So What? Using Generative AI to Build Marketing Plans

So What? Marketing Analytics and Insights Live

airs every Thursday at 1 pm EST.

You can watch on YouTube Live. Be sure to subscribe and follow so you never miss an episode!

 

In this week’s episode of So What? Christopher Penn, John Wall, and Katie Robbert walk through Google’s document leak & show you how to create a marketing plan from technical documents using generative AI.

Catch the replay here:

 

In this episode you’ll learn: 

• How generative AI can process technical documents to produce useful marketing plans
• A prompting process to develop your own marketing plans from any document
• An analysis of the Google SEO leak and LinkedIn’s algorithm with recommendations

Upcoming Episodes:

• TBD

 

Have a question or topic you’d like to see us cover? Reach out here: https://www.trustinsights.ai/resources/so-what-the-marketing-analytics-and-insights-show/

Please note the following transcript is AI-generated and may not be entirely accurate:

Katie Robbert 0:36
Well, hey everyone, Happy Thursday. Welcome to so what the marketing analytics and insights live show I’m Katie joined by Chris and John who cannot high five this week, they would sort of do a almost, that’s weird. This week, we’re talking about using generative AI to build marketing plans. So we’ve been talking a lot internally generative AI is a cool tool. But that’s really what it is, at the end of the day, it’s just part of your tech stack. And you still have things as a marketer that you need to do like build marketing plans, figure out, you know, who your audience is, how are we going to sell this thing. And so today, what we want to do is walk through how to use generative AI. To process technical documents to produce useful marketing plan. There’s a lot of documentation that’s come out recently, that we want to focus on a prompting process to develop your own marketing plan from any document. And so we’ll give some examples of documents. But there’s other documents that may be more suited to you and your industry, and an analysis of the Google SEO leak and LinkedIn algorithm with recommendations. And so Chris is gonna get into what the Google SEO leak and LinkedIn is. So Chris, where would you like to start this week?

Christopher Penn 1:51
Let’s talk about the Google leak and stuff because I know we teased it in a couple of the different slack groups and things for folks who are not aware over this past weekend as weird being have been on the show in a couple of weeks now. Over the past weekend, Rand Fishkin over at Spark Toro who many folks remember from his days, he was the founder of Moz. Got, someone got in touch with them, saying, hey, there’s been this massive leak of Google’s internal API documentation. So what was leaked was 1000s of pages of data about Google’s content warehouse. So to summarize, a lot of stuff, Google has sort of this monolithic, massive data processing system that controls what it ingests, how it processes it, and then how it makes it available to different systems at Google, like Google search. And rands article just touched on, essentially, you know, the sort of the surface level stuff in this and some of the things that that he in particular, was a little appropriately incensed by because he had been making claims for years, for example, it’s saying like, yeah, Google, clearly seems to take into account click behavior. And Google’s like, no, no, we don’t take any any click behavior at all. And then this, this documentation leaks, and very clearly in the system says, hey, when you take into account click behavior, that combined with some testimony from the lawsuit, the department justice lawsuit, Google’s actually we do take into account click behavior.

Katie Robbert 3:28
So let’s back up a little bit. So Google went through a major overhaul a couple of years ago, in terms of what we as marketers understand the algorithm to do. And I believe there were three major things that we were advising marketers to pay attention to, in terms of how, how to rank better in search results. And that was, what the helpfulness of the content, the, the relevancy of the content. And what was the third one helpful was relevance and freshness, freshness. And, you know, this is because the public facing documentation that we all collectively had, that’s what we were told by Google, that’s all that matters. You don’t have to worry about anything else. You know, everything else that you are, you know, on your plate, forget, it doesn’t matter. Just focus on those three things, and you’ll be fine. Like, that’s what we’ve been told. That’s the Kool Aid we’ve been drinking for the past couple of years. And so understandably, Rand and the rest of us should be incensed because we’ve been doing the wrong things. We’ve been spending money with Google, they’ve been taking our money, and we’re not even doing it right. So we they set us up for failure. This is a big deal.

