So What? Analyzing Your LinkedIn Strategy with Agentic AI

We're not even bothering with the intro. We're going straight in. Yeah. Uh so thank you for anyone who is still here. So we were having some sort of a strange technology issue with our streaming software.

None of us could get in. We were all kind of scrambling around. For what it's worth, we all try to log in about 15 minutes before the show for things like that. So if you're in, you know, any sort of a speaking role or you do webinars, you know that there's always a tech check. And we do that every week.

However, despite doing that, none of us could get in or troubleshoot. So we finally figured it out. We're a few minutes late, but we're gonna roll with it. So on today's live stream, so what? We are analyzing your LinkedIn strategy with a gentic AI.

I'm Katie. That's Chris, that's John. Let's just roll right into it. Happy Thursday. All right.

Yeah. So for those who don't know, uh we republish our LinkedIn unofficial LinkedIn algorithm guide, uh, approximately whenever LinkedIn creates a big update. And LinkedIn dropped a massive update uh about a th a week ago now, I think is was the about that. The the periods if you want to find it, it's uh trustinsights.ai slash linkedin guide is the URL. I don't know if we ever put a banner on that.

We didn't. I'm gonna make it while you're talking. All right. And so if you've not gone in and looked at this, this is a 138-page guide that walks through how LinkedIn works because as much as people like to say uh, you know, these things are opaque. Well, if you read engineering blogs, if you read the technical papers that underpin them, LinkedIn actually tells you exactly how the system works because there's some things that they don't give away.

Um but there's enough in here that they from an architecture perspective that we can understand the system. So today we want to talk about how would we understand and analyze our LinkedIn performance so that we can do better on LinkedIn, and part of that has to start with how the system works at all. So before we dig into something, I want to ask you two, how do you think LinkedIn works in terms of showing up in the news feed, which is what marketers really care about? I want to be seen in the either individually or as a brand. Um, I think it is fueled on misogyny and bias.

Final answer. The reason I say this is because we have actually found the data, and this is maybe part of the update uh this time with the algorithm, but we've actually seen the data where it does the LinkedIn algorithm favors male presenting profiles over female presenting profiles, which is a big problem. Uh, why is it doing that? Well, guess what? All of this stuff is programmed by humans, and humans have inherent bias in their brains.

So they may not consciously realize they're doing this, or they are, which makes them terrible. They may be subconsciously including uh just their own information, whatever it is. So basically, that's my that's my final answer. What we know is that anytime we think we have the LinkedIn algorithm cracked in terms of what to do, take out your pronouns, change your name, comment more, post more, like more, share more, whatever it is. They make a change.

And so it feels like a moving target, but the consistencies are that you're never gonna see what you want to see. That's how it works. You're welcome. John, John. You know, I have read the Marvel Comics version.

It's like 2,000 posts, enter the Hunger Games, and they all fight to the death. And if it makes it through the first round, it goes into a bunch of safety scanners and stuff like that. And it this is where all the bias and weird stuff is applied. And at the end of that, it spits out a bunch of for me, which is you know, real estate and sketchy deals and a lot of really crummy stuff. So yeah, that's the comic book version of what I read in the latest LinkedIn paper.

Okay. So here's at a very high level how it works. And this is in the notebook LM version of our paper, which if you download the paper, you get the uh notebook LM that goes with it that has all the accompanying stuff. There's three different AI models under the hood. There is the causal LLM, there is Quinn, uh, which is uh in a profile embedding LM, and then there is a uh a transformers-based sequencer called feed SR, which or generative recommender, which is sort of the third stage.

And if we think about this in the um, this is classic old school AI in a lot of ways. It is retrieve and rank. If you're if you've been in SEO, you know exactly what this is retrieve and rank, get the data and then arrange the data in some sort of fashion. So here's the at a high level. First, from things like profiles and posts and comments in any language, LinkedIn gathers up about 2,000 candidates that it could show you.

And it essentially sort of does this weeding with a model called Llama 3, which was created by Meta. They use the 3 billion parameter model, which is a very small model, actually will run on your phone. Then they use it because it's extremely fast. Back in December, Trust Insights published on our blog an evaluation of bias in this model, which is what Katie was talking about, which, as far as we know, we never got word back from LinkedIn saying, yes, we fixed this. So those biases probably are still there.

