Introduction: Why LinkedIn matters

Anybody here not know anything about LinkedIn? Right, I'm going to assume everybody does know about LinkedIn and there's a good reason for that.

It's got over 1 billion members. There are millions of posts going out every day and obviously every week.

The frustration: Great posts vs. tumbleweed

But the average post that you put out there reaches only a small, tiny minority, around 1 % of your total followers. And that was bugging me.

And it was bugging me because I was putting out posts on LinkedIn as a good boy, as he meant to do, and sometimes they would go really, really well. Lots of people would like them and comment on them, and other times tumbleweed.

And I got rather frustrated by this.

Using AI for research (not writing posts)

And so in the modern day, of course, I could Google all this and research it, but yeah, no, No, I chose to use AI. So I didn't want to use AI to write the posts, I know, called me old fashioned. I wanted to research the platform.

I wanted to find out why some posts go viral and others do not. So I asked the question to Claude, as it so happens, my favourite AI agent at the moment, but obviously other ones are available.

Not that long ago we would use Notebook LM perhaps for doing this deep research, but I find that Claude is actually really good at the deep research now, especially if you're using Opus and you tell it to do extended thinking and you tell it specifically to research, not try to fix it, but just to look into research.

And so I explained why, what I wanted out of it, and told it to get on with it. And Claude, being Claude,

in co -work especially, spun out loads of different agents, grabbed loads of different website website data research from, well, I believe it's closer to 86 different documents that it ended up grabbing and looking into.

All of the data it could find around why one LinkedIn post does better than another. But I didn't want it just to grab loads of information that I would

never read, I wanted it to go a little bit further, to analyse that, and to try to summarise all that so that I could actually make some sense of it and get some useful, actionable insights from it. So it did.

And what it told me surprised me, and I think it surprises a lot of people that use LinkedIn on a regular basis.

What LinkedIn actually rewards (according to the research)

The things that we thought it cared about, it really does not. so all those likes that we've been asking for you know like press that like button down below absolutely useless mainly decorative makes it look pretty but doesn't really help

Early engagement, links, saves, and dwell time

we've heard it's the first 60 or 90 minutes that count not anymore it's the first 30 minutes It's how people react to your post in the first half an hour that will make the biggest difference.

I used to include loads of links in my post, you know, have a look at this information over there, hey go to our website over here. Don't do that anymore. 1That will actually penalize your reach these days.

If you can get somebody to save your post, you're really making a a massive difference to how viral that post is likely to become. Really, really powerful. So if you've got loads of useful content in there that somebody feels they want to come back to and refer back to and therefore will save, fantastic.

And then, yeah, eyeballs time, the dwell time. How long somebody actually looks at that piece of information.

And this is really interesting when it comes onto images, because I'll come on to that a little bit later but the more interesting, relevant, insightful, deep the content, the longer somebody looks at it the better it will do.

Posting cadence and the long tail of impressions

And we used to be told if you want to win on LinkedIn post daily, nah Nah, you don't need to do that. You need to do it regularly, but not that regularly,

because every time you put out a post, especially if you did one yesterday or earlier that same day, it starts to cannibalize your audience because it won't promote the second one or the third one as much as it did the first time around because it will alternate between them.

That's why you see a long tail with your posts. You see impressions on posts that you posted two weeks ago and you're thinking, but I posted that two weeks ago, how come somebody's only just liked it? It's because behind the scenes the algorithm is only just now showing it to some of your followers.

Secondary factors that amplify (or limit) reach

So here's more of the information that it found. I'm not going to read it all out to you, but there are some very interesting ones to point out.

if you've got somebody with loads of followers who comments on your post that is a massive accelerator so it's actually worthwhile to try to engage connect with people with loads of followers because if they like something that you say or better still they comment on something that you

say in a post it's more likely to go viral than not if you can get different people to comment on a thread, that also magnifies the reach, surprisingly, by over five times.

Now, I'm not suggesting that you get some friendly people to comment on this immediately after every blog post, but if you do have friends and family that can help out, maybe give that a go.

I wouldn't recommend that.

the problem is of course the LinkedIn APIs are not generally available to the public so it's actually quite hard to do that though there are some slightly dodgy places that can help you if you wanted to go down that route but curiously AI generated content which was extremely common and in the past is now actually penalizing you.