Christopher Penn 4:49
This is a big deal. And so Rand did a great job, sort of summarizing it Mike King over at, I pull rank has a much longer than that. Alice’s are some of the surface level stuff. However, you can go out on the internet over a hex docs, which is a site that makes archival copies of GitHub repositories. And you can look at this, this leaks content yourself. Here’s the challenge. It’s over a million words long, 2500 pages of stuff, just the summary page here is 200,000 words long. So all of the analysis that folks have been doing, has really only started to scratch the surface of the of all these different applications. And so what I thought we would do today is something I did over the weekend, when I once I found this repository like, Oh, this is a great use case to use generative AI because A, it can read the whole document. And be, it can connect the dots in ways that you and I can’t, because we just can’t keep that much text in our heads. So I figured we might do a little bit of an exploration of that today. But I wanted to first give full credit words do Rand did a great read of my team did a great write up. And their conclusions do support a lot of the big, I guess, headline things like the age of your domain does matter. You’re put in a sandbox for 18 months. So if you run by a whole bunch of new domains, it’s going to take 18 months to for them to start ranking. So let’s so any questions before we dig in Katie?

Katie Robbert 6:26
So many, so many questions, I guess the big question, and this might be just, you know, speculation on our side. And, John, I would love to get your perspective on this as well is, why would Google lie to us? I mean, I’m sure the, I’m sure the easy answer is money. But why would they lie to us about this?

John Wall 6:47
Yeah, that’s there’s a whole kind of wave of stuff that I’d seen kickoff about a year and a half ago, where people just realize that it’s cheaper for Google to just have their PR team say, Well, this is how the thing works. It does X, Y, and Z. And it’s cheaper to have the PR team do that than actually, you know, get developers to make the algorithm do that, for it to be real. It’s cheaper to just spin a story and let people believe it. And so yeah, there’s been a lot of, because in a bunch of these SEO groups that I’ve been in, like they’re, they’re on dark social, the rest of the world can’t see them. But they’re like, Oh, hey, look, I tried putting some white text on a white background. And my page still got indexed. And you know, it got in there. And all these things that for years, Google has been saying, yeah, no, it doesn’t do this. Actually, it does do some of that stuff. Another big one is having mechanical Turks, people who spend their whole day like going through websites and rating them as to what they can do. You know, supposedly, there was none of that. But these guys have seen that they even got an employee in to do it and do the page ranking. And he said, Look, if you get your site optimized for the 20, things that these people look for, your rankings will improve, regardless of any algorithmic stuff. So yeah, you nailed it, it’s, you know, follow the money and you get there.

Christopher Penn 8:05
Yeah, it’s money. It’s also the Blackhat stuff. If people know how the system works, they can game it.

Katie Robbert 8:14
Okay, all right. All right.

Christopher Penn 8:16
So let’s get started. The first thing we’re gonna want to do is you’re going to want to go to the hex Doc’s repository. Yeah, I had to Google for it. But if you just Google for Google API content warehouse, Hex Doc’s dot PM, you’ll be able to find the repository. It as with any API documentation is broken up into two portions. One is sort of the broad descriptions to the individual modules. And there is no way to make this any bigger, because it’s just so convoluted. However, if you scroll down all the way down to the bottom, you can actually download an epub and ebook version of this if you want to have a reference. And I actually have done that. Because I wanted to, what you would do next is just in very straightforward, old school olds, old old fashioned, take this whole page, copy it, and paste it into a text document, that’s gonna be an easy way to do to get a hold of this. This is this API Summary page tells you what all the modules are, and what their functions are, what they do. And this is all code stuff. So to give a bit more insight, Google has built essentially, a massive system of internal API’s that all connect to and from this content warehouse. And so what was leaked was the documentation that Google employees use internally, to remember how the heck to talk to their own systems. So internal developers, or Google will reference this documentation to plug things into it or to extract information out of it. Okay, so let’s go over to the iron Google. Let’s go to Google Gemini.

Katie Robbert 9:58
This is I’m feeling some irony I’m feeling a little uneasy about this.

Christopher Penn 10:03
Exactly. And the reason we’re using Gemini is because of its context window. It has a short term memory of 1 million tokens, which you can see here in the corner, which is about 700,000 words. Others Anthropic Claude three has a context whenever similar. ChatGPT has context window, I believe it’s still only 120,000. So it’s 90,000 words, so it can’t handle nearly as much information. I’m gonna switch the model.

