So What? How to Automate Content Transformation

Happy Thursday, everyone. Welcome to So What, the Marketing Analytics and Insights Live Show from Trust Insights. I am Chris. John is here with me. Katie is off this week.

Uh she is at the Marketing AI conference, Make Con Workshop Presenters Briefing. She'll be doing a uh a half-day workshop on all things Clawed. Uh so the Claude for Business Works Up. So they are working out how to uh how that's gonna go, which is why she can't be with us today. So today's episode, we're talking about automating content transformation.

We talked about this in the podcast earlier this week, but in the age of AI, there is no reason why any piece of content can't become any other piece of content, especially if it is human-led to begin with. So our guinea pig crash test dummy, I don't know what we want to call it. Uh, of course, is the Venerables the best and second oldest continuous marketing podcast in the world. Marketing over coffee. Yeah, I don't know.

I gotta, I'm still gonna go fact check that. I we still may be the yeah, the the oldest around, but we'll have to see what uh what the internet says. Yes, the oldest continuously operating. Um we have, I mean, John, we have over a thousand pages of podcast episodes. I want to say we're in the 900s for the actual, you know, mp3 files, but over a thousand pages of stuff that date back to 2007 when you and I were still uh recording in the uh the dunks on on Route 9 in Natick.

This is the venerable uh birthplace. And so if we were to pull up some of this content just to see what's there from you know days past, uh this is a post from 2009. And there's there the media files there. In fact, I'm guessing, let's see if we if I click on this link if it actually the MP3 still loads, but there's a bunch of stuff that doesn't work anymore. That's a huge win.

I would I was expecting the link not to work, so we're at least uh we at least providing the content. But yeah, because that's insane over the years. You know, we've had different email service providers, we've had different web hosts, we've had different templates, like yeah, this is gonna be an interesting uh walk down adventure aisle here. Uh yes, it is, and we have there's a bunch of things like there's old sponsor blocks like this episode from 2009 was brought to you by Blue Sky Factory, which hasn't existed since 2012. Um, when it was bought by what counts, which was bought by uh J2 services, then got bought by some other.

So it's been like five acquisitions, and literally everything about that platform is gone. Um, it is a distant memory of better days in Baltimore. But there's still a sponsor block. We used to have a question of the week thing um back then, have an upcoming event that is long over. In fact, Blog World New Media Expo has been gone.

MME, yes. Props to marketing props. Marketing props still going strong 20 years later. Um, we've got links theme song. We've got an old block here that does nothing, and then the a sign up for the marking over coffee newsletter, which is from Blue Sky Factory.

That that has not worked in in 15 years. Yeah, the get your coffee delivered too. I'm trying to remember what that was. That was something. I think at some point you could sign up via RSS and it would email the show to you.

Yes, with uh Feedburner. Uh-huh. Okay, so that was a feedburner thing, which has also been gone for like 10 years now. Exactly. So today's topic is about how do we automate content transformation?

How do we make new stuff out of the stuff we've already got? And in this case, also fix stuff up. But the keyword here is automate. I could, and any one of us could just hit the edit button, go in here and be presented with oh my god, there's all the cruft from from you know very, very long ago. Some things that would be nice would be if we could get a transcript from that episode on the site.

We've talked about that in past episodes. The MP3 exists. Uh this was this was show 122, but neither of us want to download 900 episodes. That that sounds terrible. Um, what could you do with that?

You could send it to a transcription service once if you had the audio, pass it over, and then you know, come bring it back and upload it back to the site, but that's still work. What we want to do ideally is not once we get a system set up, we don't want to see it again. We just want to say, you know what? You do it, and I'll I'll check back in with you when when you're done. So, how would we do this?

A couple of different ways. Um, first, every we have to think about where we have services that we could use to do some of these transformations, and some of them are free, some of them are not free. And then how do we orchestrate it? How do we make it happen? So, from just this one episode, Sean, what things would you like to be able to do?

Yeah, I think like you said, transcripts is obviously a way that would really beef up the content. I mean, I don't the one thing that's a little bit weird, is we have so I had done a project like five or six years ago where we were going through episodes, and we've found that like more than 70% of the shows are topics that are now just dead and gone and irrelevant. So I think there's a portion of episodes that probably need to actually be deleted because it's just wrong info. I don't know, well, and uh like your just your opinion on that too. I mean, that's just me talking, but maybe there's no value to doing that, maybe for historical value and for geo and SEO, it's still worth keeping.

