So What? Whisper and Claude for content repurposing

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? we focus on Whisper and Claude for content repurposing. We walk through how to set up an AI-based content repurposing workflow, considerations for content repurposing and when to build the system yourself or hire experts. Catch the replay here:

So What? Whisper and Claude for generative AI content repurposing


In this episode you’ll learn: 

  • How to set up an AI-based content repurposing workflow
  • Considerations for content repurposing
  • When to build the system yourself or hire experts

Upcoming Episodes:

  • TBD

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

AI-Generated Transcript:

Katie Robbert 0:29
Well, hey there happy Thursday, everyone. Welcome to so what the marketing analytics and insights live show. Today I am joined by both Chris and John. It’s been a while since the three of us, we’re all in one place. And of course next week that will change again. But at least for this week, we’re back together, the band is back together. How is everyone?

John Wall 0:48
Good. We have to put together a sophomore album, right? We have to get back on the dredges and come back together because we need the money.

Katie Robbert 0:56
That’s right. And then Chris will go take it on tour next week.

Christopher Penn 1:00
exactly going on tour.

John Wall 1:02
That’s right getting on the bus.

Katie Robbert 1:04
But this week, we’re talking about using artificial intelligence for content repurposing. Specifically, we’ll be looking at whisper and clawed, which are two different artificial intelligence tools. I’m sure Chris will tell me that there’s hundreds of tools out there that do similar things or, you know, do pieces of things. But today, specifically, we’re focusing on focusing on whisper and clods. So where do we want to start?

Christopher Penn 1:31
Let’s start with the purpose. The first of the five P’s cany. What is the purpose of content repurposing?

Katie Robbert 1:40
The purpose of content repurposing is basically to get more bang for your buck to scale. So you create something once and then you reuse it in multiple different ways to reach different audiences on different platforms. So it’s, the purpose is efficiency, the purpose is scaling. And the purpose is broader reach.

Christopher Penn 2:02
Okay? But does that really that that doesn’t really fit into a user story very well, like that does it?

Katie Robbert 2:10
It could fit into a few different user stories. So as a CMO, I want to repurpose my content so that I can keep costs down. Or as a CMO, I want to repurpose existing content, so that I can find efficiencies in my process or so that I can have my team focus on other more meaningful tasks.

Christopher Penn 2:39
Got it? Okay. I like that first one, because that segues nicely into the process that we’re going to talk about today and the platform. Because there’s two different avenues you can take, there’s an easier avenue that costs money. And then there is a more technical avenue that costs no money. And so the question that folks we’ll need to entertain is okay, which which road do I want to take?

Katie Robbert 3:04
Well, and this goes to something we were talking about in our free slack group analytics for marketers yesterday, which you can join at trust, for marketers, I was asking the question of how do you decide which AI tool solves your problem? And it was, in some ways, it was almost a trick question. Because you never want to start with the tool, you want to be very clear about the problem you’re solving. And so with the user story that I gave you, Chris, as a CMO, I want to repurpose content to keep costs down, adding an additional tool might not solve that problem for me.

Christopher Penn 3:43
Exactly. And it is we haven’t talked about the second P yet, which is in this in this journey, if you will, the people it depends on their skills, right there, the easier way cost money, but does not require people to be technical. The the lower cost way from a dollars perspective requires your people to be a lot more technical than then the easy way.

Katie Robbert 4:11
But it sounds like then there’s a trade off because then you are asking for technical skills, which could cost more even though the tool itself could cost less. And so those you need to factor in if you’re, if the purpose is to keep cost down. It’s not just looking at the cost of the tool. It’s looking at the cost of the people who have to operate the tool as well. So you may be trading low cost software for higher cost talent.