But I think you already knew that. So

Turning insights into a post scoring system

I'm lazy. I didn't want to read all of this.

The insights were useful but I then of course spoke to my mate Claude and said right okay based on what you now know how would I score these posts how can you help me make sure that my posts reach as many people as they could possibly

and it went away and thought about it and said actually what I need to do is really break this up in a number of different domains and score it against different things but you know what I'll score it out of a hundred and I'll include the hook the opening the engagement triggers the the content structure, the content value, technical optimization, but you know there's something else

that I need to do. I didn't ask it to do this by the way, it automatically came up with a final one of the needs to be relevant for your audience. Who are you targeting?

because if your post is about cats and your audience are dog lovers it's not going to resonate so tell me about your audience and through that we built up a profile of who my ideal audience was and from that it then added a domain just for scoring how will it match the games the audience so very straightforward forward, it's a score out of 100.

If you're over 70, go ahead and post, you'll do fine. If it's under 70, you may want to rewrite it.

And that's where it really comes in to its own, because not only does it give you the score, it gives you some ideas as to which parts of it are doing well, and which bits of it are not, and even suggests some of the fixes.

And of course, if you can and measure it beforehand, fantastic. So that went really, really well

When LinkedIn changes the game: Brew 360 and 'virality' scoring

until LinkedIn decided to scupper all my great plans and just rewrote the algorithms, bless them.

So I don't know how many of you follow LinkedIn algorithms, but they came up with 360 brew, no idea why. And they've slowly been introducing that, and that's pretty much the algorithm that drives LinkedIn posts.

um virality today i made that word up well i didn't claude did but i'm going with it how viral it will be and it's actually based on ai itself which is intriguing but it's got loads of different parameters within it we have no idea realistically how that works so i've been using claude to try to reverse engineer as much of it as we can from what we know about it that has been published

at least in general terms but also how LinkedIn itself seems to be reacting to the different posts that everybody is putting into it and what I found is that again AI generated content is being suppressed they're trying to keep it

clean and relevant and of interest to the audience and not only does it affect the reach but it's actually having an interesting effect on the engagement so the humans are starting to recognize AI generated content and are turning off or getting turned off by that content as a result and what's interesting though is

that it's now actually reading the post this isn't deterministic this is actually going by the sentiment analysis this is going by what you're talking about.

And if you're writing about something that is not in your sphere of influence, knowledge, expertise or what have you, it hurts you.

If you, for example, are running a dog kennel service and you start to write about ferrets or about balloons or something completely irrelevant, it will say, well that's not not what I know you do. So it will suppress that.

And when you do talk about your area of expertise, it will suppress those as well. So top tip, stick to what you know, talk only about what you do, and that will help you loads.

AI-generated content suppression and 'AI tells'

But of course, I talked about AI posts looking like AI posts and the penalty that that has.

Indeed, it's not just about the words it uses and the phrases it uses, but that standard formula that you get out of ChatGPT or pretty much any AI that you ask to write a LinkedIn post is recognised as such. And you can start to look for these as a human being as well.

Well, there are lots of rhetorical figures that come up, and each one is flagged. How many of these do you recognise? The single sentence rhythm, short, punchy, hollow, as it says here. We see those quite a bit.

But other ones are the rhetorical questions that you often see at the bottom. Yeah. Do you agree? Let me know below.

Is it from LinkedIn? actually interestingly not linkedin this was uh from independent research but compiled oh the brew 360 is linkedin's algorithm for scoring things now so these are just some

tells as i call them ai tells and so the scoring system's been updated to obviously identify these and rate that in the scoring.

Which leads me on to the new scoring system. It's actually version three now, but we'll go with V2 for this talk. And it covers the lexical tells, but it also covers those rhetorical patterns and identifies those as additional AI tells.

Putting it into practice: Write like a human, then score it

What I really want to get to is to help you maximise the reach of your LinkedIn posts and to do that write it like a human is the best advice I can give you. Run it through and then if it's over 70 publish it.