But that first model says, okay, out of all the posts, here's 2,000 that we think semantically match what a target member would be looking for. It then takes that 2,000 set of posts and then calls a second model called Quen3, uh, the 600 million parameter model, it's a very small model that looks at your profile. Um, it pulls your profile data, turns that into embeddings, and then grabs about a thousand of your last interactions, likes, comments, reposts, dwell time, etc. And with a formula that they did not publish, they say, okay, out of these things, let's form a chronological sequence of your last thousand interactions. We then bring in your profile, the profile embeddings of the people nearest you.

We bring in the semantic match candidates, and we say, okay, out of these 2,000 candidates, based on all this sequential data, what's this user going to interact with? Because their goal is interaction, they're they want you to engage, they want you to like, they want you to comment, they want you to share. And so these systems in combination all say, okay, 2,000 candidates, here's all your past history. Katie, I think you're gonna like this post the most, and so here it is. What's different about this than what is than previous versions of LinkedIn is because is because they're using language models, it's now very heavily conditioned on two things: language consistency and your interaction history.

So if Katie is constantly posting about strategy and digital transformation and change management and organizational behavior, it will see this long history and the language associated with it and say, this is what Katie's about. And so if there's somebody else, say like John, pretend let's pretend John is it uh the well, well, John is the CEO of Marketing Over Coffee, um, and so John's scrolled around on LinkedIn and um is interested in change management. If there's an alignment between these two entities and all their data, John will see Katie's post before mine, because maybe I'm posting about like you know, super tech nerd stuff. If Katie posts about something totally random like football, and that's never been a part of John's interaction history in the last 1,000 interactions or 30 days, John won't see that post, even if John's interacted with Katie a lot in the past, because the there's no like semantic language overlap between the two. That seems like a huge missed opportunity, because so one of the functions of LinkedIn is you can follow someone without connecting with them.

So if I really like, you know, the writings of John Wall, I can click follow on his profile, and theoretically, in my human brain, I'm gonna see the stuff that he posts. But if he's posting about hockey and I'm looking for football, then even though I'm following him, if I'm understanding how this works, then I'm not gonna see his post. I have to seek him out, search him, and say, what did John post about recently and then interact with it, but still not necessarily going to show me the next time I open LinkedIn. Well, and we've had that experience, right? You've gone to like a friend's profile and go, there's like three posts here, I never saw it.

Right. Like how did this happen? And that's why, because it the system did not see a semantic overlap of interest between the two. That and I think that that's the big frustration for a lot of LinkedIn users right now is the machine is trying to guess at what you want to see, despite you saying, I follow these people, I always go to these people's profiles, I always like these pages. So, you know, it's sort of in some way to very oversimplify it, it wants us the humans to be singularly focused on one topic and very uninteresting.

So, you know, God forbid we post outside of what we normally post to sort of like expand our audience, then everybody loses that information. To a degree, yes, that's true, because it all revolves on semantic distance. So it's not exact match keywords, it's it's words that are related in from a statistical relationship. So if I'm talking about, say XG Boost, which is a classical machine learning algorithm, that is semantically within distance of generative AI. So someone reading about who's interested in chat GPT has a higher chance of seeing my XG boost post than a post about football because football and generative AI are much further apart in the training data.

Okay. So it sounds like what you're describing to me is sort of like playing the SEO keyword game. And so if I have my, let's say I have five core themes I always write about, as the author, I should probably do keyword research to figure out what are all the related keywords to a certain point, and then make sure that as I'm writing about these topics, I'm including those keywords. Again, if I'm following you correctly, because this is complex, the LinkedIn algorithm now kind of operates the way that a search engine does in terms of those semantic keywords. You're on the right track, but keywords won't help.

Um because again, uh when we're talking about semantics and embeddings, we're talking about statistical probabilities of words. And that may mean that we're there are words that are not keywords that are that are topically related. So think about topics, not keywords. What topic should I be writing about? Should I be writing about change management?