Katie Robbert 10:26
What are the odds? So you’re using Google Content warehouse, Google Gemini? With Google leaked documentation? What are the odds that the generative AI says it was gonna be like, Oops, that information doesn’t exist and delete it right there on the spot, like, just from your looking window, and it’s gonna be like, You didn’t put anything in here? You were imagining the whole thing.

Christopher Penn 10:49
Funny you say that. Because in regular Gemini, I tried to load the webpage. And the first thing it spit up was when I tried to run it was I’m a I’m a text based AI. I don’t know how to do that. Even though it can jump out I can clearly read other pages. It took several tries just to even get it just partially look at the page and it still won’t. So the consumer version says I’m not programmed persistent that I can’t help you with that.

Katie Robbert 11:16
I guess that is exactly what I thought was gonna happen that that’s why it makes sense. Like, they they’re probably pooping themselves right now. Like, oh, my God, this is awful. This looks really bad for us.

Christopher Penn 11:30
Yes. However, AI Studio is a separate interface with with a separate set of systems. Switch the model from flash to pro because slash is dumb as a bag of hammers. It’s good for summarization is not really good for more deep analysis. And as I always do with it with Gemini in development, warning development, I turned the safeties off. If you’re using this code in production, please do not do this. This is an unsafe practice. The first thing we do, always is we prime the model, right? So we say we’re going to examine the API’s documentation. So what are the best practices for reading through an API? Right, we want to get Gemini to load an understanding of how to think about API’s. This is important so that as we feed it actual API documentation, it will know what to look for.

Katie Robbert 12:24
Okay.

Christopher Penn 12:25
Okay so, and if you are unfamiliar with this step, we have I put it we have a download a free download, no form required. Trust us as far as power questions for the pair process, we just have the pair, y’all should just put them here and some one of these days. That is a a explanation for why we’re doing this. Alright, so we’ve gotten Geminis own internal guidance about how to analyze an API. Next, I’m gonna say, Hey, here’s the API overview. But upload to Drive. And from Google Drive, I’m just going to take that big ol plain text file of the front page from hex docs. And we’re gonna let it chew on this massive document is 192,000 tokens. So it’s about 100. So D duped and digested down. That’s what about 150,000 words of just of just this page D duped that says, hey, this documentation is the Google content we ask presents the vast and complex system with numerous interconnected modules, modular structure, modular interconnections. And then it starts listing out all of the system components in here focuses on Google products and stuff like that. Debug overall picture this is this is useful. This is a helpful summary. Say, can you make a list of all the major model inter connections? I want to know what those model pieces are. Because this is going to tell us at a at a high level, or it should tell us what the big chunks are, what the key modules are in this package. Let’s see we have abuse verdicts. This could be a while.

Katie Robbert 14:25
You know, the topic of today’s episode is you know how to build a marketing plan. I quit. I’m imagining that this leak documentation is going to radically change our marketing plans in terms of SEO, because we’ve all been doing it wrong.

Christopher Penn 14:46
Exactly. Oh, look, social graph social interaction. So remember, Google has been saying for years we don’t take any social media information into ranking at all. It has nothing to do with With Google, you don’t pay any attention to it. There’s social media stuff in here. Weird. We have web documents, indexing knowledge graph of videos and media. Okay, that’s pretty cool. That’s pretty helpful. I want to know. Dig deeper into the social modules. Google has said for years that social media data is largely ignored. Is there evidence within the API documentation that indicates Google is working with social media data, especially from third party social networks. This is the power of generative AI. We don’t have time to read through 150,000 words that’s, you know, for context. And handle his book, everybody writes a 75,000 words, this document is two of these. That’s how large this this this is. We don’t have time to read through that. And you’ll probably need like five Red Bulls just deal with it. While it doesn’t explicitly several inter reveal significant integration, multiple social platforms, both Google’s own and third party ones.

Katie Robbert 16:26
I like how it starts with your right to question Google statements.

Christopher Penn 16:31
Hmm. So Google Plus is still baked in their IDs, Facebook, Twitter, Instagram, family status, hidden key fields, third party app support and type YouTube comments. This contradicts Google’s public statements about ignoring social media data, raising questions about transparency, the actual extent of social data uses within their own system.

Katie Robbert 16:53
So I guess the question, so what do we do? How do we what do we what do we do? Um, so I just, Oh, my God. The red?