Christopher Penn 4:36
Exactly, exactly. So let’s talk about the process because it makes logical sense before we dive into the platform. What we want to think about is the content that that is the richest is typically stuff like this right? Like an episode of so what are an episode of marketing over coffee, or an episode of In-Ear Insights where we as people are seeing a whole bunch right we Talking anywhere from 150 to 200 words per minute, you know, depending on how much caffeine we’ve had that day, and when you have a 30 minute podcast or a 45 minute live stream, you’re talking about 1000s and 1000s of words, that’s a lot of content. Now, not all of those words valuable right there. Sometimes they’re discussing, you know, for example, like, how do you improve the quality of chickens you can get, but those words are still content that you can repurpose that you can be part of the repurposing engine. So the first part of the process is identify what content you want to repurpose. And I would suggest for marketers who have content laying around, particularly podcasts, videos, webinars, or B2B Friends, live streams, and especially if you’ve done a good job with your public relations, guest interviews, where you’ve been the guest in other places, those are all the I would say the starting points the process.

Katie Robbert 6:00
John, do you know how many words per minute you speak?

John Wall 6:03
Yeah, we are in the over caffeinated category, you get up to 151 60 and 180. And one thing, you know, from podcast research, your brain can take like up to 2x, you can, you can handle much faster. But on the flip side, I have more than one one star review in my AUDIO BOOK, because people say it was way too fast, and they couldn’t take it. So you have to wait to the lowest common denominator on that. And that’s the way to go. But it’s really interesting. I hadn’t thought too much. I liked that idea of other interviews on other channels, being able to summarize those and bring them back as content on your site. That’s a cool angle. I’d never thought of that.

Christopher Penn 6:41
Exactly. So in fact, one of the interviews we’re going to be working with today is a interview with Katie, when she was on the b squared, TV show the b square TV, so we’re going to use that as our example. But you can do this with anything. So let’s talk about the materials that you’re going to need. So the the Meuse on plaas. If we were cooking, for today’s for today’s Cook, you’re going to need some kind of large language model can process a lot of text. So the two that are capable of it’s right now is GPT-4. OpenAI, which is having some issues right now. And one called clause two from anthropic, which, for today is my choice for doing this kind of work. Claude is not as good a coder and doesn’t do as well as with reasoning. And it doesn’t do well with like mathematical logic. But it is really good at language. And it has larger context window has a 100,000 token context window. So for folks who are unaware of what that means, a context window is essentially a large language models working memory, how much can it remember for stress to forget stuff, if you remember the old days when ChatGPT first came out, you could be typing and having a conversation with it. And then like 10 minutes in, it just starts forgetting like things you’ve talked about, like what happened like you, I told you already, this is a blog post about this. And it’s like, ah, what happened was the text, kind of scroll away, and after a certain point, it forgets what happened, because it can only hold a certain number of words in memory, Claude can hold 100,000 tokens, and a token is about a four letter segment of a word. So the word of would be a complete word, the word example, the first token B exam, and the second token would be P le. So Claude can hold on to about 60 ish, 1000 words at a time, which makes it ideal for dealing with long transcripts and things.

Katie Robbert 8:44
If I were to put on my tinfoil hat, which I do like to do from time to time, because I do love a good conspiracy. My tinfoil hat side would say that perhaps the makers of these tools purposely made these tools forget what you talked about, because that’s what it’s like to talk to an actual human. And this is their step towards making these tools more human like and sentient, because, I mean, how many times have you talked to someone and they’re like, Wait, we talked about this already? Wait, Hmm, wait, wait, what were we talking about? I got distracted. And so of course, obviously, I know logically, that’s not what’s happening. But tinfoil hat version of me was like, oh, maybe that’s how they’re trying to make it seem more human.

Christopher Penn 9:29
There is a whole rat hole. It’s a very deep rat hole around things like memory and memory chains and you know, the Lang Chain system, which we’ll do another time. But yeah, that is a thing. So you’re going to need some Cornett kind of large language models for access. The second thing you’re going to need is the media itself, right? So have your YouTube channel have somebody else’s YouTube channel where you are a guest on it. Please don’t do any of these techniques with content that isn’t yours. Right? That’s all and ethical. So, you know, if you’re if you are the guest on someone else’s show, that’s fair game because you’re there to provide value. And presumably, at least I don’t know about you guys, but I don’t get paid to be a guest on other people’s shows. So my fee is I get to use the media how I want. But yeah, don’t you practically can use these techniques with content that is not yours, you you ethically cannot and legally, probably not.

