So What header image

So What? How to Use AI for Social Media Marketing

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? AI and marketing! With so many ways to market your business, it’s easy to wonder which way works best for you.

Catch the replay here:

So What? How to Use AI for Social Media Marketing


In this episode you’ll learn: 

  • Modern social media marketing practices
  • Generative AI’s role in social media marketing
  • What NOT to do with AI in social media marketing (hint: comment autoreplies are doing it wrong)

Upcoming Episodes:

  • TBD


Have a question or topic you’d like to see us cover? Reach out here:

The following transcript is AI-generated and may not be entirely accurate:

Christopher Penn 0:35
All right, everyone, welcome. This is so what the marketing analytics and insights live show this week. Katie is out in the woods. doing who knows what, but having a good time. So John, how are you?

John Wall 0:48
Good here? Yeah, I will have to take up the stick of contrarian point and push back on whatever we’re throwing out there today.

Christopher Penn 0:56
Today we’re talking social media marketing and how to use AI particularly generative AI for social media marketing. So we said we’re gonna talk about modern social media marketing practices, general AI’s role and social media marketing. And critically what not to do with generative AI and social media marketing hint, if you’re auto commenting on everything, you’re doing it wrong.

John Wall 1:17
That’s like the number one complaint like the LinkedIn. Hey, that’s a great post. Tell me more about this. And there’s like eight of them in a row. You’re like, okay, yeah.

Christopher Penn 1:24
So let’s start off with this. Modern social media marketing, Mark, social media marketing in 2020. Fortune, what’s on your radar?

John Wall 1:34
Yeah, that’s a great question. Because, you know, we are in kind of this post social media phase of its immediate channel, right, which we’re no longer doing that thing of like, Oh, if I can get 50 million followers, and 2000 likes a post, I’m winning like that. That’s it. And it’s been bizarre to see even with threads. I’ve seen even people with 1000s of followers saying, Man, this is really weird how it’s, you know, posts, no legs, no legs, no legs. 65,000 likes no legs, no legs, no legs. So the big thing is this transition to you’ve got to nail the topic. But the good news now is if you find something that resonates, it immediately spreads beyond your own network. So you’re not kind of hampered by follower count, like you used to be. Yeah, phony comments. That’s right on the list, I think the other fight. The other thing that’s huge for me has been the stuff that you’ve been doing with generative AI to create topics and lists and just do a better job of editing and get better quality content. But yeah, I, you know, to be honest, I have definitely been the curmudgeonly old man on my porch saying, Go away with your gen AI. So I’m really excited today to hear what you’ve been playing with and digging in, because I know you’ve been messing around with a lot of stuff.

Christopher Penn 2:40
So let’s start off with social media practices. I think that’s important. You hit on a really important point. Tiktok was the first social network where the content mattered more than the Creator. Right up until that point, you’re exactly right, who has the biggest number of followers, and Tiktok was the first network that’s like, okay, we’re just going to show stuff that people want, whether or not that person is one follower, or a million. And now we’re starting to see other networks really embracing that approach. Instagram, Adam mosseri, the the showrunner over at Instagram, put out a post about a week ago saying that Instagram is changing his algorithm to no longer as heavily weighed the number of engagements and followers in order for content to spread, they’re now going to look at ad trying to boost smaller creators. Now, there is obviously some self interest there as well, because if you know, these, these large accounts, they are, they’re actually having to try to pay them to create more stuff. So if they can boost up additional smaller creators, that’s less money they have to pay. But that concept of of topic based media is really important. The second one is private versus public. So everyone knows the public social media networks, Facebook, Instagram, tik, Tok, LinkedIn, etc. There is also a gigantic ecosystem of private social media. So we’re talking things like Slack. If you actually if you wanted to see a live example, go to trust For marketers, that’s our slack. There’s about 3000 people in there that are asking and answering questions of each other every single day. There is Discord, which has half a billion monthly users. This is a gigantic network that you can’t see. Like it’s not open to the public. It’s not going to show up in any search results ever. It’s just out of way, but yet, there’s a huge number of people in it. And then there’s all these other apps that have their own private ecosystems. And so part of modern social media marketing is understanding that you have the public facing things like Twitch and YouTube, and then you have private networks, even paywalls private networks where you can’t see Patreon for example, people who you’re a paying member So with Patreon, you get access to other fans, communities and groups that are not available to the public. My, one of my favorite musicians has a huge in online community and things in her Patreon. There’s only fans, you know, they have all the private communities there. That’s it’s paywall, like you can’t see participate. So that’s a big part of social media strategy today do is to figure out where’s our audience? And what networks? Are we even allowed there?