Christopher Penn 17:12
Yeah. Oh, there’s a lot of questions we had, let’s ask the question, based solely on the documentation, what can we conclude by inbound links, right, because we’ve been hearing about inbound links. Are they important? How important are they? Google has also made the statement over time like, yeah, it’s just one of 200 ranking signals, you know, focus on building good, great content. That’s that’s been sort of the refrain for 20 years now. And to be clear, that’s not a that’s not wrong advice. Like you should still have content in a site that doesn’t suck. Sure. However, I think some things are more important than Google has said. Based on headstocks, we can conclude the inbound links to a website also known as backlinks play a crucial role in Google systems impacting several aspects of the functionality. Here’s evidence from the documentation. The existence of a dedicated anchors module indicates the importance of link data center signifies a centralized system for storing and processing backlinks. Detailed information, hey, look, there’s PageRank that they said has been gone. There’s a statistical analysis module, there’s representations of scoring signals. There’s boosts for user data for user interactions with the data. So this is a really interesting. You mentioned, boosts. And it looks like there’s something called the motions from the API documentation. Make. And you’ll note I keep saying from the API documentation, I’m telling Gemini, do not infer other stuff from your general long term memory focus only on the data that I have given you to force the model to focus just on on what we’ve provided because otherwise it will hallucinate. Make a list of the boosts and demotions in the content warehouse API, understanding that we have the modules but not the model weights. One thing I think it’s important to say is this. This documentation is essentially sort of the model dimensions if you will, if you’re if million Google Analytics, you have dimensions and metrics, metrics are the numbers dimensions are the categories. This documentation is essentially the dimensions. We don’t know the weights. We don’t know how much each feature counts for, but the very existence of a feature indicates that there’s at least some numerical value to it. So this kind of like saying, We know all the variety of ingredients on top of the pizza, we just don’t know how much of each ingredient there is, but for sure, if pepperoni is listed, there’s probably pepperoni on the pizza somewhere.

Katie Robbert 19:51
So I think foot depending on your role in the organization, digging into specifics of the document makes sense, you know, so I want to know more about social or I want to understand more about, you know, the anchor or, you know, the the booths. Is there, like, can you directly ask Google Content warehouse to say basically like, what are the 10 things I am supposed to focus on? Is that the kind of information you can get from this particular leak documentation? Absolutely.

Christopher Penn 20:30
In fact, I wrote a prompt over the weekend that does exactly that. And it goes like this. We divide SEO typically to four areas of practice, right technical stuff, that’s is infrastructure, server speed on site, things you’re doing with aspects of the content, stuff like schema markup, there’s off site, stuff like inbound links, link building. And there’s content, like the quality and quality of the quality of the content itself. Given this general framework, relying only on the API documentation. Let’s build a list of SEO best practices from the documentation in these categories. Return your best practices in an outline format, because we want to be easy to read. And this is this is a critical part of the prompts. provide specific module or function calls as supporting evidence from the documentation. This will reduce hallucinations substantially because Google has to reference its own data. If no supporting evidence from the API documentation exists omit the best practice again, we’re saying we want to focus in just on the things that are in the documentation. So this is exactly where you’re going, Katie, which is we have this big ball of data. Yeah. And we turn it into something actionable.

Katie Robbert 21:45
Yeah, well, you know, because I mean, you’re talking about Google has been hiding how search actually works. And the number one question we get from clients is, how do I improve search? And so, you know, I’m sort of speaking in general terms, but they don’t care about what’s in the documentation, they don’t care about, like, they just want to know what they need to do, that they’re not doing, or stop doing, that they are doing so that they can get better search results like this is it’s frustrating to everybody to think that like we’re putting so much of our blood, sweat and tears into, well, what is AI going to do to search? And how do I get better search and search is my primary channel, and I only have budget for search. And everybody’s doing it wrong.