Katie Robbert 10:26
I think that’s solid advice.

Christopher Penn 10:28
And now comes the third and messiest part, how you turn video into text. And there’s a couple of different ways to do this. The first thing you need is some way to grab the video itself. And there is an there are paid desktop tools that you can do this, like there’s one called YouTube Downloader. There are also open source tools, there’s one called Why T DLP YouTube download program, that if you are technically savvy, you can download this code and run it on your computer. And this will allow you to give it at the command line a very short command that would be goes like this to go i t y t DLP, and then the URL of whatever content you want to download that will grab that content from the internet, assuming it’s publicly available, and put it locally on your hard drive.

Katie Robbert 11:26
But I would assume if you’re you know, and this is, you know, every situation is going to be different. If you are looking at your own content, you likely have the source file that had to get uploaded to YouTube in the first place. You know, things don’t just magically appear on YouTube, they have to come from something. And so for, you know, for us this live stream, you know, it gets, you know, brought into YouTube after the show, but I have all the source files. So if I have that I don’t need to go through this process. Right?

Christopher Penn 11:57
That’s correct. This would be if you were guesting on someone else’s show, for example, got

Katie Robbert 12:00
it? I just want to make sure I was clear on the use case.

Christopher Penn 12:03
Yes, exactly. Or, you know, maybe you’re like me, and you were you’re you had published a whole bunch of stuff 10 years ago, and you’ve lost all those files, because they’re on like five, five MacBooks. Previously, you know, either way, you can do it that way. Or if you don’t have the source files yourself, that’s a great way of doing it. The second thing you need to do is convert the video to audio. And again, there’s a number of different ways to do this, if you are familiar with and capable with Adobe Premiere, or iMovie, or Camtasia, you can do it in those tools, you can load the video up, and you can say Export to mp3, or WAV file or whatever. There is an open source tool, very, very powerful one called FFmpeg that again, it’s a it’s an amazing tool that will convert pretty much any media into almost any other media. So that’s a tool that I typically use. But again, you you almost certainly have a piece of software on your computer that will convert a piece of video into an audio format.

Katie Robbert 13:03
So why do I Why is that step necessary?

Christopher Penn 13:07
That step is necessary because a lot of the transcription tools need audio in order to work, they can’t ingest video natively, because they’re not equipped to handle the multi gigabyte files that that a lot of these videos are. Got it you know we are we’re recording this in HD, full time ATP HD. So this file this, you know, the 3545 minutes we’re gonna spend together today is going to weigh in at about three gigabytes. And no one wants that crossing the network if they can get a 45 megabyte mp3 file instead.

Katie Robbert 13:39
I mean, I’m always saying that I always say in that job,

John Wall 13:42
it’s all about the gigabytes.

Katie Robbert 13:46
Alright, so we have our video file, we have an audio file. Now what do we do?

Christopher Penn 13:52
Okay, fork in the road time. If you want a the easy but costs money approach there is a fantastic piece of software called otter, which is one that we’ve used actually a Trust Insights for action since it came out. And all it does is you drop your audio files in you pay your I believe it’s I want to say it’s like 60 bucks a month now to do this, and it will take your file and transcribe it. And then from there, you would export it as a text file. You know, that’s, that’s probably the easiest solution. And by the way, there’s some really cool new features in otter that have just been added relatively recently one of which is auto chat, which is their integration of a language model. So you can do something simple with this like what are the major talking points from from this episode? So this is our last week’s sowhat episode, and it will go through and attempt to distill out those talking points. Which is is nice and handy. This is really good. By the way if you are recording with have permission, client calls and then you want to like get action items from your your meetings. This is a phenomenal way to do that.