Well, what's involved in change management? The 5P framework by Trust Insights, the ADCAR, the all these different frameworks, right? If you were to do a keyword search for like ad car, you're probably not gonna, you're you'll it'll take a couple of jumps to get to change management. But if you look in a semantic distribution, a probability matrix, ad car and 5P framework by Trust Insights should be right next to change management because they're all conceptually the same thing, and people will use them near each other in the massive amounts of data that these tools have been trained on. So if you want to improve your LinkedIn performance, especially in terms of showing up, what you would want to do is you would want to get an understanding of who you're trying to interact with and get a semantic understanding of their space, and then take your content and say, how closely does this match this semantic space?

And you want to do this at a person level, not a topic level, because at both at two different checkpoints in the LinkedIn system now, language models are ingesting your profile data. So that's not just I'm a I'm an AI expert and uh all AI all the time, right? I will have things like in my profile. Here's the causes I volunteer for. Like I like helping Bay Path Humane Society, here's the certifications I have, here's the education I have, because there will be semantic overlaps that that matter.

And LinkedIn knows this because it will often suggest you might know these people from your college, you might know these people from this organization. Those semantic overlaps are how it picks up those aspects. So all the things in a person's profile matter, even if they're not necessarily what you talk about 24-7. Okay. So to get into the topic of today's show, like there's a lot of information.

So what do we do with that? Because, you know, and so for those who aren't part of our community, um, over in our free slot community, analytics for marketers, the question of the day was literally just to have people drop their LinkedIn profile so we could all connect with each other. Since we get a bunch of new members every week, it's always nice to make sure people feel like they can make those connections, not pitches, just connections. Um, let's be very clear about that. You can join for free at TrustInsights.ai slash analytics for marketers.

So this is a lot of, you know, you just went through a lot of technical information, which I feel like I have a decent high-level grasp on. Like I could describe it to someone else and feel confident that I was like 99% correct. As an end user, what do I do today? How do I make sure that my yeah, what's the so what? Like, you know, it's all it's great to understand this, but then it's like, okay, but what do I do with that?

How do I make sure I'm showing up? Because I feel like it's not only your profile, but also, you know, the topic is your LinkedIn strategy. So that includes what you're posting, when you're posting, how you're posting, how you're interacting, like what do we do? So when really doesn't matter. So anyone is telling you the best time to post on LinkedIn is Tuesdays at 1 p.m.

Nope. The fact that language models are now in charge, generative AI is now in charge, means that recency really doesn't matter. And this is something that's been an issue now for a year. People have been saying, why am I seeing posts that are three weeks old? Because they're relevant, not recent.

Um, and so uh relevancy is is sort of king at the moment of LinkedIn. So we're gonna look at this from two different ways of what you can do next. We're gonna look at um should we do the hard way and then the easy way or the easy way and then the hard way. Is there like a middle way? No.

Um we'll start with the hard way. Fair enough. So, and this is the part that Katie loves. Katie loves surprises. Uh oh God.

Oh no. So I took the paper that we wrote and all the source papers, and I said, let's build an application that can actually measure two profiles and see what you should do to improve your profile to be semantically relevant based on using the exact AI models. So there's three different AI models running behind the scenes, four different AI models running behind the scenes here that attempt to replicate as closely as possible the current LinkedIn algorithm. There's Llama 3, Quen, there is um a BERT model for the methods, which are uh semantic concepts in in posts, and then there is uh NemoTron Ultra uh Nemotron Super, because we don't have access to the feed SR model, that is LinkedIn's custom baby under the hood. No one has access to anything close to it.

We're using a generic model as a stand-in. So this is not exact. So be very clear, this is an approximation. But what it is, I took my profile, I took Katie's profile, I took my last 10 posts and took Katie's last 10 posts. I put them in here and I said, I want to improve my profile.

How close are Katie and I in terms of what we talk about? Uh in terms of who we are. So Llama, remember I said that's the first gate model that says how close are these two uh, you know, these these people in the interactions says there's like a we're like a 96.8% overlap. Katie and I are very closely aligned. In terms of the content we share, we're 98.4%, highly aligned.

We talk about very similar things. In terms of the language we use, 95%, and the composite score, 96.5%. So there's a high probability. This is already saying, there's a high probability I will show up for for Katie's in Katie's feed because I'm so well aligned. And then it says, hey, um, it's going to try to essentially for I'm using the profile here, I'm just gonna try to add and rewrite parts of my profile to get an even closer match.