But yeah, transcripts, and then the other one is just yeah, clean up all the garbage. Like I would like to have all of the old, any weird old email forms out and have everything be the new email form. And um, and like you said, the sponsor blocks is another weird thing. Um, like there's a couple of duct tape and chicken wire things behind the scenes where there are certain sponsors that get redirected to other places and things like that, but it's by no means any kind of system. It's like I said, it's just literally like when somebody yells about it, I go and you know, jury rig something.

So something more formal on that front would be cool too. So that means that we need would need to make use of the CMS itself to be able to do a lot of modifications, and you can do that programmatically. WordPress, which is the the site this system is built on, um, and the company that we host with WP Engine has what's called an SSH gateway. Now, this is uh nerds speak for a command line tool, like a terminal could connect via a secure connection directly to WP Engine, directly to the website itself, and manipulate WordPress so it can download a post, edit it, and upload a post. You would do that in in terminal, you know, get get the old 1980s war game style interface uh bring that up.

And that is sort of the the secret to that word automate. When we talk about automating content transformation, is do you have connectors that can connect to and from your AI system or whatever system is that you want to design. The good news is uh, you know, for example, a host like WP Engine, uh Cloudflare, many other content delivery networks and content management systems, all have either command line tools or APIs. That would be prerequisite number one, because we absolutely positively have to have some way to programmatically grab posts, edit them, and then send them back. The second thing we would need access to is some form of AI, ideally, something that has a smart language model behind the scenes, something like a clawed code or what have you.

But you also might want to have a lightweight language model that can do grunt work, like in this post text, identify the sentence that is the sponsor message and replace it with a uh WordPress short code so that you just have to maintain the sponsor block in one section in the switch in the site. And on every page, the sponsors update as people come and sponsor the show. Because then you can say, hey, sponsor, you don't just get this episode. For while you are an active sponsor, you get run of site. And oh, by the way, when you stop paying, you go away.

And if you want to pay for a perpetual listing in the in the site, there's an added package for that. So we would need a language model of some kind to tie into it. And then the third are the different services that would do those kinds of transformations. So if you just wanted a raw transcript of the words themselves, there's so many choices. Um there's there's local models like Parakeet or Whisper, there's uh cloud models, every every big tech AI company.

If you want transcription that also has what's called diarization, which means you separate the speakers by who is speaking, you're gonna have more uh specific services, uh Otter, for example, or Fireflies, or uh you know, there's again there's a gazillion of them. But what you're looking for there are again vendors that have some kind of text-based connector, because if you can connect your clawed code or your open code or what have you to that vendor via a command line tool via text environment, then AI can just steer. It can say, Hey, I'm gonna upload this MP3, you're gonna transcribe it, tell me when it's done, send me the thing back, and I will then be able to do stuff with it. If you wanted to do things like audiograms, uh again, there's a million and a half services that do audiograms now, which one of those have a command line tool? So, what I would suggest as part of anyone's Martech audit is to just go to Google in AI mode, you can you can ask uh something along the lines of which audiogram production companies have a text-based command line tool or CLI that I can use with agentic AI.

So if I tap on AI mode, it's gonna give me things like hey, huh? None of them do. Suckers. Um, however, there's an open source version called Remotion that says if you uh which by the way, this is like really bad news for headliner and descript and all those companies. Y'all should have command line interfaces for AI already.

Yeah. Um this one says there's this free open source one uh called Remotion Audiogram that integrates with your agentic workflows, React based programmatic framework, high quality customized audiogram videos entirely by the command line, driven by text commands. I didn't know this existed. There's a new one. Um let's see, Deepgram, classic open source New York public radio audiogram.

So now we have an answer. We have of this um something that we could add into it. So now, how do we start thinking about all this? Our next step would be to write up a project plan of some kind that a tool like chat GPT work, uh, open AI codex, Google Anti Gravity, Clawed Code, Clawed Cowork, any of those could could build for us. So let's start a new document.