Does it work? Well I think it does. These are my results. Can you work out where I started using it? can you work out when version 2 came out as a result of the brew 360 hitting me and in fact you see the final uptick there that's v3 coming in so i think it works but don't just take my

Early results and an independent test

word for it i have had feedback from an absolutely independent tester somebody i met at a mindstone event in London talking about this never met him before he went off and tried it the next day with one of his posts and this was his feedback he got more impressions more likes more comments etc within the first 24 hours than he had ever had on any of his posts what was most reassuring though that the demographic, which is this one here, was exactly the one he was after, which I took as a positive.

So, who would like to see it? Thank you. At least somebody does. I was a bit worried somebody was going to say, nah.

Live example: Scoring and improving a real post

Right, can I have a volunteer? So, Ian, are you happy for me to use yours? Right, there you go. Okay.

Sorry. I'll definitely have a look at one of your later. So, oh, there you go. That was your talk from here last week.

Yes, indeed. Well, last month, rather. So, I'll just copy and paste it. I've got Claude up here.

Now, remember that this post optimizer is tuned to my audience, not Ian's. So I'm specifying that he's a friend of mine and therefore not to score it as if it's myself.

What Claude will do with that information is ignore the suitability for the audience and it will adjust the scoring accordingly.

Now I store this one in Notion so it's going It's going to take a couple of extra seconds to do this. Oh dear. I'm sorry.

So, some good, strong elements in there. You know, it's genuinely interesting. And it was, it was a great talk.

Mildly contrarian angle, you know, something useful being discussed there. Short and pretentious. improvements.

So they thought that the hook could be a lot stronger. It's just a pure announcement opener, nobody outside that room cares that the recording is live.

Oh that's harsh isn't it? That was that was Claude, not me, sorry Ian. Yeah

there was no question engagement trigger. Harshest penalty here was actually for for the external link within the body, because we know that that reduces the reach.

Yes, if you want to add a link or a single image, the recommendation is to put it into the first reply to your post. So post it, reply to it, include the link.

Is there a time text now, like a time decay text on that? No, you can do it instantly. recently.

AI tells that were found, well, non -lexical, so it sounds like you probably wrote it mostly yourself, but yes. So it may have picked up a couple of areas there, anti -patterns.

Anyway, it gives a few different options as to how you could turn it into an insight post or or lean into the announcement format, and so on. You get the idea.

So, the core idea is solid. So, that's how it works in operation. I could do a shit sandwich too.

Yeah. Sorry, Ian. I used to have a friend.

The bigger lesson: Research first, then build the tool

So, the real lesson here, it's not so much about the tool, it's about the approach. That's what I just wanted to get across.

You could have jumped straight to building a scoring system, but if you do that, you're missing out one of the key steps, which is the proper research. Get it to do a deep research, first of all, to summarise it, to gain insights from it.

Then you work with it through interrogation and idea exchange and so forth, and then create obviously the scoring mechanism afterwards. 1But use it as a way to bounce ideas off.

Many people just get AI to do something. It's actually really more interesting to quiz it and push it and get it to, you know, critique, to comment, to give you other ideas.

And that's what I enjoyed the most from this process and And it continues to evolve. So the tool is a byproduct.

I find it a useful one and it's quite a fun one to use. But the thinking that this process taught me how to actually work with the AI to get what I wanted from it was the most useful aspect of it for me.

Getting the tool: Where it is and how to use it

But you might want to have a play with the tool yourselves. If you do, it's here, it's public domain. Please grab it. There's a URL. It's up in GitHub.

So if you click watch, then you'll be alerted when there's a new version and so forth. It's on version 3 .1 at the moment. And this is really very straightforward to use.

I know if you haven't seen GitHub, it's a little bit scary. I'll come back to this slide in a sec. The GitHub repository looks like this but, you know, basically you can read through it. What it will have is the actual research but the setup is pretty straightforward.

You just take one of this skill files, sorry let me just go back there. You've actually got all the data behind it and this key skill which is this one here that you just effectively shove into your AI assistant. I did that earlier with chat GPT just to show how it works.

You just copy all of that in there and then as you can see here it tells you how wonderful it is. Why does AI do that? But anyway it explains the purpose of it and if you don't have a LinkedIn strategy profile you just say I don't have one and it will work through a series of questions to understand who you're targeting, ideal audience, which industries and so forth, and help create that for you.

And then once you have it, you can just use it as you saw me demonstrate just there.

Simple. Thank you.