John Wall 5:31
Yeah, that’s the whole private community thing is like Discord is kind of where I still get that feeling of the old internet, you know, where you get a small group of people really niche topics and a chronological timeline, which is just, you know, you think we wouldn’t be having to beg for that these days. But here we are.

Christopher Penn 5:48
Exactly. So when it comes to using generative AI and social media? Well, okay. There’s the obvious right, let’s Hey, write me 100 tweets, I refuse to call them whatever they’re called. Now on the network, formerly known as Twitter, I will continue to dead name it as Twitter. It, you know, write me 100 tweets write me a LinkedIn post for this. This? I mean, that’s, that’s kind of the obvious stuff. But people don’t necessarily do that real well.

John Wall 6:21
Yeah, well, it’s I mean, you’re literally photocopying stuff that’s already out there on the web, right. I mean, you’ve got nothing unique or original on that kind of takes. So yeah, if, if you’re a slave in the corner, who has to crank out 500 things this week, then I can see the appeal of that. But yeah, if you’re trying to do something new or different, or breakthrough, that seems to be just a complete waste of time.

Christopher Penn 6:40
Exactly. However, you can use gendered AI to help you, if you if you got to do that. You’ve been told to do that. And you’ve just heard someone say, like, hey, go go make the donuts, right. And then there are ways to do that, that are better than others. Here’s a relatively straightforward one. In fact, I’m gonna go ahead and share my screen here, what you want to do is not just use the generative AI tool to make things generically right, that doesn’t help anyone, what you want to do is you want to use the generative AI tool to make stuff for specific people for specific audiences. So let’s do this I’m gonna go to, I’m gonna go to LinkedIn. And I’m going to pull out I’m going to here to John Wall, I’m gonna say, give me John’s LinkedIn profile as PDF. And now I’m going to say, I’m going to go over to Gemini ins and say, Let’s build let’s build, build a linked in audience profile for social media content creation. Before we begin, what do you know about best practices for creating great content? On LinkedIn, so what we’re doing here is we’re following the Trust Insights pair framework, which is prime, augment, refresh and evaluate is, looks like Gemini may or may not might be on vacation here. Let’s reload that. And we’re asking the model to preload with its knowledge. I’m going to turn all the safeties off too, because I’d like to live dangerously. And Gemini is gonna pull out its basic knowledge of what you should do on on on LinkedIn in particular. And after it, does that say, Great, let’s calibrate on our audience. Now, here is the link in profile for our ideal customer. And let’s go ahead and bring in John’s LinkedIn profile. And we’ll just go drop the PDF right in. Hit Run, it’s going to analyze your LinkedIn profile and kind of get an understanding of who you are. So your demographic and background management level tech savvy, your location, interests and challenges. So this is all seems like it’s pretty right on right. This is this is not unsurprising. What you would then do is take some of the content ideas you have and say, run this by my customer profile. Does my customer profile like this? Let me pull up here, a social post. Let’s see. Oh, actually, oh, Paul, Pope a promotion for our webinars. I’d say, Great. Here’s a actually yeah, I’ll use LinkedIn social posts. I have a social post drafted I’ve been meaning to publish one of these days. Here’s a LinkedIn post. What do you think our ideal customer would think of it What should we do differently? If anything, I’ll put in three quote marks to demarcate the area, I’ll put in the actual post. And now, instead of just make stuff on with gendered AI, what I’m happy to do is do an analysis of the content and say, Will this resonate with our ideal customer audience? Because you don’t have to pay shouldn’t make content for everyone, you should make content for. Ideally, the people you want to take action. So it just has gone through it said, Hey, your post is not relevant to your target site. So what? Exactly there’s no connection to data driven decision making there’s, you need to do some more ethical considerations offer some practical takeaways. So it has gone through and said, Yeah, your posts not so great.

