Third field report. Measurement window: 13 August 2025 to 12 August 2026. Sources: LinkedIn analytics exports (post and follower data), my own content calendar database, and this site's nginx access logs.
Summary
Seven months ago I had one connection on LinkedIn. Today I have 139 followers and 135 published posts behind me. Here is what the data says, before the detail.
The reach subsidy is real, and it is ending. In April a typical post reached 4.7 times my follower count. In August, 0.7 times. The audience grew sevenfold; reach per post did not move at all.
Virality is worth less than it looks. One post reached 34,708 people, 58% of my entire year. It produced roughly 23 net new followers and fewer than ten website visitors.
Outreach beat content, and it was not close. Follower acquisition quadrupled the week I started actively contacting people. The viral post four weeks later gave one week of +43 followers, then a week of +1. The trend it sat on top of was not created by posting.
Bigger reach means worse engagement. Posts under 120 impressions engage at 2.58%. Posts over 200 engage at 1.08%.
A one-year export would have overstated my performance by 90%. LinkedIn caps every analytics export at 50 posts. Ask for twelve months and you get your best fifty; ask month by month and you get everything. The one-year view put my median at 258 impressions. The true median across all 135 posts is 137.
Why this measurement exists
In 2021 I left the Amsterdam agency I had co-owned for twelve years, sold almost everything, moved to Mexico, and deleted my LinkedIn account. The agency is still there. The account is not.
I had joined LinkedIn early, within a year or two of its launch, so that account held most of a professional network going back to the start of my career in 2000. LinkedIn displayed the connection count as "500+", which is where the counter stops, so I do not know the real number and will not guess at it.
I came back in January 2026 with a new profile and one connection: Peter, my business partner at ClockWork League. New country, no local network, and a deliberate choice to build internationally rather than regionally, because the work travels. I said so at the time, which is the only reason I can date the start of this precisely.
That makes this a clean measurement of something most people cannot measure: what a professional audience costs to build from zero, in public, with receipts.
The instrument
Three data sources, joined on dates and post text:
- LinkedIn analytics, exported month by month. This matters: every export is capped at 50 posts, so a twelve-month range silently returns only your best fifty. Eight monthly exports of the same period give all 135. Each file also carries a follower tab, giving daily new followers across the year.
- My content calendar database, 265 rows with a category on every post.
- This site's nginx access logs, which see every visitor regardless of cookie consent.
135 posts, 129 of them joined to a category by matching post text against the calendar. 88 have both impressions and engagements.
Everything below is that dataset. The methodology is published as a reusable playbook so anyone can run it on their own account.
Finding 1: the date range you pick decides the answer
The first thing the data did was correct my own assumptions.
I started with a twelve-month export, because that is the obvious thing to ask for. It told me my median post reached 258 people. Running the identical analysis across eight monthly exports gives a median of 137.
Same tool, same account, same period. The difference is the date range.
LinkedIn caps every analytics export at 50 posts. Over a year that removes most of your posts, and the ones it removes are the weakest, because the export returns your top performers. Nothing warns you. The file simply arrives with 50 rows.
Monthly is not automatically safe: a month in which you published more than 50 posts is capped too. Mine never were, so eight monthly exports gave me everything.
| Top-50 export | All 135 posts | |
|---|---|---|
| Median reach | 258 | 137 |
| 25th percentile | 232 | 90 |
| 75th percentile | 382 | 213 |
If you have ever analysed your own LinkedIn performance from a twelve-month export, your numbers are wrong in the flattering direction. Mine were. The fix costs nothing: export the same data one month at a time.
Finding 2: the reach multiplier is collapsing
This is the finding I did not expect and cannot stop thinking about.
For every post, I calculated the impressions it earned relative to my follower count on the day it went out.
| Month | Median audience | Median reach | Ratio |
|---|---|---|---|
| March | 15 | 214 | 16.7x |
| April | 21 | 94 | 4.7x |
| May | 36 | 140 | 3.9x |
| June | 57 | 103 | 2.0x |
| July | 79 | 140 | 1.7x |
| August | 141 | 104 | 0.7x |
My audience grew sevenfold between April and August. My reach per post never moved: it sat between 94 and 140 impressions the entire time.
In August a typical post reached fewer people than follow me.
The obvious reading is that LinkedIn front-loads distribution for new accounts, showing your posts well beyond your network while you have no network, then withdraws that subsidy as you accumulate followers. I cannot prove the mechanism from outside. I can show the shape, and the shape is consistent.
It also means the growth I saw in my early months was partly a loan, not an achievement. That is worth knowing before you conclude your content is working.
Finding 3: what a viral post is actually worth
On 26 July I published a post about my children growing up trilingual. It reached 34,708 people, which is 58% of my entire year's impressions. It is 254 times my median post.
Here is what it produced.
| Impressions | 34,708 |
| Followers, 14 days before | +21 |
| Followers, 14 days after | +44 |
| Net gain attributable | ~23 followers |
| Website visitors | fewer than 10 |
One in roughly 1,500 people who saw that post followed me. One in 3,500 or worse visited the site.
The daily shape is instructive. 28 July brought +28 followers, the largest single day of the year. Then it stopped. The following week was +1.
On the traffic side I have server logs rather than guesses. The biggest LinkedIn day of my year produced 21 recorded human sessions on this site, and that number includes my own visits and crawlers that fetch the page bundle. The honest estimate is fewer than ten real people, possibly fewer than five.
A post that reached 34,708 people sent me almost nobody. It was a nice story about my kids. The people who enjoyed it were not looking for an AI developer, and there was never a reason for them to visit.