And this is just a text editor, nothing fancy, nothing special. I'm gonna say, today we are going to build a content management system. The local programming language is Python, and you'll find all the coding standards in the docs folder. What we want to do is connect to our website, MarketingOverCoffee.com via the WordPress command line tool, WPCLI, which is already set up as an authorized user of the SSH gateway so it can grab, download, edit, and upload WordPress posts. We'll also have access to the remotion tool, which makes audiograms.

We'll have access to Fireflies, the transcription company via the Fireflies command line tool. And our goal today is to do a few different things. First, download all the episodes at the text files and find the MP3 links in them. Two, make sure the MP3 links have the correct libsin URLs for modernity. Three, download the MP3 files, store them locally.

Four, send each mp3 file to the Fireflies transcription service to be transcribed programmatically and then return the transcript in plain text format and store it locally, probably in something like a SQLite database. Five, fix the WordPress post itself to remove mentions of old sponsors, which are always denoted with a sentence, something along the lines of this episode brought to you by or this episode sponsored by, with a WordPress short code that we can programmatically change later. Six, we'll add the transcript cleaned up back to the post at the end of the post so that we don't lose the existing show notes because we want to preserve what's already there. Seventh, we'll upload the post back to WordPress and make sure that it looks okay. That's our process.

And I need to add in my Python tools. So now I have my Python rules, which I'll want to put in here. And we want to tell it how to get to Fireflies. Fireflies has already been authorized on this computer. Now we'll finally finish up this general project plan with ask me up to 20 clarifying questions using your brainstorming superpowers until you have enough information to succeed at this task.

What did I forget? The only thing, you know, there are episodes that have transcripts already on them. Um so keeping an eye for that would be one thing. Um the newsletter subscription forms. Yes, that's right.

Forgot about that. All right. Let's see. You will also see expired sections in posts leading to the old feedburner distribution sections, which is gone. That software hasn't existed in some time.

And you will also see very old Blue Sky Factory newsletter subscription forms. Those haven't existed in a long time and need to be removed and replaced. Now on the website, there should be the new newsletter, right? The which is where people can just goes to a Google form. Okay, at the moment, which goes there.

All right. This is where users should be sent. All right. What else did we forget? That's everything that's there.

It would be interesting to know which episodes don't have images. Because from the old content management system, the I basically I think all the image links are broken. Okay. Well, let's add that in. Why not?

Part of your work will be a content audit to identify missing things like missing images within the episode so that we know we'll have to go back and generate others later. We're not going to create those images now, but we do want to flag them for use replacement. Okay, what else? That's definitely uh as much of the mess as I know today. Um all right.

So the best practice here is in this kind of environment. Let me touch recipe. We'll put all of that foaming at the mouth into what's called a recipe file. And this is just a plain text file, but it contains all the information that we would need. Oh, I forgot to tell.

Feel free to use your web search tools to obtain any missing information. Once you have come up with the plan, have the advisor review it. All right, so now we can go into our coding tool of choice uh using clawed code here. Let's do a new session, and now we do the extremely boring uh read and run recipe.md. And the reason that we keep that recipe file in a text file instead of just as an ad hoc prompt is when you restart a coding agent a lot of the time or do a new task, it forgets the original ask.

Like it builds all these elaborate plans like, oh, I've got to do this, gotta do this, and you know, 25 minutes later it's doing something else. You're like, what are you doing? Like go back and read the recipe. It's like, oh, you're absolutely right. You this is a content management system and not you know a uh a chicken cooking timer.

I'm like, how did that happen? But it'll probably have a chance to go through and read um any existing project materials because it uh as part of the prep for today's episode, it was a lot of messy stuff and um getting software connected. But once that's all done, now it can now it can just ask me the questions that it wants answers to before it likely builds the software to do this task. And that is the other part of this automating content transformation. We don't want an AI tool like a clawed code or a clawed co-work to be doing the actual work because that AI usage, especially now is more expensive than ever, right?

There's no shortage of companies have burned through their entire coding budget, you know, in two months for the for the whole year. So what we want to do is instead we want to encourage our AI tools to build us the software we need that will then turn into uh a thing that we can re run and rerun all the time. That way, when a new episode goes up, if this is on a scheduled task to run once a week, it will automatically handle all that processing, and you and I won't have to do it. Yeah, that's a wonderful not only cleaning up the trash but less effort going forward. Exactly.