So if I wanted to specifically target a certain kind of buyer on LinkedIn, I would instead have Katie's profile, I would have, you know, uh the CEO of Apple, Tim Cook, right? I would have Tim Cook's posts and Tim Cook's profile, and it would say, if you want to close that gap semantically, here's how you should change your LinkedIn profile to match. And then if I wanted to, I could put a post in here and put and copy and paste that post in and say, okay, now if I you know if I want to show up in Tim Cook's feed, what, if anything, could I do to modify this post to be more semantically aligned with who Tim is and what he cares about? So that's interesting, but that's going one person at a time. Could you do something like uh run your ICP, your ideal customer profile, and get the technographics and the firmographics and the demographics and all those things more in a cumulative way.

Put that information here as the target profile. Because it's hard to do it one by one because then you're sort of like changing your profile every other day. But like could you do it in a more cumulative way with the ICP and say, what things do I need to have in my profile so that I show up for a larger potential audience. The problem with an ICP is that it's an averaging out of a bunch of different characteristics. So for example, uh university names uh would be a good example degree names, diploma names, charity names and stuff.

Uh a Tim Cook is going to look different from like a Larry Ellison, uh even though they're both high-tech CEOs, right? And that would be your ICP would be the high tech, you know, Fortune 50 CEO, those two individuals are going to be very different. So what you would do, which is some thematically the same, is you would go and grab those five LinkedIn profiles, you know, Larry Ellison and Mark Benioff and so on and so forth, grab their profiles, grab each of their their last 10 posts and you would use a tool like Claude Cowork, hence the agentic AI in today's show title, to programmatically take control of your browser and gather up their posts and store them. And then you would instead of having one person's profile and conversation history, you'd have the group's conversation history and the group's profile as the target. And then you could do your calibration against the group as a whole.

So not quite the ICP but almost a similar idea. Exactly. But with all of those weird little nuances in people's profiles that you would want to that you would want to do, um, that you want to have access to. So this is the hard way of doing it. Um because this this uses the exact model.

So it's going to give you the most faithful replication of what LinkedIn does for people who are not LinkedIn employees, essentially. Now, if you want to do something that was more practical, um, you would fire up a tool like Claude Cowork, and you would say something along the lines of, I want to build a semantic understanding of these two profiles. So let's start off with a basic prompt. Say today we're going to do semantic comparison of two users on LinkedIn, Chris and Katie. I'm going to provide you with their profiles in text file format and the last 10 or 15 posts from each person, each in JSON format.

So that's the starting point there. That did not transcribe, did it? Oh, there it is. Um what you will do is write a small piece of Python code that will compare using cosine similarity and BM25 similarity, how close Chris and Katie are, and then provide some suggestions for Chris to adapt his language to more closely resemble what Katie is semantically relevant about. So there's our basics.

And now we should probably put a uh final thing. Your final output should be an HTML file styled with CSS that's easy to read and prescriptive, telling Chris exactly what he needs to do to be found more semantically relevant by Katie. And we're gonna do that. I like how this is the simple version, but you still had to understand things like I'm looking in your post for like BM25 and uh cosine similarity. Like you still have to understand that that's part of it, yet this is the simple version.

This is the simple version, and this is part of the reason why there's so much snake oil on LinkedIn about oh, the ultimate LinkedIn algorithm hack, just do this, because there is no just do this when you're dealing with language models. You're dealing with very complex systems. But if we understand how those systems work based on what LinkedIn has literally told us in their technical papers, we can we can use today's very, very smart generative AI models to get to the closest approximation possible or to approximate what we know the outcome is. And we know the outcome is semantic similarity. And if you've been in the linguistic space, the SEO space, uh, the GEO space or whatever, you know that semantic similarity is sort of the the heartbeat of all those systems in terms of how do we figure out what language these things have in common.

So, Chris, while that's running, we have a question that I'm not a hundred percent sure uh we can answer, or really quite honestly what the question is, but it's sort of a two-parter. So, first part: if you had to rank semantic understanding, member profiles, content posts, last 1000 actions of the author reader in terms of what has the lowest to highest leverage. Yep. And then um, go ahead. So the answer to this question is use the algorithms that LinkedIn uses.

