The LinkedIn algorithm is not your audience strategy

I have spent roughly a month posting on LinkedIn almost every day, and I think I am becoming less interested in whether the algorithm likes me.

That is not because distribution suddenly stopped mattering. LinkedIn is still the main place I am using to put Editorial Intelligence in front of people who do not already visit the site, so it would be strange to pretend reach is irrelevant.

What has changed is the question I am asking when a post gets 100 impressions, when another gets 10,000, or when something I thought was strong appears to disappear almost immediately.

There is a familiar complaint on LinkedIn that the algorithm is against creators. You publish something you think is good, barely anybody sees it, and the obvious conclusion is that the platform has buried the work. I have felt that frustration too. But after looking more closely at a month of my own data, I am starting to think the more useful response is not to complain that LinkedIn did not distribute something enough. It is to understand what the distribution is telling me, and then work with the system without letting the system decide what I should think about.

One post distorted the whole month

From 10 August to 6 September, LinkedIn recorded 21,129 impressions and 10,891 members reached across my account. Those numbers look quite healthy until you look underneath them, because one post did an enormous amount of the work.

On 28 August I wrote about OpenAI and Anthropic hiring for senior communications and editorial roles, and used those jobs as evidence for an argument I had been developing about editorial work moving upstream. The post was published at 3:30am UK time and generated 11,679 impressions, reached 8,312 members, produced 62 profile views, gained 27 followers, was saved 44 times and sent 22 people to the related article on Editorial Intelligence.

In other words, that one post accounted for more than half of all the impressions recorded across the four-week period. If I wanted to optimise the account purely for reach, I could draw a very simple lesson from that: write more posts about famous AI companies, jobs and salaries.

I probably would get more impressions if I did.

I am not convinced I would build a better Editorial Intelligence audience.

That distinction is becoming important to me because I always knew the work I am trying to develop here would not necessarily travel easily. Editorial Intelligence sits across editorial strategy, AI, knowledge systems, organisational memory, thought leadership and the mechanics of how ideas become useful inside organisations. Some of those subjects are fairly abstract. Some require a reader to care about problems they may not have named yet. They are unlikely to have the immediate recognition of OpenAI hiring for a well-paid job.

Most of my existing LinkedIn network was also built before I started writing about any of this. It accumulated through Sage, journalism, agency work, previous employers and fifteen years of professional life. There is no reason to assume that everyone in that network should suddenly become interested in Editorial Intelligence just because I am. Judging the work entirely by whether that inherited audience responds to it would be a weak test.

I need some ideas to escape that network and find people who recognise the problem.

A post with 35 impressions can still tell me something

That became much clearer with a post I published this morning about context becoming a form of capital. At the first snapshot, a few hours after publication, it had only 35 impressions and had reached 19 people. On a normal social dashboard that looks like failure.

But seven of those 19 people had reacted, one had followed me, and 52% of the audience was outside my network.

More interestingly, when I looked at who had engaged, I found people working in AI strategy, software engineering, transformation, SaaS, business leadership and executive expertise. Several were second- or third-degree connections.

Another recent post showed the opposite shape. It reached 57 people and generated 10 reactions, which is again a high response relative to the tiny audience, but 82% of that audience remained inside my existing network.

I would be making too much of a very small sample if I claimed that these numbers prove anything about LinkedIn’s ranking system. They do not.

What they do show is that reach and relevance are separate variables. A post can be distributed poorly and still resonate strongly with the people who encounter it. A post can also get a decent response while remaining trapped inside the same network.

That is why I am becoming more interested in audience quality as a metric. If one person working directly on the kind of AI, knowledge or editorial problem I care about discovers the work and wants to see more, that can be strategically more useful than a much larger number of impressions among people who have no particular reason to care.

That is not an argument against scale. It is an argument that fit should come before scale, otherwise you can get very good at accumulating an audience for something you did not actually want to build.

The algorithm is not the audience

I think this is where some of the LinkedIn algorithm conversation gets muddled. The algorithm is a distribution mechanism. It is not your audience strategy, your editorial strategy or your reason for publishing.

LinkedIn controls a huge part of who sees a post, and that can be irritating because the creator has very little visibility into the decision.

But I can still control the things on my side of the exchange: what I write about, how clearly I frame it, how often I return to an argument, when I publish, which conversations I participate in, who I respond to, what I link people towards and whether there is a coherent body of work waiting when somebody becomes interested.

That last point is especially important for me. I am not trying to make every LinkedIn post carry the whole Editorial Intelligence proposition. The website now contains enough connected writing that a post can act as an entry point.

Somebody might arrive through a job post, an argument about AI writing, a note about context, a discussion about GEO or a comment I leave under somebody else’s post. The job of that individual interaction is not to explain everything. It is to make the next useful step possible.

This is also why I am less worried than I was about posting frequently. I wrote recently that the LinkedIn number I cared about was the share of an audience coming from outside my network, because it showed that an idea had escaped the people who already knew me.

