19 July and 20 September 2026. Fifteen of the 30 profiles on the roster have posts on the shelf today; the rest returned nothing above the floor. Each account contributed between 1 and 136 posts.
• What we did not read. No customer's posts, no private analytics, nothing from anybody's account inside Sonar. The study reads the public shelf and the public watchlist captures, and nothing else.
• Engagement. Reactions plus comments plus reposts, which is every number LinkedIn shows a stranger. Impressions are private to whoever wrote the post, so no reach figure exists here. What the public numbers mean is covered in what LinkedIn impressions are.
• Medians, not averages. Engagement is heavily skewed: the mean on this shelf is 2,303 and the median is 1,459. One post at 26,976 would move an average and cannot move a median.
• Thin buckets. Any hook type with fewer than 30 posts is marked thin and should be read as a hint.

Because of the engagement floor, every post here already worked. Nothing in this study can tell you which hook gets a post from 40 reactions to 400. It can only tell you how already-successful posts opened.

The nine hook types, by median engagement

The top three by median are the question hook at 2,116, the vulnerable admission at 2,110 and the contrarian take at 1,998. The question hook's bucket holds 27 posts, under our threshold of 30, so it is marked thin and it is the one row in the table that would move most if we ran the sweep again next month.

What the numbers say

Four hooks cluster at the top and the gap between them is nothing. The first four rows sit between 1,981 and 2,116, a spread of under 7 percent. Choosing between a question, an admission, a contrarian line and a flat claim is not a decision this data can make for you.

The bottom of the table is a real gap. A news take, which opens by reacting to something that happened in the industry, has a median of 609, under a third of the top row. Posts about what somebody else did earn less than posts about what you did. The data point hook, which opens with a statistic, is second from the bottom at 942, which surprised us.

Admissions carry the widest right tail. The vulnerable admission has the highest mean of any hook at 3,179 against a median of 2,110, which means its biggest posts are much bigger than its typical one. Nothing else on the shelf has that shape at that size.

What the numbers do not say

The account matters more than the hook. The 15 accounts on the shelf have per-account medians running from 220 to 5,242, a spread of more than 20 times. So we recalculated: instead of comparing every post to every other post, we compared each post to the median of the account that published it. The ranking almost disappears.

Read against its own author, the vulnerable admission is the only hook that clearly beats the line: 59 percent of its posts do better than the median post on the same account. Seven of the remaining eight sit between 0.86 and 1.05. The question hook falls from first place to sixth, which is what a thin bucket does. Restricting the same calculation to the 11 accounts with at least 20 posts each gives the same answer.

The spread inside a hook type dwarfs the spread between hook types. Between the best and worst hook the median moves 3.5 times. Inside a single hook type, the tenth percentile to the ninetieth moves between 5.6 times (numbered promise) and 18.1 times (question hook). Whatever separates a good post from a great one is mostly not the hook.

One label is a catch-all. The classifier picks one of nine labels, and a reply it cannot match falls back to "bold claim". That bucket is the largest in the table at 195 posts, and some unknown share of it is failed parses rather than genuine bold claims. Removing it entirely leaves the order of the other eight unchanged, which is the only reassurance we can offer.

The labels were never checked by a person. One classifier pass, no second opinion, no inter-rater agreement measured. A post that opens with a question about a statistic could land in two buckets and there is no appeal.

Fifteen accounts is a niche, not a population. These are accounts that publish about building companies, to audiences that expect that. A B2B founder writing for 900 people in one industry is a different physics problem, and the honest comparison for them is their own last thirty days.

A second shelf, a different order

We ran the same count on a separate capture: 188 labelled posts from the profiles individual accounts have chosen to watch, 166 on LinkedIn and 22 on X, with no engagement floor at all. The median post there earns 154 rather than 1,459. There is no overlap with the first shelf.

Seven of the nine buckets hold fewer than 30 posts, so this table proves very little on its own. It is here because it fails to reproduce the first one. The data point hook goes from eighth to first and the question hook from first to eighth. If hook type were a strong, portable cause of engagement, two independent samples would agree more than this.

Nine openers for the three leading hooks

Every line below is invented for this post. None quotes a real post, and none names a real person or company.

Vulnerable admission

• "We missed our own launch date by five weeks and I told the waitlist on the day it slipped."
• "I have rewritten our pricing page four times this year because I still cannot explain it out loud."
• "I let a bad hire run for seven months after I already knew, and the delay cost more than the mistake."

What makes these work in the sample is a specific cost, arriving early. An admission with no number and no consequence reads as a performance of honesty.

Contrarian take

• "Most advice about founder content is written by people who have never had to ship in the same week."
• "Our weekly all-hands made the company slower, so we cancelled it in March."
• "We stopped answering RFPs and our close rate went up."

The contrarian posts that carried the shelf all had the evidence in the post, usually in the second or third line. A contrarian claim with nothing under it is a pose that the comments will test.

Question hook (thin bucket, 27 posts)

• "What would you cut first if your revenue halved on Monday?"
• "How many of your customers could name the thing you sell?"
• "Why does every pricing page in our category start at the same number?"

These are questions the post then answers. "Thoughts?" and "Agree?" are a different move and they are not what is in this bucket.

What a hook cannot do

A hook buys the second line. That is the whole job, and the reason the top four rows of the first table are within 7 percent of each other is that four different openings all buy the second line about equally well. What happens after it is where the 18-times spread inside a bucket comes from.

The practical read for a founder: stop auditioning openings and go and get something worth opening with. The posts at the top of both shelves have a specific event, a number, a decision or a cost inside them, and the hook is just the first sentence of that. If you want a rule, borrow the one the second table supports: say the thing that cost you something, and say what it cost.

This is also why Sonar starts at the other end. It reads your calendar, your meeting transcripts, your Slack and your notes once a week, read-only, and pulls out the moments worth saying before anybody writes a line. The opening sentence is the last decision, and by then the post already has something in it. How it works is the mechanism, Google Calendar is the smallest source you can start from, and how to write LinkedIn posts is the writing half. If you would rather buy a hook library than build a week, the Taplio review is our honest account of that trade.

Questions

What is a hook in a LinkedIn post?

The hook is the opening line or two, the part LinkedIn shows before the "see more" fold. Its only job is to earn the click that expands the post. Our own post formatter draws that fold at about 210 characters, which is the figure Sonar's preview uses, so the hook is whatever survives that cut.

Which LinkedIn hook type gets the most engagement?

On our shelf of 918 high-engagement public posts, the question hook had the highest median at 2,116, followed by the vulnerable admission at 2,110 and the contrarian take at 1,998. The question hook's bucket holds only 27 posts, so the more reliable answer is the vulnerable admission: it is the only hook type that beats the median of the account that published it, at 1.18 times, with 59 percent of its posts above that line.

How many hook types are there?

There is no official list. Ours has nine, chosen because they covered the openings we kept seeing without leaving a bucket that meant nothing: story cold-open, contrarian take, numbered promise, question hook, bold claim, vulnerable admission, how-to promise, data point and news take. Any taxonomy is a choice, and a different nine would produce a different table.

Do question hooks work on LinkedIn?

In this sample, yes, with a caveat. The 27 posts labelled question hook had the highest median engagement in the raw table and a below-average one once measured against their own authors. Every question in the bucket had an answer in the post. A question posted as bait, with nothing behind it, is a different thing and this study says nothing about it.

Does the hook matter more than the post?

The numbers say no. Between the best and the worst hook type the median engagement moves 3.5 times. Inside a single hook type it moves between 5.6 and 18.1 times. Most of what separates posts is happening after the first line.