Um, so LinkedIn use they stated this in the causal LLM paper last fall, which you can look up on archive.org. They use Llama 3B and they specifically use Matrioshka embeddings with a 3,072 dimension um embeddings. So you all you need to know is the words to put into Claude to say, I want you to help me install Llama 3, 3 billion parameter version, 8-bit quantization. I specifically want to use it for embeddings. I want to generate Matrioshka embeddings.

I want to generate 3,072 dimension Matrioshka embeddings to compare pieces of content. And effectively, what you're doing is you're using the vanilla version of the same model LinkedIn is using in production, but you're doing it with you know your desktop computer instead. So in terms of what gets you leverage, what gets you leveraged is higher similarity. So if you have one pool of embeddings that says this is all about change management, you have another pool of embeddings that says this is all about college football, those are pretty far apart, and so it's gonna be difficult for there to be enough overlap in those topics for LinkedIn to go, Well, should I show this post? Nah, there's nothing here.

All right, so um, John, I hope you were taking notes. There's gonna be a quiz later. The second part of that question is uh do we know how those one to two thousand candidate posts are shortlisted from a bigger pool? And then further down, which one we will see first? Right.

So uh we know how those are shortlisted, those are shortlisted by using the Lama three, the Matrioshka embeddings, it's processing everything in the LinkedIn system. That's why it's such a small model, it's a 3 billion parameter model. 3 billion parameter models are small if they can run on your phone. But LinkedIn uses that because it has to process everything. When you use regular generative AI like Chat GPT or Gemini or Claude, you are using models that are 10 to 15 trillion parameter models.

And that's why they're so slow, right? When you type in a Gemini, you're like anytime now, Gemini would be would be great to have the answer. When you use a 3 billion parameter model for embeddings, because it's such a finely focused task, it can process all of that. So that's how it's processed. Now, how do they know which you see first?

That's what phase two is. So it's retrieve and rank. So the phase two is that ranking to say, okay, now you got 2,000 posts to sort through. How do we sort them? So Quen pulls your profile embeddings.

The feed SR pulls in your thousand last thousand interactions, and it essentially kind of does a forecast saying here's profile similarity, here's past history, here's the content. If you were to do a statistical prediction of what post is going to have the highest call, the highest score of those three different measures, that's what shows up first in your feed. That would that's an oversimplification, but that's kind of how it works. So let me ask you this question. Um, and maybe it's a question for John first.

So obviously LinkedIn has tools like the sales navigator, you know, add-on tools that they want you to purchase so that you can do more advanced searching, you know, for you know, prospects and lead generation and you know, whatever kind of targeting you want to do. Does this version of what you're saying, like you know, that we're running in Claude Co-work, is this sort of the I don't have the money to spend on a sales navigator, therefore I can do this instead. Like, where does sales navigator fit into it? And John, I guess my question for you is as the head of business development, is this something that you were like, oh yeah, I could probably do that to appeal to a you know our wider audience because your profile, I haven't to be honest, I haven't looked at your profile because I know you, you know, but like do you feel like it would, you know, accurately represent where you are today versus you know, uh maybe five years ago. Yeah, you know, there's so much going on with this.

As far as I mean, sales navigator is kind of just a whole separate thing. I look at that more as um data enrichment and you know, being able to just kind of put together lists of people and giving you some additional access to people. Like that's part of it too, just being able to message and hassle these people. Um yeah, I don't know. I kind of look at this as a whole different thing.

This is more of a content play. You know, this is like how do we post stuff that gets in front of the people we want to get in front of? And how do we make a convincing case? But I, you know, I'm interested to hear more on that because so far to me, this is just all black box. Like we don't have any of our even our own last thousand interactions that you can easily get to unless you're doing some kind of extra exporting that's probably against the terms of service.

So yeah, I don't know. It's kind of you can run a lot of models and get a pretty good idea of what should be done, but you're you'll never be definite as far as like what works and what doesn't work. You can't promise that to anybody. Um, but yeah, but and I'm also hoping that Clippy is gonna, you know, help guide us along and show the way too. Here, I have to give him a shout out as special guest star because we haven't talked about him yet.

Okay, so flawed co-work finished up an assessment. Now, cowork used a different similarity method. They use what's called TFIDF, term frequency inverse document frequency. This is not the same as embeddings. This is a much older version of natural language processing.