John Wall 10:55
So this will push you back to the it’s like busting out the red pencil and getting you redirected.

Christopher Penn 11:01
Exactly. But now, instead of just saying you’re in wallowing and sorrow, yeah,

John Wall 11:06
it’s got some alternatives for you, too. Exactly.

Christopher Penn 11:09
Great. What I’d like you have to do is re write my post incorporating as many of your suggestions as possible, about the same length, matching my original writing style, voice and tone, let’s see what it comes up with. So again, we are taking exists something existing that maybe was not great, and retuning it using our ideal customer using our audience for this. Now, who should you do this with? If you’re on Twitch, or LinkedIn, or Instagram, or whoever, look at your own data, and say, who are the top five people that engage with us, or looking at your CRM, who are the five customers that we want to see and five, we’re looking at marketing automation system, who are the five prospects that we want to land as customers this year, build a profile based on them, and then start pushing your social content through it to say, reread the rebuild this rebuild this for our customer. So we have here who decides what’s true in your data, we make this a bit bigger here, we have this, this is essentially done a rewrite for us. So now we’ve we’ve gone from your basic generation, write a LinkedIn post right to rewrite this LinkedIn post for our customer, or our target audience. So that’s an example for modern social media marketing that is actually useful, you can say I want to recalibrate my promotions, make help me make a video script for YouTube helped me get a show outline for a live stream, all these things you would now have with a with an ideal customer profile. So that would be number, the number one most important thing you can do is using generative AI, build your customer profile based on real people, and then start running stuff through it.

John Wall 13:18
Very good, that’s taking the pain away from your day, that’s for sure.

Christopher Penn 13:21
Exactly. Next thing you should be doing with these tools, you should be using them to analyze your data using them to analyze in particular social media data that you want to understand better. So I’m going to move over to ChatGPT. Because of the different major tools, this one has the best image recognition. Suppose you are looking at popular content, maybe its competitors content, maybe it’s some of your audience content. Maybe you’ve got a bunch of it. Maybe you’re looking at hiring an Instagram influencer, and you want to inspect your last 200 posts, these tools can do image recognition. So let’s say analyze the following image and document what you see in a bullet point list. Now I’m gonna take an image that I pulled right off of Instagram earlier today. And let’s see what it does. So it’s got basically some groceries on it, as basically we should provide here’s listed the items items visible. Now this, you know, it’s kind of fun. You know, this is this is a very basic example. But it’s gone through and done. Image Analysis, image recognition. One image is kind of interesting. But if you were to start doing this at scale, with images on a hashtag, or with images on on deviant art or any of the image based social networks, you could start putting together a topical analysis of things that people share, especially You also happen to have accompanying post data, what post images seem to resonate best with an audience, if you were to pull a competitors, Instagram account, and their metrics, there are competitive tools that can do that. You could pull a list of all their content for the last year, do image recognition on via the this case of the ChatGPT API, and come up with a list, here’s the, here’s the topics that from an image perspective seem to work best for our competitor, maybe we can get some inspiration, when you have those topics, to your ideal customer profile. And you say, give me some Instagram ideas, to add some things to shoot that would mirror the best performing posts of our competitor.