Um, so this may take uh some time, and by some time I mean this is likely to take probably two to four hours for it to process and think through. But that is that is sort of the essence of content transformation is giving figuring out which tools to connect to, and then figuring out what the process and the flow is going to be. And then how about so what would be just the general approach? So after it runs those three hours, it will come back with basically a package for you that you would just install and set up and have run, or what how what's the how would it normally go if it goes as planned? If it goes as planned, you could either have it be something that lives on your laptop, which for you know a hobby show uh is totally fine.

Uh that wouldn't cut it. Um, or you could build it, you could direct it to build something that runs full time on a server, right? That has like it lives on the Google Cloud. And I imagine if you're a super high volume podcast, then it's just running every single day, just keeping up, doing maintenance and things like that. Now let's see, it has questions for me.

What is what do you questions you have? The current site has 1008 to 1012 published episode. Given Firefly's transcription of cost and time of live editing everywhere press posts, how should this pipeline run? Pilot batch first. Yes.

Recipe says sponsor sentences. Do you know it with the sentence. Um delete sentence, insert new short code. Uh edit sponsor lines inside the transcript. Ooh, that's I didn't think about that.

Um, it's saying, do you want to remove the sponsors from the transcript? Yes, we do. Um after editing a post, how should it go back to WordPress? Uh save the draft for manual review. Um, because it would well, here's the thing.

Here's the thing. These posts are already live. So if you were to publish them as drafts, then you'd be delete the originals. Right. So let's do a test of five or ten posts locally to see if the system works.

And if it does, then you'll publish directly once we've worked out all the bugs. How about I don't? I've never used the manual review process. Is that something different? Or is it that just means what you were just saying is actually what you're saying.

It's not, I was thinking there was some kind of WordPress content approval or something, but there's not. There can be. Uh, there actually is a whole like approval pipeline process. And uh it's especially robust in WordPress multi-user, which we don't use. Um what's nice about this particular process, especially in Claude, is it's gonna keep asking questions.

Like told it can ask up to 20 questions. One of the things that is new with um the current fifth generation models, so Opus 5, Sonnet 5, GPT 5.6, etc., is that they're very uh agent focused and agentic. So you have to be specific and say, don't just say ask me questions. Say ask me up to 20 questions so that it knows. Oh, I can ask a lot.

It says feedburner, blue Sky Factory both appear two ways in the data of oil plate widget block and one on editorial mentions. Uh let's get the widgets and the editorial mentions. MP3 links use three different URL patterns. Normalize all to direct traffic.libsen.com, which is the correct one. 68 posts use the old audio URL uh short code to embed the player.

They're just use a plain link. The the audio shortcode player hasn't worked in years. I do not know what the deal with that is. There is one thing with the show with the MP3 links. You know, anything in the past three or four years has to be kept intact because there's all kinds of different ad network tracking things.

Like it I have to handcraft every show episode now, depending on which of the ad networks it needs to report to. Interesting. I thought it was all going through traffic.libsen.com now. No, I mean that's where the mp3 file sits. So that's always where it resolves, but it there's anywhere from one to five redirects in front of that where it goes to all their ad agencies.

It's yeah, it's it destroys the stats. The stats are a complete disaster all the time. Um that's got that's the pod roll stuff too, right? That's like their stuff. Uh pod roll and there's like three or four, it just depends on who the advertiser is.

They all use different platforms to track their ads. So if they sponsor a specific episode, yeah, that episode has to get routed through. The call has to get routed through that before it ends up at libsen. That seems very complicated. Yeah, well, I mean, it's this is something we never talk and brag about, but we actually do protect.

We uh unlike most podcasts, our listeners, their privacy is protected. Like the first hit that they do is through um uh through our agency, and that's kept anonymous. Like even the agency doesn't know who you are when you're requesting the that first hit, and then the agency passes it on to any other ad tools after that. Um so and like I said, it makes two or usually no more than two or three hops after that before it heads to Libson. Huh.

But yeah, it's you know, if you're an advertiser, you can both get your stats and we can maintain the privacy of listeners. Um, but yeah, it just makes it all an ugly mess. Um the other thing that while this is churning away over here, the other thing that you can do is if you have the time and you have the the credits uh in your AI plan of choice, you can build your some of these utilities yourself. So if you wanted to have a an audiogram maker that was completely local, I haven't tried remotion, so I have no idea whether it's any good or not. But I was making one for myself in Python called byliner.