And so you can see, because it's that much older, it's word-based, sort of that keyword-based that you were almost talking about, Katie. There's a much bigger gap here than when you use the models that LinkedIn actually uses, uh, which is was the one that we built into our tool. So this is kind of a actually a really great example of the difference between going keyword-based, which says, hey, things suck, versus embedding-based, where it says actually you two are are very have a lot of strong overlap, but there's still useful things in here. So it came up with here's your theme distribution, where uh who does what? So Katie talks a lot about leadership and change.

Chris, not so much. Katie talks a lot about people and process, Chris not so much. Katie talks about framework and method, Chris not so much. AI strategy and business, Katie and Chris talk about a lot. Data and analytics, Katie and Chris talk about a lot about comms and content.

Chris talks about more than Katie. And so there's kind of this nice little skew. Here's the vocabulary gap. Again, this is keyword-based, not embedding based. You can see where are the gaps that the if I wanted to close this older semantic gap, or at least get some ideas about what I should maybe include in my posts.

These are the the blatant terms that Katie uses that I don't, right? So leadership, product strategy, team, people, architecture, decisions. I talk about none of that. I'm talking about like, hey, I just found this cool thing in Clawed cowork, right? Or uh, you know, we built this this funny thing.

So that's an interesting little gap. And then, you know, sort of head to head, which terms uh are frequent for both people. And then it gives me some prescriptive recommendations. If I want to talk more like Katie to attract her attention, I should instead of saying, here's this is one of my posts. Here are three ingredients for success with any significant project.

I would start by saying the leaders who actually see ROI from AI are not starting with code, they're starting with the decision document. So it's it's still the same thing, but couched in the language that Katie uses more than I use. So even this is helpful to see, all right. Well, I guess I would have to, and I'm not going to, but if I I would have to change how I write to focus more on the things that would fit in Katie's language world. So I started a post saying agentic AI switches system two to system one, right?

And it reframed in Katie's language when you take the thinking out of a task and hand it to an agent, the people on your team stop building the instincts they'll need when the agent fails. That's a leadership problem, not a tech problem. And I'm like, oh yeah, I totally sound like I would have written that. Exactly. I'm like, wait, does I write that?

Sounds like something, but no, what I like, what's interesting about this, you know, and obviously you need to find the case or the use cases for it, because you, you know, just modeling yourself after someone who's already out there, you still have to find your own voice. You don't want to just replicate somebody else who's out there because then you start to get into IP infringement and lawsuits, and it's ugly. But if you're trying to get the attention of, say, a prospective uh hiring company, so let's say I want to go work at McKinsey, done with Trust Insights, you guys are nerds, I'm over it. I want to go work at McKinsey. I could take a look at theoretically, if I'm following this, the McKinsey business profile, take some of that information about what they've written and say, well, I write about similar stuff.

How can I, you know, start to use some of their language so that I'm a more appealing candidate to them when they're making those business decisions about who to bring onto the team to demonstrate that I understand their culture, I understand the work that they're doing. Like I can see that being a really good use case for modeling yourself after someone. Mm-hmm. There, the use case is a couple of really important use cases that you just touched on. Number one, if you are looking for work and you know who the hiring managers are at the kinds of companies you want to work at, you go and you grab all that stuff, you follow the exact prompt that we just used in clawed code and say, help me understand the kinds of language I want to be using.

Because look at this action button. This is the if I wanted to uh attract Katie's attention in Katie's feed, number one, trim the old crap in my profile. It's not helping my similarity scores. Two, write some content about leadership failure framing because that's kind of Katie's jam. And if I can get if I can figure out how to say it in my voice but touch on those things, great.

Number three, you know, maybe change my my uh LinkedIn summary. I won't, uh, but I could uh if I needed to. You know, here's so this is a uh a nice little 30-day action plan. So that's use case number one. Number two, if John was on uh had a deal on the pipeline, and let's call it like a million dollar deal or a billion dollar deal, it's a one billion dollar deal in the pipeline, and John's like, I have got to be seen in this in the LinkedIn feeds of all the C-suite of this company for the next 30 days while they are deliberating.