Katie Robbert 15:08
Well, and as we’re in the spirit of content repurposing, having something like auto chat, summarize the main talking points gives you an opportunity to repurpose these small snippets on social media where you might have a character limit or you just want to have a very short summary of here’s what this thing is about, rather than having to rewrite it over and over again.

Christopher Penn 15:32
Exactly. Now, if you want to go the the free route. There’s a package called whisper from OpenAI. So this is the same OpenAI that makes ChatGPT. But they open sourced their their whisper software, a couple of years ago, actually, I believe it was couple years, it might have been a year ago. And this software, again, when you download it, and compile it runs locally on your computer, you will will feed the audio files you were making to whisper. And what it does is, and it does some things that that otter and other tools just aren’t capable of, it will go through and it will, a do a transcript to make your subtitles for for YouTube. One thing that it does, that I’ve not seen any other software do even closely well is it will automatically translate. So if you have audio in another language, you can feed it through and automatically translate into a target language, which is super, super cool, and very helpful if you’re consuming any kind of international media. But this is this does what otter just did to which is does the transcript and makes you sell you a nice text file.

Katie Robbert 16:38
Okay, so we started with a video of us talking, we started with some kind of an interview, we got the source file, we basically converted the video into an audio file, transcribed it. So now we have the written content from the original video. What do we do now?

Christopher Penn 17:01
Okay, so the next step is you find that video file of that text file that it made. Let’s go ahead and go to our text files. And we load we just attach it as an attachment into this into Claude. And now the question is, what do we want to do with this right? What kind of thing would we want to do? So I’m going to use a prompt that I use for this sort of thing. And I’m going to say we’re gonna summarize the transcript. This is a conversation between booksellers of E squared, media and Ed robear. of Trust Insights, write the transcript is not diarized. You will need to infer the speakers from the words. Katie and Brooke talk about equal amounts of time. So what we’re going to have to do is say first, you’re going to draft a blog post in a professional tone of voice that summarizes the transcript. Use active voice. Avoid adverbs. Avoid business jargon, these last three commands are really important because otherwise you tend to get sort of a very bland, uninteresting summary, whereas if you tell, okay, you can’t use adverbs, and you’re not allowed to use passive voice, it, it tends to create better results.

Katie Robbert 18:34
You would spelt trust wrong and of Trust Insights. Oh,

Christopher Penn 18:37
no, we’ll fix that. Actually, it fixed on its own. Okay, so now what it’s doing is taking this this multi page document here and it is spitting out essentially what what you and Brooke talked about. So, you talked about predictive analytics empowering marketers, the two main types of predictive analytics for social media marketers, you gave an example of using time series. So this is essentially what the the the discussion was about. Now I’d say now write out the top five bullet points from the transcript

type driver analysis time series now its predictive analysis clean customers analytics data is crucial for accurate predictive and predictive analytics as marketers be proactive rather than reactive. This looks good. Okay. Next, create five different tweets of 140 characters each that could be used to remote the viewing of the original episode. Be sure to use what is the b square or handle? Is it Hello b squared?

Katie Robbert 20:01
I believe so yes,

Christopher Penn 20:04
it will be squared. And at Trust Insights, as the Twitter handles and use at least one hashtag in each tweet, the URL should always be present, use Trust Insights as the URL, let’s see how it does.

Katie Robbert 20:34
So you’re in like a step wise, do you have to go through this exact set of steps to get to like the five tweets? For example, do you have to first you know, clean it up and say, you know, provide the summary. You know, we don’t know who was talking when then the four points and then I’m wondering, is there a process to getting to this set of content? Or can you put in the transcript and say, give me five tweets,

Christopher Penn 21:03
you can put in the transcript and say, give me five tweets, you can actually skip to that step. It there is a slightly higher chance of hallucination if you go straight to the the end, because one of the quirks of language models is that they need runway to think they need time to think and the more processing and input and output you have them doing, the better their thinking tends to be.