John Wall 15:49
Yeah, that’s amazing, I can see that. And I hadn’t thought of this before. But if you have a library of images already on your blog, or wherever that need to be tagged, being able to auto tag, stacks of images off, that’s another pain point that you can take hours out of somebody’s day, there is

Christopher Penn 16:06
a script that I have that I’m still trying to get get the last bit of working, but it’s starting to get there. On your computer, there’s a folder somewhere that is filled with things that say screenshot this date and time, right, hundreds of these. So I’m trying to figure out how to pass each one to the GPT-4 API, have it recognize what the image is, and then haven’t renamed the file, like image of of two things this that there’s, there’s paid companies that do that, I want to write my own because it shouldn’t be that hard. Ya

John Wall 16:42
know, that every marketer on Earth, it’s just like, Oh, my God, that pile on my desktop, that would be wonderful.

Christopher Penn 16:49
Exactly. So that is another example that’s in this case, this is the use case of extraction using gender of AI to extract data from one format to another, you can do all sorts of crazy stuff with this, you can take content from one format and move it to another with it. Another useful, useful use case, suppose you wanted to analyze your stuff on threads. How would you go about that today? John, how would you go about analyzing the performance of your content on threads?

John Wall 17:21
Yeah, that I’ve got nothing on that, right, because there are at least now is a web client. So you could, you know, scrape some stuff off your browser. But for me, it’s literally just scraping through the phone and seeing which ones have hits. Is there even a toolkit is that in the Instagram toolkit, so

Christopher Penn 17:40
it’s not an Instagram, per se, although you can get to it through Instagram. But there is a portal that meta offers in your settings and privacy, that allows you to export all of your data. So you can say I want my data, give me my data. And it is in your meta account center. So if you go to, in this case, go to Account And then you will be able to see your information and permissions, download your information. And then you can basically make a request of what information you want. So you could say, for example, here on my service, there’s my Instagram account. Because threads is part of Instagram, I want some of my information. And you will you’ll go down and say where it is here threads, I want my threads data, and you want it downloaded to a device. So you queue up this all this information. This is all of the data from threads. What’s in the box, what do you get, you get a very, very unhelpful, extremely large pile of data, I pull up what they send you. This is a big folder, they send you big folder that looks like this.

John Wall 19:03
And in JSON, that’s good.

Christopher Penn 19:05
It’s Yes, it’s threads. So it’s all JSON files. But it shows you your threads and the replies the things you’ve liked, on threads, etc, etc. Now, if you’re presented with a JSON file, what do you do with it?

John Wall 19:26
Just open up your IDE and pray you can see what you need.

Christopher Penn 19:32
You could do that. Yeah. So here’s an example. This is this is my most recent threat. Today, I got a working model. And you can see there’s all sorts of data about this. However, this is not super scalable. This is not super helpful. So what you could do is you go to your generative AI system, let’s say let’s create a new prompt. Okay. Today, I need help writing some code in Python. or the lack of material choice to process a JSON file and turn it into a spreadsheet? What do you know about this process? Generally, what are the best practices. So again, we’ve got to preload the model with knowledge about converting JSON into something else. And then I would provide it with a JSON file, and say, Here’s the file, write the Python code to interpret this file, turn it into a Python script. And then for all the stuff that’s in that folder, I cannot convert it all into spreadsheets. So I would have working software then. And then anytime I need to run analytics, on my threads data, I queue up the export, private end of month, if you’re doing it for your, for your company, and the monthly export your threads data, run your Python script, because once it’s made, it’s made, and now you’ve got a spreadsheet. And if you want, you have the tool to do the analysis, especially but you can convert this data, you can extract and rewrite the data into a different format. And now instead of having to go scroll through threads, now you’ve got it in an easy to use form.