Christopher Penn 22:36
And so we have, and by the way, with Gemini, the Gemini reduce, it renders in Markdown format, which is super handy, and super nice. And so what we’re going to do, what I’m going to do from today’s show, is I’m going to put this markdown format, I’m going to render it as a PDF. And we’re going to put up for free in the analytics for market a Slack group. So if you are a member of that group, totally free to join, go to trust insights. As such, I look for marketers, and after today’s show that PDF is going to be in the analyst community at No, no other registration, just got to sign up for the community. And you will get a copy of Geminis analysis of the Google leak. So our our results, our technical SEO fast load times. And we knew that that was that’s useful. Use structured data markup. Yep. And it specifies which modules optimize title tags. And that’s that hasn’t seen because title tags is one of those things that we have thought for years, Google has tried to essentially eliminate things that we can do on our site to influence rankings. And it turns out that actually matters. So title tags, have clear, relevant well formatted title tags, use unique content, the presence of shingle based analysis and shingle info productive suggests a Google identify some potentially penalizes duplicate content. Google has been saying for years go to go to a webmaster forum and stuff like that go to the their channels say oh, there’s no such thing as a duplicate content penalty. That’s a duplicate content penalty.

Katie Robbert 24:03
Especially with generative AI now people using it to write content. They’re like, Oh, it doesn’t matter. It matters.

Christopher Penn 24:09
It matters, it says. So in the documentation, there’s there’s a duplicate content penalty, provide accurate business information. So this is one of the interesting things in here. Your Google business pages. There’s a whole separate section of the content warehouse that specifically deals with local with Google business that seems to have an outsize weight on everything else. And what we see this in Google’s AI answers, right. The demos that gave Google I O A little while ago said hey, here’s all these things that you know that AI knows about your business. It turns out content warehouse, really prioritizes your Google Business Listing. So if you don’t have one, because you’re like, Oh, well, we’re a virtual business. We don’t have offices, you need to set one up. And then in the case of trust insights, it’s probably time you know, later on today for us to go and look at our listings to make sure that they’re up to date, but and clearly that matters. build high quality backlinks, authoritative sources prioritize backup with high anchors, source type sources, relevant context, avoid spammy backlinks. Okay, that’s not surprising. Content SEO, create relevant fresh and engaged content. There’s the freshness we were talking about earlier, Katie freshness is in there. Focus on relevant keywords and topics, quality salient terms that salient terms and content and your content readability. Okay, I have a question. Talk amongst yourself.

Katie Robbert 25:35
Well, and so, you know, John, this is something that, you know, you look into, so this is saying, you know, high quality backlinks, maybe I’ve been misunderstanding, but I thought that, basically, Domain Authority didn’t matter anymore. So it sounds like it does, like, how does that change? The way that we’re approaching things? If, like, I guess the first question is like, okay, so if Domain Authority matters again, how do we get good domain authority? And then how do we make sure we’re getting links from it?

John Wall 26:08
Yeah, that’s, that’s another one of the things that came out of this was, yeah, there’s absolutely domain authority, even one of the other ones that was crazy. There was some stuff about author reputation. Like if you’re a certain person writing on different sites, the the you can get domain reputation by getting the right person to write for your site, which, ya know, you hit that one, that’s another one that we’ve, you know, we’ve been told for years that that’s not true, but it definitely is the case. And so now, the one thing where it does go back to the standard parrot headline of the EA T, right, just having expertise authority, that will eventually get you traffic one way or another. So if you’ve been doing that, then you you kind of are working on your domain authority. But yeah, I didn’t know there was some stuff about even subdomains right to it and domain names actually mattering? Which is another thing that they’ve said forever, it doesn’t matter. So yeah, there’s there’s a lot of a lot of shifting going on.