Finding 4: outreach beat content, and it was not close
Follower acquisition more than quadrupled at the start of July.
| Period | Followers gained | Per day |
|---|---|---|
| 1 April - 30 June | 42 over 91 days | 0.46/day |
| 1 July - 12 August | 87 over 43 days | 2.02/day |
The obvious explanation is the viral post. The weekly data rules it out.
| Week of | New followers |
|---|---|
| 15 June | +4 |
| 22 June | +1 |
| 29 June | +10 |
| 6 July | +5 |
| 13 July | +10 |
| 20 July | +11 |
| 27 July | +43 (viral post) |
| 3 August | +1 |
The rate stepped up in the week of 29 June and stayed there. That is four weeks before the viral post, which produced one week of +43 followed by a week of +1.
What changed at the end of June: I started actively reaching out to people. Not waiting to be found - deliberately identifying people worth knowing and contacting them, with a Sales Navigator subscription to filter for the right ones. I also rewrote my profile around the same time, following my own playbook.
I cannot fully separate the outreach from the profile rewrite. What I can say is that the durable change coincided with starting to reach out, and that the single biggest post of my year did not move the trend at all.
For anyone hoping that good content alone builds an audience: it did not, for me, in seven months of trying.
Finding 5: bigger reach, worse engagement
Across 86 posts with both metrics:
| Reach band | Posts | Engagement rate |
|---|---|---|
| Under 120 impressions | 28 | 2.58% |
| 120-200 | 26 | 1.46% |
| Over 200 | 32 | 1.08% |
My smallest posts engage 2.4 times better than my biggest.
This inverts the metric LinkedIn puts in front of you. When a post travels beyond your network it reaches people with no particular reason to care. Impressions rise; relevance falls.
A post that reaches 90 of the right people and gets three replies is doing more for a business than one that reaches 800 and gets none.
Finding 6: category matters less than expected
Every post in my calendar carries a category. Joined to performance:
| Content type | Posts | Median reach | Engagement rate |
|---|---|---|---|
clockwork_league | 9 | 171 | 1.83% |
free_gpt_friday | 12 | 150 | 1.76% |
working_on_now | 11 | 95 | 1.55% |
personal_story | 13 | 138 | 1.48% |
article_day | 12 | 242 | 1.32% |
agency_life | 13 | 157 | 1.02% |
ai_insight | 11 | 153 | 0.99% |
The spread is narrower than I expected: 0.99% to 1.83%. There is no format that doubles the result.
Two things are still worth noting, and the first is awkward.
free_gpt_friday was my second-best category, and I ended it. Twelve posts in this dataset, consistent performance, and every one handed over a usable tool. The series ran to fifteen free GPTs before I closed it and moved the tools to their own site, Sjenkie's Workshop. That was a deliberate decision about where my attention should go, made before I had any of this data - and the data says the format was working.
I am not restarting it. Fifteen felt like the natural end, and a series that runs on past its own logic is worse than one that stops. But it is a useful reminder that the thing performing best is not always the thing you notice, and that "this ran its course" and "this stopped working" are different claims.
ai_insight is my weakest, which is uncomfortable, because it is closest to what I actually sell. The difference between it and free_gpt_friday is that one is commentary and the other is something you can use today.
Cross-posting site articles earns 75% more reach than a regular post (242 vs 138 median) and slightly lower engagement. It buys distribution, not conversation.
Length does nothing. Every band from under 120 words to over 300 sits between 1.20% and 1.68%.
What this does not establish
Seven months is not long. 139 followers is not a large audience. One account is not a sample.
The follower curve is reconstructed by summing LinkedIn's daily new-follower figures, which are gross additions - they total 148 against a stated net of 139, so nine people unfollowed at times the data does not specify.
The viral post's follower attribution is a before/after comparison, not a controlled measurement. Other posts in the same fortnight contributed some share.
The site session counts are an upper bound: they include my own visits and crawlers with browser-like user agents. That makes the conversion finding stronger, not weaker, but it means "fewer than ten" is an estimate rather than a count.
And the July inflection has at least two candidate causes. I have named both.
What I am doing differently for the next six months
Reaching out more, not posting more. The data is unambiguous: the only thing that durably changed my audience growth was contacting people directly. Content supports that; it did not replace it.
Measuring engagement rate, not impressions. Reach and engagement move in opposite directions in my data. I have been watching the wrong number.
Ship things people can use, not opinions about them. The two categories that handed someone something concrete - a free tool, a specific programme - outperformed my commentary about AI, which is the thing I sell. That is the lesson from free_gpt_friday, and it survives the series ending: the mechanism was usefulness, not the format. What replaces it should keep the mechanism.
Making posts that require the article. My site articles get reach and produce almost no traffic, because the post contains the whole thought. A post that states a finding and withholds the method converts. One that summarises the article does not.
Treating follower acquisition as the compounding metric. With the reach subsidy ending, impressions no longer grow on their own. The audience does, if I work at it.
Not chasing another viral post. It bought 23 followers and no visitors. It is not a strategy.
Conclusion
I came back to LinkedIn expecting content to build an audience. Seven months of data says it mostly did not.
What content did was make me findable and worth connecting with once someone looked. What actually grew the audience was deciding, at the end of June, to stop waiting and start reaching out.
The most useful thing in this data is not any single number. It is the gap between the metrics that feel like progress and the ones that are. 34,708 impressions felt enormous and produced 23 followers. A quiet week of outreach in late June changed the trajectory and looked like nothing at all.
I will run this measurement again in six months. The prediction I am willing to be wrong about in public: reach per post will keep falling relative to audience size, and follower growth will track outreach effort far more closely than posting volume.
If that turns out to be wrong, I will publish that too.
The full methodology, including the export steps and the joining logic, is published as a reusable playbook.