Oh, looks like Claude has more questions. Sponsors need to be removed from the transcript of LM assisted cleanup recommended, yes. Plain HTML5 audio controls, uh match the Libsyn frame used on recent posts. Uh let's see where should the new transcript pipeline data live, new tables in the existing database, the fireflies. That is fine.

So we're gonna say this is fine to use this Fireflies account. And the way we will handle this is that you will load five MP3s at a time, wait for Fireflies to transcribe them, get the transcripts back, then delete the MP3s to free up space, and you'll do this in batches of five so that we don't overwhelm the system. Yeah, because it is totally storage-based, isn't it? Like you can keep it grinding 24-7 as long as you don't let this the recordings pile up, right? Exactly.

Um, so the real that will let it work within the boundaries of that system. There are other systems that are just like per minute per month, like you have so many minutes per month. Right. And that's another area where it's not a bad idea to have a local version. So uh in a in another thing, I've got a uh a Python project that I'm working on to do that diorized output to fix basically to to create those diorized audio transcripts so you don't have to rely on a service at all, particularly for old old material, old stuff that you don't really want to spend money on if you can avoid it.

If you can just spend compute power. But this is an example of the byliner, and I'll just put that on screen. Um that you can't hear the audio to it, but it's just an old episode, and so it's using the audiograms to create the waveforms, and then it sticks the tr the uh captions in the bottom. The oh, all right, yeah. This is just this week, Jerry Pullman's here from uh Red Art.

Yeah, exactly. So this takes something from a style guide I was messing with and has a Python script in the back end that essentially says, okay, here's the audio file, turn it into this style of audiogram and make it an MP4, which is a video. There is a free open source project. Um, let's see if I still have it up here. It is called YouTube.

Let's go to the GitHub repository that allows you to programmatically control your YouTube channel. So you wouldn't do this. You would install the software you which is free, uh, it's open source. You would security check it, obviously, with your your coding agent of choice, and then you would give it to a tool like Claude Code to say, okay, once you've generated an audiogram, here's the target YouTube channel. Generate the title, generate the description, generate the tags, maybe make some cover art, and then programmatically push that video to YouTube.

So that and maybe even schedule it. So if you want to do like you know, a best of marketing over coffee 2009, uh, you could push you know all 52 episodes from that year into YouTube. And the key about this with a tool like the local generations is again, you don't do this manually. Like this is all handled by machines because this is this is grunt work. Yeah, right.

This is the worst of the worst. It really is. It's it's just terrible, uh, terrible laboring for things that are a proven process. So that's when we talk about automating content transformation. This is the overall process.

If you have an agent tool that can essentially write the programming, the software for you, and you can be super clear about your requirements, you can do a great job. And the framework that we recommend to do this, huge surprise, Katie's not here to talk about it, is the 5P framework by Trust Insights, purpose, people, process, platform performance. For this kind of project, the purpose is to transform content and also fix up the website to clean up all the cruft, right? Um, the people is you and me who don't want to be doing this, um, but also the audience that might want this information, and it falls into people, it's not people, but by having full transcripts, it will feed AI, it'll feed the machines that are crawling the marketing over coffee websites. So that's an audience as well.

The process is exactly what I dictated out aloud. The platforms are all these different connectors and that we talked about, and then the performance is did the software do the job and get the final stuff on say. If we were to go back to the marketing of a coffee website in two weeks when all this processing is done, is it up to scratch? Um, you know, are do episodes have full transcripts on them? Are the sponsor blocks been replaced by you know short codes?

Do the libs and URLs all work correctly? So all of that would be the performance criteria to say, yes, this did the thing, and then later on, in tools like Google Search Console and Bing Webmaster Tools, we would see an increase in traffic, in impressions, in clicks, in AI citations, because we gave our audience, which includes these machines, much more content to work with. Yeah, that's just uh even the fact that it's just less offensive to the eye, you know, less broken stuff lying around. Like that's just providing value. But yeah, unlocking the content and cleaning stuff up is huge.