Christopher Penn 27:04
Yeah, there is, in fact, an actual site authority metric within the content warehouse and the compressed features, file format. Interestingly, this goes and says, while it does not explicitly address press releases, because my here’s my question, Google set for years, a press release, links don’t count. We’re going to devalue press releases, they don’t count. Don’t Don’t do press releases. It’s not a good way to build links. And yet, we know when you publish a press release on a reputable wire service. It shows up on like Wk, RP, and Cincinnati and and all these local news sites and a share and it shares is released. So like, Does this matter? Is that advice true? It turns out, there’s enough signals and stuff in this documentation, that would suggest actually, you might want to send out press releases, because the links that you get from local news sites, because this is the emphasis, there’s a lot of emphasis on local news and local links, or just localized link. Reputation. Yeah, the news wire itself probably has been discredited. But Wk RP in Cincinnati isn’t. And so I would put that, especially with our what we’re teaching people now in our generative AI course. Yeah, you probably want to be doing press releases. Again, they’re probably are relevant. And they may have been relevant this whole time. And we were just being lied to. So there’s one other thing in here, let’s ask him, What is in the API documentation about the overall topic of a site and the content on a site in terms of vectors, embeddings, and vector radius. So one of the things that was mentioned was that Google seems to be able to, to try to pigeonhole a site, like this is what this site is about. So if you are writing content, that is off topic from what it thinks your site is about, it will you will rank lower on it. And so if when you dig into the documentation, if there are, there is evidence that Google is essentially doing vectorization of the content on your site, and then turning that into a there’s a module called Site radius that says, here’s what it might say, hey, trust insights, has been blogging about Google Analytics for three years. So we think this is a site about analytics. So when you put up a post about organizational change, and an organization behavior, it’s not semantically close to Google Analytics. So your content about organizational behavior is going to do less well. So it’s sort of saying I want to pigeonhole you into your about this. If you try and broaden your interests, we’re going to penalize you for doing that because you’re not allowed I was what we think you should be. And so here we are with a new and improved SEO guide, that is based on the documentation. And again, this is a very large document. So using generative AI is the way to go. And using it in a way where it is just some of the six use cases we talk about with generative AI, it’s all about this is summarization. And this is extraction, extract the relevant information from here. And then, you know, summarize it into into a useful working guide. So that’s, that’s what really what’s in this this massive model. And so now, from a marketing planning perspective, you now have an SEO plan.

Katie Robbert 30:50
So this this particular channel, well, and I think that that’s really the key takeaway. So yes, we could, you know, bash Google for another few hours around how they’ve, you know, done us wrong, as marketers, like, that’s just not cool. But they’re Google’s, who they’re gonna do whatever they want to do. Anyway, the real point here is that, you know, to summarize, Chris, you’re using generative AI, to take really large technical documentation about how Google search engine works, how it’s actually factoring your content to rank it. And you’re saying, Give me a checklist of what I as a marketer need to do. And so now, savvy marketers who not only sign up, for our analytics for marketers community to continue the conversation, but go there and get the resource that we’re going to provide based on this analysis, they can now start to get that competitive edge over other marketers who aren’t paying attention to what this leak means further SEO.

Christopher Penn 31:50
Exactly. And I want to emphasize this, you don’t need to wait for leaked documents, as much as fun as this is, you don’t need to wait for leaked documents to be able to do this. Because it turns out that most technical companies in most places, publish a lot of information. And if we are, if we are clever, we will figure out ways to make use of that. So let me switch gears here. I’m going to pull up a new instance of Gemini and let’s call this LinkedIn algorithm. We know that there are a lot of people on LinkedIn, who were talking about, hey, here’s the secret to getting your stuff noticed on LinkedIn. And we’re pretty sure that those folks probably don’t have the secret, right?

Katie Robbert 32:35
If they’re telling everyone, it’s no longer a secret.

Christopher Penn 32:38
Well, yes, the first one we’ll fight club is gonna talk about fight club. I’m gonna switch off all the safeties and things as we usually do for these sorts of exercises. Here’s what we do know. LinkedIn publishes a really good, very technical blog, the engineering blog. And in this engineering blog, which is now like 15 years old, they have posts on pretty much everything that they are doing from a technical perspective, because the goal of this blog is to attract engineering talent. So there are things like flight interactive building public groups, and how to prevent abuse of the system recommendations, how people you may know, works, etc. What if, what if you were to glue together all of these blog posts and like a detective, say, I mean, the little nuggets scattered throughout all these posts about how LinkedIn works? And what if we could assemble a a set of protocols and practices from what LinkedIn has said publicly? So we’re not making stuff up? We’re saying, we know the secrets are in here, we just don’t have time to read all 83 page, blog posts?

Katie Robbert 33:55
Well, and I’m guessing that that’s maybe their intention is they’re like, No, we totally put it out there for everybody. But they’re not assuming that, you know, the non engineering and non technical marketers are spending their time reading that information. But then they can claim like, No, we totally, it’s there. I don’t know what you’re complaining about. We just didn’t put it together in such a way that you could understand it. And we gave you bits and pieces all over the place, but like it’s there.