John Wall 21:15
That’s really cool. Now, are there any large language models that can handle JSON and do any work with that? Or is that beyond that? Let’s

Christopher Penn 21:22
find out. Here’s an example file to use in this project. Help me write the code to process this file, the Hit file, file upload, and let’s get my let’s get my threads and replies because that’s kind of an important file, we’ll see it’s a 1.6 megabyte file because I am up, it says error extracting text from this file. Why? Because it’s too darn big. Okay, I should let’s try. Alright, let’s try a smaller one of these files, File Upload threads, posts, let’s just do posts. See what this says here? That’s a much smaller file. Now, it does not like but I bet just the JSON file format,

John Wall 22:14
it doesn’t realize format, probably.

Christopher Penn 22:17
let’s turn to a text file, I will just add a dot txt at the end of the file. And we’ll put this in and see if it likes it. So yes, just adding a dot txt fixes the problem, there’s 515,000 tokens or 400,000 words in this file. So this might take some time. But But to your point, maybe he couldn’t read it directly. But for sure, it will help us be able to write the code to process this data and get an understanding of what’s working.

John Wall 22:53
Alright, yeah, that’s just amazing just to be able to grind that through and, and I did not know that is crazy how far down that’s buried in the meta tags there to get to that data. But if you know where it is, you’ve got the keys.

Christopher Penn 23:04
And this is true for every social media platform. GDPR requires them to give you your data like it is required by law in all the EU, it’s required by law in California under CPRA, and seven other states. And so as a result, they are required to give you your data. So, for example, one platform that is extremely difficult to get analytics on LinkedIn, right? It’s extremely difficult to get data out of tool unless you request the data from LinkedIn. And you say, Hey, give me my data through the export facility. And when you do that, LinkedIn will spit back Hey, here is here’s all the content that you have posted on LinkedIn. So LinkedIn gives it to you as a big ol spreadsheet, you can of course, load this into like Excel if you wanted to. Or if you wanted to do something more interesting, you could open that up in Excel. Let’s go ahead and start up a new session here. Oops. So there’s my LinkedIn data. That uh, let’s, let’s go into this. And we can see here is my engagement rate, my impressions, my reach, etc. I’m going to do a sort on the reach column. And now, let’s take this LinkedIn data, say today. Let’s do some social media content analysis to identify what is high performing content on Linked In as measured by one of our columns that are available here. Impressions, reach and clicks. Let’s go with impressions. Even though it’s generally not a great metric, what do you know about this topic generally. So we’ll preload the model with some of the some of its basic language so that it knows how to interpret our data. Turn off the safeties. And now what I’m going to do is I’m going to create a new text document. And this is going to this will be my high performing posts. Now, I’m going to go to the bottom of my LinkedIn down here, and grab a handful of my least performing posts. Okay, now, let’s say, Great, I’m going to provide you two files, high performing links in posts, and low performing. LinkedIn posts, what I’d like you to do is analyze the contents of the high performing posts and contrast them with the low performing posts, and provide your analysis. So now we go and do our file upload. I’m going to rename bottom performing to low performing so that the AI does not get super confused. The plop both of these in here. And let’s see what it says. So here, we’re using generative AI to do analysis of unstructured data. So this is data where we’ve done the binning, we’ve sorted it out. And now we want to say, what, what’s the difference? Why, why is this not working? low performing posts, repetitive format, it’s just, you know, lack of engagement, limited content, variety, weak calls to action, high performing posts, variety of content sites focused on engagement, trendy topics are the practical approach, strong calls to action, promotional content with value, these are recommendations. Right. So this is this is gone through now and did this assessment of our content. This is from the Trust Insights, LinkedIn account. So let’s say great, let’s turn this into a checklist for our content, write out a checklist, like a pre flight checklist that a pilot uses to assess any new content before we post it on, linked in using the criteria from your analysis of what makes a post high performing versus low performing. So now, we’re going to take this summarize knowledge, and we’re going to rewrite it, we’re going to turn it into something else. And now we’ve got a preflight, we’ve got a standard operating procedure that we can use that could hand off to our example, our account manager, Kelsey and say, hey, when you’re scheduling social media posts, here’s the checklist, make sure you’ve checked off each of these things. And Kelsey can go yep, I’ve done this, or she can say, hey, this, this content is is not appropriate for our target audience or the it’s the tone is wrong. And that you can push back and saying, Yeah, we shouldn’t post this is a bad idea. Or if you want to get real fancy, let’s say great. Now convert this checklist into a prompt for a large language model like Google Gemini, right in second person, imperative, which is second person point of view. I’m now going to generate a prompt, you are assisting a marketing team creating high performing LinkedIn content before the team posts anything on LinkedIn, you guide them through the following checklist. I can now take this prompt, turn it into a system instruction. And now integrate it into my social media workflow. So if LC is not available, it will I can take post and put it in sake, check my work valuate this post? Should I post it or not?