John would take those C-suite profiles, run this analysis, and then just start cranking out stuff that is the highest probability that would be seen by that group of people to just just to show up. This is not something you know you would do for every single profile and stuff. But if there's something that we're where you gotta throw everything in the kitchen sink to win that deal because it's a billion dollar deal or a trillion dollar deal, this is what you do. I think that that's a really smart way to use this because you know, all of us, small businesses, large businesses, we're all trying to appeal to our prospective customers. And we can't change everything about our profile all the time, but what we can do is create the right kind of content, which goes back to first you need to understand who your ideal customer is.

Um, which, by the way, if you want help with that, go to trust insights.ai slash contact. John will hook you up. We just redid the services page. It's lovely. The uh underlying methodology has not changed.

We just made it look a little nicer. But I really like the idea of, you know, it's sort of like that next level of appealing to the right people. Like we talk about how to show up, but once you show up, are they gonna stick around? Mm-hmm. Exactly.

What is interesting is um it says that you and Katie are saying the same things, you're just saying to different people in a different dialect, which actually for what we do as Trust Insights is fantastic because we want to appeal to different people. We don't want to be going after the same people. We actually want different networks. Um your dialect is precise, technical and practitioner facing hers is operational organization and leader facing, uh, which is again is fits who we are. Mm-hmm.

Yeah, no, and I think that that uh what's nice about using us as the examples is we can look at this and go, yeah, that's correct. You know, you know, in your example, like you would be appealing more towards me. But in reality, like I, you know, I know where your stuff is, I know to look for it, I know your general publishing schedule. So I'm not worried about you showing up in my feed. But, you know, when John posts, who's admittedly a less frequent poster than you or I, just because the nature of the work that he does and what he's working on, he's not the one generating content.

You and I are. I want to make sure I don't miss someone like John's content when he does post. And so, you know, it's if we didn't work together, like I would be like, okay, how do I make sure John is showing up in my feed? Because when he does post, it's important to me. And I think that for people in John's position, and John, correct me if I'm wrong, it's a great way for you to be keeping an eye on, you know, to Chris's point in that example.

If we have a million dollar deal, you want to make sure those people are showing up in your feed repeatedly so you can see what they're talking about, what their pain points are, if they've made a new hire, if they've brought on another agency and they just never bothered to tell us. Those are things you probably want to know. Yeah, it's definitely uh, you know, it's the prime place to stock these contacts and see where they're at. Um, and another, you know, side door on that too is to see who else they're engaging with, whether it's competitors or other things. Um, yeah, I I'm not going to talk too much about that, but you know, if someone was enterprising, they would be able to go find all the prospects for competitors and you know, pitch more appropriate content and get a better fit than everybody else that's out there.

Yep. And so if I would say at the highest level of this, you would end up with a system where you have, like, for example, our friend Ashley Zekman does a ton with influencer marketing, our friend Justin Levy does a ton with influencer marketing. At the highest level, you would probably have cohorts of pre-gathered curated data for like the major influencer segments in a system like this, in this code version here. And then if you brought on a client uh that wanted to do influencer marketing, you would say, okay, we're going to use our curated uh artisanal, organic handcrafted uh influencer segment for you know this sort of executive or business you want to reach, and we're gonna give you an audit of where you are and the and you know, take your editorial calendar for your next 30 days, and we're gonna put it through this system, and we're gonna make sure that it aligns with from from again using the exact same technology LinkedIn's using, get it aligned um to make sure that it works. So that's how you use the LinkedIn algorithm, which is several algorithms in concert.

Um, you you cannot game this system. I think that's the other thing that I I want to beat very loudly. You cannot game the system. There is no secret hack that will make LinkedIn work better for you. Because one of the things that LinkedIn published in this totally separate paper on infrastructure is that the weights of the models, the internal calculations recompute hourly.

Um, and they totally flush daily. So you may discover a hack that works for like 12 minutes and then it's gone because the system has recomputed its weights. There is no gaming it. What you should be focusing on is stuff that profile coherency, content that's targeted towards the audience you care about, and behavioral consistency over a 30-day window. If you read nothing else from the paper, that's the three things that you should be focusing on.

And you can get that paper for free at trustinsights.ai/slash LinkedIn guide. Um, I want to acknowledge um our friend who asked the question. I think I get it. Thank you. Sorry if the phrasing of the question was off.