And like you said, that's just the most horrible grunt work. So to automate that away is fantastic. Exactly. So obviously, this is not going to finish during this episode or the next episode, or probably the episode after that. This is going to be a very long running project, but this is the overall process, and you will be most successful following the 5P framework by Trust Insights.

So once this is done, then we can measure it and see how it goes. But also, this now will give you, as the producer, a lot more leverage with sponsors to say, like, yeah, you gotta get if you want your name on still, you know, in the in the historical archives, you gotta keep paying. Yeah, well, and that unfortunately, with the vertical that we're in, you know, there's a good 30, 40 percent of them that just don't exist anymore. You know, they were either bankrupted, acquired, you know, became some other tool. Like there's a lot of just stuff that doesn't exist anymore.

Mm-hmm. One of the things that and this is something that again, content transformation, theoretically, theoretically, um the when you do the host reads in the show, that actually has a different audio signature than when we're recording, because you know, when we're doing when you're doing two parts, you get different audio signal. You could have a tool that splices the MP3s and then you know goes back and adds new advertisers. Yeah, right. Just cuts in a new choke.

Well, and that yeah, that's always been, you know, that there are a number of networks that do different stuff like that that you know, you just tag where the ad goes, and it's always dynamic. In fact, some is dynamically inserted where different listeners get different ads. It could, you know, it can go beyond just the fact that they can be swapped in or out. But as I, you know, as we're talking about this too, this totally gave me a gave me an opportunity that I hadn't considered, which is that there's all of these episodes with sponsors who do still exist that it's pointing to their old event or old white papers that don't matter anymore. I could definitely, you know, just run a list of like, okay, X sponsor, you know, did you know we have 300 pages going to your stuff?

You know, where would you like these links to go today? And you know, for a nominal fee of you know whatever we can 1 million dollars if you for just one million dollars we can point this to somewhere where you can have this track instead of us being the number one source of 401s to your website. Like you don't want that. That's a really interesting point. So if you were to do an outbound link audit of all the using an SEO tool of all the links that are going from this sweat from our site to dead content.

In fact now I'm just curious. Now we're just going to go completely off the reservation here and take a look around that that was I probably need to retire that term because that's offensive. Let's take a look at our SEO tool. Let's do a site explorer. Let's do marketing over coffee dot com.

And let's take a look at outgoing links and let's look at 404s. All links that are 404s 165 that are now 404. Yeah that's not horrible given that's that's like about what I would expect which is bad but not horrible. You know, the other interesting one too, though, is I hadn't thought of this. You know, you can go to the sponsors too, but you can also go to the competitors.

Like, I'm not going to take that's a when I was talking about the duck taking duct tape and chicken wire. That's you know, there were times where certain sponsors wanted certain keywords to go to their site. And, you know, if they were replaced by another sponsor, those links moved, you know. That's a it went with the sponsor. Mm-hmm.

So this would be the report to feed to a system like Claude to say, oh, by the way, we need to repoint all of these to internal pages or remove the links entirely because they're they're just dead. They're they're long gone. And um, in the database that that tracks all this information on the back end, put a field saying like sponsorship opportunity, uh, so that you can then make a report from it to say, hey, for the category of email marketing or for the category of marketing automation software, CRM software. We've got all this stuff in our system we could dynamically repoint if you want to sponsor, you know, us uh run of site for this particular term or concept. Yeah, that's a good, and I realized too, I've missed a shot with I've always been tagging the episodes as podcasts, you know, because it's audio content.

That's my first thought, but they should really be tagged as like if the majority of this episode is about email or a website or branding or you know, just whatever, having a few extra tags in there makes it easier to categorize that stuff as far as stuff that could be sold later. The other thing is that, and and this is more of a general content transformation, you can transform content in reverse directions. So I'll give you a quick example of this. If I go to my YouTube channel, and this is something that I do for myself, if I go to YouTube, I go to content and I go to an older episode here. Uh let's see this one here.

Yeah, falsifying CRM data. Uh, these are my my kitchen recordings each week. Of course, the content is there, right? The video is there. But then on my website, what I have is I built a utility that connects to my YouTube channel, grabs the transcript for each uh episode, and then creates a let's see if I can just view this page.