Katie Robbert 21:27
So hallucination is basically that unfocused output that you get back,

Christopher Penn 21:34
yes, or just outright lying.

Katie Robbert 21:37
Well, if the machine it’s not lying, it’s just not giving you the right information. Right, right.

John Wall 21:43
Exactly what you said about focus is definitely about right, because it’s like the 10th query is always going to be more accurate than the first because it doesn’t, you know, it’s still hunting around to try and figure out where it’s going.

Christopher Penn 21:56
Exactly. So now, we have five lovely tweets. And these are actually pretty decent, like, I would be okay with putting these up on on our social channels.

Okay, yeah. So now we’ve gone through and the last thing we can do, is, we’re gonna say, let’s take that transcript. I’m gonna say do not summarize, rephrase, rephrase, and protects only fixed grammar, spelling and formatting. And we’re going to have it tried to process the actual transcript itself, because it comes in as just a big pile of words. And by having it go through and start to process this down, it will try to clean up now I can guarantee you, it’s going to get the diarization wrong, which is the assignment of speakers because I’m using the Whisper version, whisper camp in its current version, do speaker differentiation. So if you’re doing speaker, if it’s important to delay who’s speaking, you either have to give it those explicit instructions, or use the the version that comes out of the paid product otter, which has the speakers already delineated, and then you can, then it will clean that up. But this is just an example of how you would now use this to make the maybe a longer form transcript for your website. So if you think back, for example, to maybe earlier podcasts that we did, or maybe you know, in John, in your case, it would be you know, episodes 100 through 600 remarketing over coffee from 10 years ago, this would be a great way for us to be able to take those old audio files, get them transcribed through whisper at no cost, and then have massive amounts of content for the marketing over coffee website.

John Wall 23:41
Yeah, it’s funny because it just some we’ve used tools over the years that are so horrible. If you could even just run a cleanup on some of the existing ones, I’m sure they would actually come up better.

Christopher Penn 23:51
Exactly. Now, what we’ve done, we’ve done a lot of this, this step by step, piece by piece, one of the things that if again, if you have technical access to technical talent on on team, a good portion of this process can also be automated, a good portion of it can be glued together to the point where all you’re really doing is copying and pasting into a system like anthropic actually anthropic itself has an API. So if you’re technically skilled, you could do the entire process from beginning to end. So I have a piece of code that I wrote, that essentially takes in the video URL, and it processes all the steps up to anthropic clause. So it will download the video transformed the audio, it will clean it up, it will run through whisper, it’ll create the transcript and then all I have to do is go take that into put into clause. And eventually I will take I will have a piece in here that will talk to antibiotics API directly. But that’s the most advanced version for scaling your content production if you’ve got 200 episodes of In-Ear Insights Do you probably want to have at least a good chunk of it automated so that the amount of time you spend on is minimal for the processing side?

Katie Robbert 25:09
So let’s talk for a second about marketers who aren’t Chris Penn, marketers who are more like myself, John sort of straddles the two worlds between me and Chris. So, you know, we’ll see what side of the fence you land on, John, but, you know, the process you just described is not something that feels attainable for me, I can’t code. You know, so I would be doing it in a clunkier way. I’ve also gotten a lot of sales pitches from companies who’ve stated that, like, oh, I listened to your podcast, and I want to repurpose this content for you. At what point do marketers or companies rather start to think about investing in doing it themselves versus bringing on a company or an agency to set up a system for them, you know, to either fully do it and in the back different kinds of content, or to bring them, you know, as you were just saying, like, up into the point of just dropping the file into a system like, like ChatGPT, or whisper, or Claude,

Christopher Penn 26:17
this is gonna be the answer that we give all the time. It depends. And it depends on really on the purpose, you know, the set first P is really important. If this is something that you’re doing just for maybe SEO purposes, and you don’t have a ton of content, then you might want to look at one of those all in one tools, just because their costs are not egregious for the most part, you know, 100, couple 100 bucks a month. And if you’re doing a piece of video content a week, this process, this process that I do, like for example, Katie, you could do most of it with otter and Claude, right? The only part where I think it’d be a sticking point? Actually, no, because you have all the video and audio files, so you could do pretty much all of this. If you’re doing one file a week, I would just do it yourself. Right? There’s the if you’re if you have to do hundreds of files a week, maybe you’re a media company, then yeah, you probably want to look at some kind of system that it really was soup to nuts. And that critically had someone you could call for help when when the plumbing got jammed.