I was trying to absorb what Chris was saying. Same. We are all in the same boat of information overload. I think you did a fantastic job. Um, no, and that's the thing is it's a lot, it's a lot of information to try to understand, which is why Chris distills it down into a guide for you.

Because here's the thing: a lot of these major technology companies, because that's what they are, they're tech companies, they put all the information out there. They're like, what do you mean? We're not keeping it a secret, we're not black boxing it, but they don't write it in a way that the average user think people like me and John can understand. So we rely heavily on thought leaders like Chris to be curious enough to go find that information and distill it down for the rest of us. So if you haven't checked it out, like honestly, it's a ringing endorsement, not just because it's from the company, but because Chris does an excellent job on this particular paper every time LinkedIn makes a major update.

So go get it. Trust insights.ai slash LinkedIn guide. I can't think of a better resource that breaks down what's happening in a less technical way. I'm not going to say non-technical because there's still a lot of tech in it, but in a less technical way, so that you can try to understand it. Hey, give the paper to a large language model and say, summarize this for me.

Um say write it in terms of ordering pizza. Like what are some of the great things that you like to do, Chris, that sort of make things a little bit more distilled down and easier to understand? Pizza every time, because most people know what pizza is. You would think so. But like I find most people can figure out how it's made.

But to that point, find find the thing that makes sense to you. So if you are someone who loves horseback riding, you know, take this paper and describe it to me in terms of, you know, a horse event, or describe it to me in gardening terms, or whatever it is that makes sense to you. That is one of the beautiful things about a large language model is you can take these highly technical concepts and break it down in a way that makes sense to you. But first, you got to get the paper. Exactly.

So go get the paper. And one of the things in the guide is four pages of citations uh so you can find the underlying sources so that you can do your own version if you want to. I I don't know why you'd want to, but you could. Um, yeah. So that's that's it.

You we we use clawed co-work today. You can use any system. So Google Anti-Gravity, OpenAI Codex, Claude Cowork, Quen code, Claude Code, any system that allows you to put in a bunch of data and just programmatically generate code and and fact check it stuff. Um do it in those systems. You will save yourself so much time.

The hardest part is getting the vocabulary down to understand these are the things that I should ask for, and that's where a lot of people do run into trouble in general with AI, but especially with with something like this, which is very highly technical. You want to have that the vocabulary base down so that you know what to ask the models for and what kind of outcomes that you're after. So, any final parting thoughts on this uh very interesting topic that uh and and a little bit of a crazy live stream. I think the big takeaway is that you know, algorithms are a moving target. They're you know, to your point, there is no gaming it.

The best you can do is try to understand it and find the three things that you can do that are gonna make your experience on these platforms more enjoyable, more meaningful, more connected. John, it's in the book. Check the book, and the book is free. Yeah, go get it. Somebody did comment, they're like, Well, I don't understand why you don't charge this.

I'm like, well, maybe it is we charge go ahead. I was just gonna say it's going like crazy. We get real-time notifications. We've got everything set up here, and we just sit here all day, and Slack is just like copies of this are flying out the door continually. So it's yeah, it's definitely roadproven.

There's a reason we're re-releasing this every time they do the updates. Yeah. I honestly, you know, to answer that question, I think that we're not in the business of gatekeeping. You know, obviously, like we have our services, we have our courses, but this kind of information, it benefits such a wide audience, and it really helps. Like the name of the company is Trust Insights, and we need people to trust the insights that we're giving them.

And this is a great proof point because, you know, if we start to charge for this, it sort of loses that credibility. Like we're just trying to make an easy dollar. But, you know, this is just one of those things that I think is just so helpful and useful and valuable that I'm I'm actually grateful that we do give it away for free because so many people benefit from it, including ourselves. Exactly. And hey, we write our own software for it.

Maybe maybe we'll charge for the software. All right. That's gonna do it for this episode, folks. Thanks for tuning in, and we will see you all on the next one. Thanks for watching today.

Be sure to subscribe to our show wherever you're watching it for more resources and to learn more. Check out the Trust Insights podcast at TrustInsights.ai slash TI podcast at our weekly email newsletter at trustinsights.ai slash newsletter. Got questions about what you saw in today's episode? Join our free analytics for marketers Slack group at trustinsights.ai slash analytics for marketers. See you next time.