Christopher Penn 34:21
Exactly. It is there. And so, what we’re going to do next is we’re going to take a assortment of 77 of the blog posts that I thought were most relevant how LinkedIn makes recommendations. And we’re going to feed this in to Gemini to say, Okay, let’s see what’s in here. So, great. Here are 77 blog posts from Lincoln’s engineering team about the way linked in technical architecture. Works. Synthesize and create a broad outline of content from these blogs specific to LinkedIn news feed, and how LinkedIn makes decisions about what content to show. Be sure to reference specific technologies, algorithms and platforms, supporting evidence, or your outline from the blog content. So we’ve got our prompt, let’s upload from Drive. We’re gonna go into and drag in those 77 blog posts. Now, I will say the part we’re not going to show here because it’s boring, is I wrote a piece of Python code to extract blog content so you don’t have to copy and paste manual accents to pain.

Katie Robbert 35:57
So how know that is something you can use generative AI to assist with. Standard AI is really good at writing code. But again, not part of today’s episode,

Christopher Penn 36:06
right. So let’s see what we got here. LinkedIn is newsfeed a deep dive into content decisions overall architecture to pass ranking system. First pass record data the candidate updates from different sources. Second pass rancor combines and ranked FDR puts into a single personalized feed the follow feed, timeline based indexing rocks DB storage content ranking and selective multi objective optimization utility function Bayesian optimization, Candidate Selection deep learning for feed ranking factors influencing the well Time Skip prediction embeddings personalized embeddings. So that’s interesting. fairness and equity, transparency and explainability supporting technologies. So we’ve now got a really good technical summary of what LinkedIn does behind the scenes. So as we did with SEO, we say, Great, this is a good technical summary. Next, based on the blog content, and your outline, citing specific technologies, algorithms, and platforms, were appropriate. Put together a guide for social media marketers, of best practices from the engineering content to increase the likelihood that their content will be seen and featured.

Christopher Penn 37:39
In the news, beat, okay. Let’s see what happens. So now we’re going to take the technical stuff and boil it down to useful practices. This may take some time?

Katie Robbert 38:00
Well, I think you know, so a lot of the software that we use has that engineering documentation. I know Mehta has their engineering documentation, which covers Facebook, and WhatsApp and Instagram and a lot of the other platforms, even our marketing automation systems, our CRM systems, they have technical documentation that most of us just skip, we don’t bother reading it because it’s so long, because it’s so technical. And what I really like about this approach is that now, we can take the documentation that tells you exactly how things work, and put together that punch list of okay, I’m a non technical marketer with this technical documentation. Just tell me what I need to know. Tell me what I need to do, so that I can then confidently say I’m now doing it correctly.

Christopher Penn 38:55
Exactly. So the recommendations for social media marketers understand the two paths ranking system. You want to capture attention quickly and signal relevance. Leverage, follow feed, build your network and post consistently. staying fresh follow feed has a freshness component, drive engagement, prioritize content that sparks conversation, ask questions and currently encourage comments and shares. Consider dwell time, viral actions, likes, comments and shares are highly valued by the ranking algorithm integrate content residents prompts those actions be timely, respond to comments and engage in discussions prompt feedback to create as a ranking signal. So not just responding but responding quickly. To things harness the power of embeddings includes skills relevant to target audience your hashtags such as Islam, you’ve talked about getting past Katy. Use relevant hashtags to increase discoverability. Avoid spammy behavior, and pay attention to your analytics. Consider your target audiences activity patterns that time of day matters. So if you’re using social media management software, like our friends over at Agora Pulse that they produce time of day analysis, You may want to refer to that to see if it works, optimize for mobile and engage in relevant LinkedIn groups, the redheaded stepchild of LinkedIn thought it might be time to dust off the marketing over coffee group.

John Wall 40:12
Yeah, you know, there’s 7000 people hanging out there looking for a party. And you know, if there was a link from there to get there, it would be great. But yeah, it’s amazing how that has just kind of been left in the, in the garbage bin. Exactly.

Christopher Penn 40:26
So now, we have our documentation powered SEO strategy. And now we have our documentation powered LinkedIn strategy. So we’ve got two great marketing plans, from the vent from the mouths of the vendors themselves.