John Wall 29:22
Yeah, that’s great as well. And same deal. You can have somebody run that and confirm that everything’s in there. It’s just a great way to make sure you’re hitting all the marks before it goes out the door. Exactly.

Christopher Penn 29:33
So there we again, we’re using summarization, we’re using rewriting now even to some degree recent question answering. So we’re using all these different techniques for language models to make our social media content better. So this is derived from our own data. I would strongly encourage you to do this with your data don’t I mean, yes, a lot of these things are pretty general best practices. But you may be doing something differently wrong than we are. And so you want to make sure that you’re using your data and you can see it we’ve we’ve not used any crazy sophisticated tool, we’ve been copying and pasting out of an Excel file. Right. So this is this is not rocket surgery. This is pretty low tech stuff. So we’ve covered image classification, we’ve covered generation and building an ICP for your social media content. Now we’ve covered tuning up your social media content. The last thing I think is worth talking about when it comes to using these tools. And using generative AI for them, really comes down to, in some ways, understanding social networks. So I’m gonna start a new session here. A lot of people, I do mean, a lot of people talk about hacking the algorithm, here’s the, the 28, LinkedIn algorithm hacks that will make you a better marketers. And I did, John, you’ve seen all those? What are the what are the hacks that you’ve heard?

John Wall 31:03
You know, post once a day, always answer all comments, tag everybody and their uncle, even if you don’t know who they are. Yeah, those are the that’s the low hanging fruit. Oh, and ask questions and have surveys. Let’s see.

Christopher Penn 31:19
Now, here’s my, here’s the question. Is that valid?

John Wall 31:24
Now, I’ve seen plenty of, you know, post tank using all of those, those tools. So yeah, that is all anecdote all the time.

Christopher Penn 31:34
What would you tell someone? Or how to learn what actually works?

John Wall 31:42
Yeah, examine what’s out there. Like, this is a great one for see what your competition is doing and what’s working, but try and get to the data and find out the common thread against of stuff that is successful.

Christopher Penn 31:54
I’m gonna suggest one even better than that, I’m gonna suggest that we just read the instructions. What I’m specifically referring to and this is true of every social network, every social network publishes an engineering blog. Right. Matt has got an engineering blog, LinkedIn is got an engineering blog. Their engineers are on podcasts and stuff like that. And these folks share how the how the technology works, right. So they have this one here, new approaches for detecting AI generated profile photos. So it says, here’s the the technology, we use this dataset, we have these results, this is what we did. Here’s our conclusions, etc. The downside for most marketers is that these posts while extremely informative, right, this is how LinkedIn detects bad behavior on LinkedIn. And they show you the the sequencing of the layers, the model, the problem is most marketers are not technical marketers and technicians and engineers. Understanding how it LSTM works is probably not super high on the list of things that you wanted to do this week. I mean, I may be wrong.

John Wall 33:03
But yeah, it’s not doing that this weekend.