Creates a blog post with the summary, the YouTube video, key takeaways, and then the transcript. So exactly what we're doing with marking over coffee. But now instead of me having to manually do this, because this sucks, um, I record the videos, and then once a week I run this utility that syncs it back up with my website so that all my content that I've put posting on YouTube, YouTube has a place to live here. Right. And so the initial YouTube outload that is that through Adobe, where that transcript is originally created from, or is YouTube creating the transcript?

So YouTube creates a transcript, and what I do is I will grab I grab the YouTube video and then there's a free utility called um FFmpeg that will turn any video file into audio. And then there's a free utility for the Mac called Parakeet that takes an audio file and transcribes it. So I can I just have it when it downloads the video, it does the transcript. And then my local AI tool, uh, which uses uh Quinn 3.6 for this, uh essentially just distills this down into this this lovely thing here. And and what you get is a nice summary, the key takeaways and stuff, but it's very clearly based on my original human luck work.

That's very cool. Yeah, that's uh great way to have a whole chain going, so it's not like all separate mess everywhere. Right. Because I this sucks. Like daily content having to maintain this, just is terrible.

It's it's awful. I don't I never want to do it. Um and I for years I did not do it. When I ran this for the first time, it said, I have 700 of your videos to go back and catch up on. I'm like, all right, I'll see you in two days.

Yeah, right. But if you've got, and this is a challenge that a lot of marketers have, if you've got content all over the place, this is a this methodology that we're talking about of automated content transformation, allows you to sync things up. So if you got podcasts over here and YouTube over there and stuff, what I we always, and we've been advocating this, you and I have been advocating this for 20 years now. It has your content should live on a site you own. It's fine for it to be on YouTube and in those places, but things happen, things change.

Yeah, right, always fall back. And the great thing is, yeah, a site, if someplace goes away, it's all still at your place. And then when the new king shows up, king or queen comes to town, you can just automate pumping the stuff into the new thing. So it's yeah, if you don't have a storage bin of your own, you know, land that you own somewhere, you're it's uh isn't that that's an idea it again becoming the crabby old man yet again. But there's just the like the landscape is littered with people we've known who became huge on like this platform over here, and you know, then it died, and that was it.

They just went back to you know, Joe win a cubicle after you know having made tons of money in certain verticals. So yeah, it's it's real. Like you you rise and fall. Yeah. You're right.

I remember now that you've mentioned that I remember some very uh big fish and small fishbowls back in the early days of podcasting. You were like, Mr. Podcast, and then you know, then you had podcasts winter for 10 years, they vanished, they're back in a cubicle, and now you know you have you have the the Joe Rogans of the world. But yeah, if you can stay adaptable to the different platforms without having to do all the reworking yourself with a gentic tools, you'll save a lot of time and you will keep your content. Right.

Yeah, yeah. You stay out in front of the world, and as the new thing comes up, you've at least got content ready to go. You're not starting at zero. Exactly. Um, so that's that's automated content transformation.

Develop a requirements document, which is essentially uh software development stuff. Use an agentic system, use the 5P framework by uh Trust Insights. Figure out which Martech tools you have that allow for connectors that are optimized for AI agents. And that by that it's gonna be what's called a command line interface, CLI. That's the magic word.

Forget MCPs, MCPs suck. Uh use a CLI because it doesn't consume AI usage. Um if a vendor does not have a CLI and an competitor does, think about jumping ship or telling your vendor, hey, you either got to keep up with the time to give us a CLI or give us a rate cut because this other vendor has the thing that makes AI work really well with it. Connect your tools in the requirements document and all the CLIs. Think through the process through user stories and things like we were talking about today, and then let the tools go off and do their thing.

And within uh a week or so, you'll probably have a working system. That's I hadn't thought of the angle of skipping the MCP because that's just saves you tokens, but that is totally it. Like the why would you use MCP when you've got CLI at no charge? Exactly. No charge.

It's faster. The results are better. Um, no security threats. Fewer fewer of them, anyways. Yes.

No one's spying on your work. Hubspot. Um anyway. Uh that is going to do it for this week's episode on next week. I forget what we're talking about next week, but the week after that, we're going to be talking about uh diagnosing AI writing styles.

So make sure you you stick around. Uh subscribe to wherever it is you are watching or listening this show, and we will see you all on the next one. Take care. 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. And a 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.