Katie Robbert 27:28
So John, is after seeing all of this, are you now sort of considering going back to those older episodes of marketing over coffee and repurposing them as fresh new content?

John Wall 27:39
Yeah, I mean, that’s always been the list is to try and get more of that stuff on the library. You know, unfortunately, the website traffic is not primary income generation, you know, it’s really more about getting the content the advertiser square. So, you know, keeping the blog up to date is pretty low on the priorities for getting stuff done. But, you know, I do know, there’s podcasts networks where this has become insane. Like they have millions of hits a month, because they’ve got now 510 years of transcripts up there, whatever. So it can definitely be a path if you just want to get right on it for advertising, especially if you have a ton of content is a great way to go.

Christopher Penn 28:22
The Go ahead, there’s extra value to for marketers where you don’t control the source content, right. So real simple example here. This is one on my blog, I was on Mike Stelzner, Social Media Examiner show. And it was a fun chat. He put the video up on YouTube. And so what I did was I do exactly the process we just went through is okay, summarize the show, right? There’s a big ol warning up there saying it was in from May, so it’s already out of date. But here’s the major bullet points. Here’s the video itself. And then here is the machine transcript. Now again, I feel comfortable doing this because I was the one providing, you know, 95% of the content for the show. I have another one I do with Evan Christelle. So for me, this is a good way for me to reuse the time I spent on somebody else’s show that I don’t necessarily get a lot of direct value, we get exposure to their audience and things like that. But from search purposes, this might be a way to take advantage of this. I would even say this is probably something that from a process perspective. I would also eventually want to be able to do on the Trust Insights website put it right on our blog, too, because hey, why not get it gets much of mileage out this content we possibly can. Any kind of guest appearance where you’ve been on shows Katie, you’ve been on marketing products, you’ve been on b squared. You’ve been on Mar tech, you’ve been on a whole bunch of these shows, talk on things that you gave up your time, essentially in exchange for exposure, but we could then take the words that you had to say and put them to put him to work again.

Katie Robbert 29:53
Ironically, I’ve done quite a few podcasts interviews recently that will launch soon. And the question is always, you know, what can I do with generative AI? And hey, look, it’s so it’s like very meta of itself, where we’re using generative AI to repurpose the content as we’re talking about generative AI.

Christopher Penn 30:17
It’s like Inception, except with blog posts.

Katie Robbert 30:20
Not nearly as interesting,

Christopher Penn 30:22
not nearly as interesting. And so that’s the, that’s the structure that I would recommend. You can use. For example, you can use Claude to for just repurposing other stuff too, and for doing advanced editing, so you could take old blog posts, perhaps, and have it refreshed and have to update them. So for example, you could take a blog post that was about Universal Analytics. And as long as you had your own text file, like we did with the transcript, you had a text file, here’s everything that’s changed. Since blog post came out, you could tell quote, here’s the original blog post, here’s everything has changed. Rewrite this blog post using Google Analytics 4 As your knowledge base instead of Universal Analytics. Sometimes it will work, sometimes it won’t, but it is capable of at least that substitution logic to go ahead and do those rewrites. We were talking in analytics for marketers earlier today, which again, as Katie mentioned, if you haven’t been there, over there go to trust for marketers about developing a prompt to do sensitivity reading. So in another community, we’re in the spin sucks community, someone share this really nice piece of background information about decolonizing language, non violent language demilitarized language from the American Psychological Association, and I said, Okay, Claude, here’s your knowledge. Here’s a prompt, I want you to read through this blog post I’ve written and identify any problematic language and tell me how to fix it. I put what I put our so what transcript from last weekend, and it said, this is a great post, no changes needed. I’m like, thank goodness. And then I put it in an episode of all in the family, the 1970s. TV show, it’s like, Hey, here’s a long list of problems.