Christopher Penn 33:07
Okay, so what do we do about it? We use generative AI. Huge surprise, right. So today, we’re going to look at the engineering behind LinkedIn, please read through and provide a high level summary of this LinkedIn, engineering blog content. And what we’re going to do, I’m going to take the engineering content from LinkedIn, all how many words is 112,000 words, over 68 blog posts that were just taken from the LinkedIn blog. And I’m going to turn the safety off. Because there’s some topics in there like abusive behavior that attract detect that will will trigger Geminis warnings. And it’s going to provide an analysis of all of the content together. This is really important, because there’s a lot of detective work that goes into understanding an algorithm, right? So it’s it, LinkedIn doesn’t just hand it to you on a silver platter saying, Hey, do these eight things and your stuff will do better. Instead, what they talk about is, here’s all the different systems and technologies that we use under the hood. If you if you gathered all that together and say, AI helped me connect the dots, help me understand what’s going on under the hood in a coherent, cohesive, big picture, then it will be able to to distill that down and say great, based on the engineering content, we’ve analyzed, build a set of practices for developing and publishing. High performing content on the link in new was feet, we define high performing as content that gets great reach and engagement, write the practices in bullet point format. So now we’re gonna say you’ve got all this knowledge, we’ve distilled down the entirety of the relevant portions of the LinkedIn blog. And by the way, I would if you were going to do this in production, I would also recommend grabbing podcast interviews, from this week in machine learning and a bunch of other podcast, very tactical podcasts where LinkedIn engineers have done an hour and a half of talking about what’s happening on LinkedIn. It says, Okay, here’s your content creation, relevance, value, diversify content types, use trending topics and hashtag encourage conversations. Right? That’s the basics. optimizing for LinkedIn algorithm understanding dwell time, right audience, people you may know, etc. So great. This is a good start. But this is all fairly obvious content. I think we’d agree this is, this is not a surprise, right?

John Wall 36:06
Yeah, right. That’s the engaging content. That’s all you have to do just right, engaging content.

Christopher Penn 36:11
Exactly. Give me a set of practices based on the technologies described in the engineering blogs. For high performing content that are not obvious. This is an example of obvious content, do not provide obvious content like this. And now I’m gonna copy that entire response that I just gave me, right, because we’re with something called contrastive prompting, it’s a more advanced prompt engineering technique. You basically say, hey, all that stuff you just gave me disqualify it, it’s out. Instead, find me find me more things, better things, newer things. So you have skill optimizations, sure your profile skills, engage in collaborative articles understand the two paths ranking system triggered downstream impacts, embrace embeddings. Consider timing and freshness. I like this, explain in greater detail. The point give me step by step instructions for how to do this. Anything that you don’t understand as a marketer, just ask the machine explains to break it down for me Give me a step by step instructions for leveraging this particular technology. And assuming that the data is there, a model will be able to do that for your model be able to do justice. Now what we’re doing, we’re using Gemini 1.5, you could use Claude three Opus for this, you could use ChatGPT, the paid version, any of the big paid models will that you can do this with you cannot do this with any of the free versions. So if you’re using the free versions, this is not going to work for you. Here’s a step by step one, grasp the basics, understand what the Rancors do optimize them capture attention early. The first pass rancor makes quick decisions about content relevance. So it’s crucial to grab attention within the first few seconds, strong visuals, compelling headlines, etc. Leverage keywords and hashtags posts when your audience is active. Second, pass rancor encourage engagement, consider dwell time, build relationships. And you might say, Okay, well, that’s great for the first pass rancor. Okay, I understand this idea of capture attention early. Give me some practical tips on how to do this. Again, you may read it and say, Well, okay, that’s great. How, how should I do this? So one of the things that I’ve done this experiment a few times, the one of the things that we’ll recommend is cross channel, that when you if you’ve got a LinkedIn post that you want to do really well email your list about it, send out an email, say, hey, go engage on this post. So doing that cross channel, promotion, put up a YouTube post, if your audiences on YouTube put up a tweet about it, whatever the case is, use that to boost the performance in a previous iteration of this and said, You You you’re trying to target that first 60 seconds if you can, first 60 minutes if you can get that first 60 minutes, you’re in good shape. If you have co workers tell them in your internal slack or Discord, whatever, hey, our new posts is up everybody go like and comment on it. So these are, you know, again, these are all the things that you can do to deconstruct social media algorithms using the engineering content step guessing. We just take what the engineers are telling us that this is how it works and we follow instructions