Katie Robbert 40:43
Well, and I feel like this is a good opportunity to make a plug for the five P’s. Because just because the documentation exists doesn’t mean that the software is necessarily relevant to you. So I can see this becoming almost like an overwhelming, overwhelming obsession for marketers to try to figure out every single system, how to game everything, based on the information, but you really still have to, you know, focus, because now we have what, like 10, very specific steps we have to follow every time we build on LinkedIn. So the question is, where does LinkedIn fit into our overall, you know, business plan? So what is the purpose? What is the question we’re trying to answer? Is LinkedIn part of that? Do we have the team to follow all of these steps, you know, to the tee, so that we can say, we’re doing the thing? You know, what is the process, platform being LinkedIn, but you also have to create content. So it’s not just I’m posting, you know, random gibberish and nonsense, the content still matters. And then you can measure their performance. And so I want to bring that up. Because, you know, we’re talking about Google search and SEO, we’re talking about LinkedIn. There’s other technical documentation out there. So the question you have to ask is, Does it matter? Is that a platform I need to care about? Therefore, I’m going to go through this exercise to figure out how exactly my results are the best shown how I reach people, and therefore, because you’re adding more stuff onto your plate, you’re not taking away you’re adding. So it’s not enough to just post something on LinkedIn and say, Oh, well, it did or didn’t perform. Now you are armed with the knowledge of how it works. So you have to follow all of those steps, you’ve just given yourself more things to do.

Christopher Penn 42:37
You have however, the great thing about these tools is that if you have a template for your own internal standard operating procedures, my next step would be to do this analysis, say, okay, great. Now I want you to take, you know, second pass fine to ranking from from blank to documentation, here’s our standard operating procedures template, write out a standard operating procedure for our marketing coordinator on LinkedIn to follow that will the way the activities are optimized for that specific part of the algorithm. So again, we’ve got this great source of knowledge. And we’ve got these amazing language models that can translate one thing to another, we can have them translate into our standard operating procedure. So here’s exactly how to do this. And then it’s like printing your own cookbook, like, here’s the recipe, here’s the recipe for you for your LinkedIn every day. Here’s the recipe for your personal brand. Here’s your recipe for your LinkedIn group. Here’s your recipe for your you’re optimizing your content on your website. That’s what’s so powerful about these tools is you can use they’re not just just analysis tools, they are creator tools to help you make those standard operating procedures.

Katie Robbert 43:47
The first thing I would probably do is take my existing standard operating procedure and say, okay, great, you’ve just given me this list of things to do, update my sop with this new information, and then hand that out for everybody to follow or if I if for some crazy reason I don’t have an SOP, I could then say great, you’ve just given me all this information helped me put together a standard operating procedure that I can hand to my marketing coordinator and have them follow like a checklist that I can hand to John Wall our chief statistician and he’s gonna get marketing over coffee like up in the stratosphere now.

John Wall 44:27
Exactly. Love us again.

Christopher Penn 44:31
And to your point key going back to the five P’s if you have performance, if you have marketing goals you need to reach your next step would be to take your existing marketing planning documents, your marketing strategies, your marketing tactics, and bounce them off of these summaries say okay, we know organic searches 48% of our traffic or whatever. We know that we have to generate X number of leads for organic search this year. Go back into that Gemini. As I say, here is our marketing plan, adjust this marketing plan with the knowledge from Google’s content warehouse so that our, our tactics and our execution plans and operating procedures are, are aligned with what’s in the documentation, but also aligned with the business outcomes we care about.

Katie Robbert 45:22
Great, John, final thoughts.

John Wall 45:26
I’m still, you know, heartbroken that Google’s been lying to me for all these years. So I’m gonna need time to adjust and take this in.

Katie Robbert 45:33
You and me both.

Christopher Penn 45:36
The good news is that Molly’s the truth out there. Truth is in here now in gendered AI and in our Slack group. So we’ll we will share that information there. That’s going to do it for this week. Folks, we will talk to you next time. 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 but trust insights podcast at trust insights.ai/t I podcast and a weekly email newsletter at trust insights.ai/newsletter Got questions about what you saw on today’s episode. Join our free analytics for markers slack group at trust insights.ai/analytics for marketers, see you next time.

Transcribed by https://otter.ai