The bigger lesson is that my own experience of posting is completely different from everybody else’s experience of following me. I see everything because I wrote it. LinkedIn shows everybody else a fraction.

A lot of the content I publish does not travel very far at all. That is not necessarily a problem if the publishing is helping me build the territory, test arguments and occasionally find the people for whom the work is relevant.

Commenting is part of audience selection

I have also started to think differently about commenting. The standard creator advice is to comment on other people’s posts because it increases visibility, which is probably true but does not tell me where I should spend my attention.

If I comment indiscriminately on high-reach posts, I may get more profile views without becoming any clearer about what I am interested in.

If I repeatedly have substantive conversations around AI strategy, knowledge systems, editorial work, organisational change, thought leadership and the future of content, something different happens. I am not only trying to increase engagement; I am choosing the intellectual neighbourhood in which I want Editorial Intelligence to be encountered.

That makes niche commenting much more interesting than generic engagement tactics. A useful comment can introduce an argument to people already thinking about the same problem.

It can also show me which language practitioners use, what they disagree with, which examples they recognise and where my own framing is too abstract. In that sense, commenting belongs inside the same evidence loop as publishing rather than sitting outside it as promotion.

This is one reason I do not want to automate the whole thing into a growth routine. There is a difference between mechanically leaving comments to make an account visible and using conversation to understand a territory.

The second requires judgement. Sometimes the useful outcome is not that my comment gets likes. It is that somebody makes a point that changes the next article.

I am experimenting with time, but not looking for a magic hour

The posting-time question is another area where I think the usual optimisation advice can become too simplistic. The breakout jobs post was published at 3:30am UK time. The context post went out at roughly 4am and, in its tiny initial sample, reached a surprisingly relevant group outside my network.

That does not mean 4am is the best time to post on LinkedIn. Two examples would be a ridiculous basis for that conclusion, and there are too many variables in the subject, framing and initial response.

It has made me curious about something more specific: whether posting time changes not just how much early distribution a post gets, but who that early audience actually is.

That is a better experiment. I can publish across very early UK mornings, normal working hours and evenings, then compare the first audience rather than simply asking which slot produces the most likes.

The four-week content demographics already look much less London-heavy than my overall audience: London accounts for 36% of my listed audience demographics, while the content-viewer demographics for this period put both New York and London at 10%. That is not enough to attribute the difference to posting time, but it is enough to make geographical distribution worth watching.

So I am going to keep varying the times rather than standardising too early. The aim is not to discover a hack. It is to learn how the platform behaves around the kind of work I actually publish.

Not every post has the same job

The biggest trap in looking at the breakout post would be to assume every future post should behave like it. It will not, and I do not think it should.

Some posts are naturally good at reach because they attach an idea to something already carrying attention. Jobs at OpenAI and Anthropic are a good example. Some posts are better at making an existing argument more precise. Some attract practitioners who recognise themselves in the problem. Some produce saves. Some generate a useful disagreement. Some send somebody to the website. Some mainly help me find out whether an idea is understandable outside my own head.

That is close to an argument I have been making elsewhere in Editorial Intelligence: give every piece a job. If different pieces are doing different jobs, measuring all of them against impressions alone makes very little sense.

For LinkedIn, the measurement chain I want to build now looks more like posting time → initial reach → response → out-of-network share → audience relevance → profile activity → followers → site discovery. I also want to record who is engaging in broad terms, because a number without any sense of audience composition can hide as much as it reveals.

This is not a sophisticated attribution model. It is deliberately simple enough to use. After another 10 or 15 comparable posts I should know more than I do now, and that learning can go back into the Editorial Intelligence evidence bank rather than disappearing into a dashboard.

Working with the algorithm does not mean serving it

There is an obvious counterargument to all of this. Once you start analysing distribution, it is easy to let the platform train you into producing whatever it rewards. The jobs post travelled, so write more jobs posts. Short posts outperform long ones, so stop developing complicated ideas. Outrage gets comments, so become more outraged. If that is what “working with the algorithm” means, I am not interested.

The distinction I am trying to make is between learning from distribution and being directed by it. LinkedIn gives me behavioural evidence that I did not have before: what escaped my network, what people saved, what generated a follow, what led to a site visit, what attracted practitioners in the territory and what disappeared. Editorial judgement still has to decide what any of that means.

Sometimes the right response to weak reach will be to improve the framing. Sometimes it will be to return to the idea from a different angle. Sometimes it will be to comment in a more relevant niche and let the audience build slowly. Sometimes the conclusion should simply be that an important idea is not especially social-media-friendly and deserves to exist anyway.

That is very different from treating every low-impression post as evidence that LinkedIn has failed you.

After a month of posting almost every day, I do not think my aim is to defeat the algorithm. I want to understand enough about it that I can use LinkedIn for what it is good at: exposing ideas to people I do not already know, creating evidence about which problems resonate, and helping a specialist audience find a body of work that may be useful to them.

The number I care about is therefore not simply getting bigger. The question underneath the number is getting better:

Did this idea find the people it was meant to find, and did I learn anything useful from what happened next?