John Wall 40:00
Yeah, I love that as far as taking advantage of the first pass, like I did not know it was a two pass system the way that the does that that’s pretty interesting. If you miss that first window, you’re completely out of luck.

Christopher Penn 40:10
Yep, exactly. So these are some of the practical examples for how you use generative and marketing is way more than just, Hey, make me a LinkedIn post make me make me a pile of tweets and stuff. We’re talking about processing your data, reviewing and digesting technical data, doing data analysis, building an ideal customer profile, or running your social media content by it either manually or programmatically. But there’s all these different things you can use generative AI for. So the last couple minutes, the things you shouldn’t do with generative AI. Clearly the bot responses to everything, they all have the same little rocket ship emoji, and that’s a great, some great insights, John.

John Wall 40:55
There everywhere

Christopher Penn 40:57
Everywhere as every platform there everyone is, is sinking. And the reason why people will say well, why can’t they stop this is because these being done by browser extensions. So the software company can’t detect the difference between a human typing that in and a Chrome browser extension, just copying and pasting it in. So it’s very difficult to detect, even though we can see the broad pattern. Because you can see, you know, the last 10 People have basically the same comment. You can’t tell that any one of those people individually is using this on your post. Thanks. So there’s, there’s no good way of filtering that out. The best you can do, if you’re tired of is fine. For people doing that on your profile, just block them.

John Wall 41:39
Just kick them to the curb. That’s it

Christopher Penn 41:42
Second thing you should not do is you should not use generative AI tools to harvest data that you don’t have privileges and rights to, you can do a whole bunch of things that are definitely a different color hat than that we try to wear around here. The tools can do it. But as Ian Malcolm said to Jurassic Park, your scientists were so preoccupied with whether they could they never stopped to think about whether they should.

John Wall 42:08
Yeah don’t find yourself locked on Jurassic Island. orange jumpsuit is not the way to go.

Christopher Penn 42:15
Or being eaten by a Tyrannosaurus. And third, please do not deploy content directly out of a generative AI system to social and public social network without human review, because the potential for that to go wrong. This actually dates way back to 2016. When Microsoft released Tay their their twitter bot, and Tay was turned into a racist porn bot within 24 hours because it trained on the replies it got and no one thought, Hey, we should put some guardrails on this.

John Wall 42:53
Yeah, that’s runs afoul of itself quickly when you throw it straight into the system just gets ugly quickly.

Christopher Penn 43:01
If it does, any final parting thoughts, John?

John Wall 43:05
Yeah, be a good citizen, you know, use this stuff to make great stuff. Don’t be a jerk. I think that’s the right message.

Christopher Penn 43:14
All right, so that’s gonna do it for this episode, we hope you found it useful, gave you some ideas for using generative AI. With your social media marketing. Again, it’s way more than just make stuff. You can even do things like copy and paste a competitors, you know, public posts and say, what is their social media strategy? Or if you have posts from a couple different networks, say identify their social media strategy? What are what is it they’re trying to do? What’s the big theme, you name it? These tools can handle enormous amounts of data. Leverage that capability, leverage that capability to to process data, and to give you insights from it. It’s the best thing you can do with them and is way more than writing a blog post. Alright, folks, we will talk to you 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 trust I podcast and a weekly email newsletter at trust Got questions about what you saw on today’s episode. Join our free analytics for marketers slack group at trust for marketers, see you next time.

Transcribed by

Leave a Reply

Your email address will not be published. Required fields are marked *

Pin It on Pinterest

Share This