Katie Robbert 32:05
Well, I mean, that one, we knew, if anyone under if anyone knows the TV show, they know that would be problematic. But I understand that for the sake of example, it’s probably a good one to use. But ya know, I think, you know, it’s interesting as you’re describing, sort of the use case of converting Google Analytics content, Google Analytics 4, but you have to first list out everything that’s different. You know, in those cases, I can see pushback from writer saying, wouldn’t it just be as easy for me to rewrite the content myself. And so this is where you have to take it on a case by case basis of how much needs to be written how many versions need to be rewritten for what different audiences the different tones, the scale of it all, versus using a system like Claude to do sensitivity reading, that seems to me like a really good use case for all of your content, you know, versus bringing on a personnel I would highly recommend, you know, consulting with an expert to create those sensitivity prompts to make sure you’re not inadvertently introducing new problems into your content and that you’re getting the language correct. But once you have that initial consults done, then it’s just a matter of rerunning the prompt on all of your content over and over again. And that’s where a lot of that automation comes in.

Christopher Penn 33:26
Exactly. And here’s the thing, as we saw with clods interface, it’s got the little paperclip icon where you can attach files. If you are a smart marketer and a clever marketer, you will start building libraries of knowledge that you can drop into language models, right, maybe you have a just a running set of notes. In fact, I remember back in the day, when we worked at the agency, we had one of our team members, wrote like a 22 page document on everything she learned when she was taking the Google Ads course, right. And she just summarized it, boiled it all down for the team. And we were like, wow, this is really impressive. Today, I would tell that same team member, this is awesome. We’re not going to use this as training data for when we use a system like law to say, Okay, I’m gonna write a blog post about Glads. Here’s all the background and face those clouds going to need to write an intelligent pose or ChatGPT or the system of your choice. But if you if you’re a marketer, just like we talked about, you know, plenty of times on on In-Ear Insights, which if you’re not subscribed, go to trust podcast. Now, should you have a prompt library of the prompts that you’ve written that worked well for you, but you should also have a training data library of a valuable background information that can really help make models do better.

Katie Robbert 34:43
It occurs to me so you know, John, off the top of your head, how many podcast episodes are there of marketing over coffee?

John Wall 34:52
730 ish.

Katie Robbert 34:56
So it strikes me that if you wanted to build a marketing over coffee, sort of branded toned model, you certainly have enough. You have more than enough content to work with, because it’s always you consistently hosting it. And, you know, talking about it like, you could build a John Wall large learning model very easily because you have enough volume of content to do so.

John Wall 35:22
Yes, and I see a file pile here. I think there’s something.

Christopher Penn 35:27
Yeah, we have so much content.

Katie Robbert 35:29
Yeah. I think that, you know, it’s, it’s interesting, because John, you had said that, you know, updating the blog is such a low priority. With the advances in this technology, do you think that that’s an opportunity to use some of the automated processes that Chris has outlined, and bumped that up higher on the list? Because it’s not as much of a burden to you, as the individual?

John Wall 35:50
Yeah. And then there’s the whole value of using that as training data, like you said, to have something that can answer questions or come up with new topics. That’s really where, you know, there’s also some upgrade to that, besides just SEO, which, you know, in itself does bring in money if it’s done right. So yeah, it’s all good.

Christopher Penn 36:08
Think about this, John has had some of the biggest names in marketing, the most successful people in marketing, on marketing over coffee, John has a training library to make blog posts that sound exactly like Seth Godin, or David Meerman, Scott, or you know, the name of person of your choice, because you have their language, you have the uniqueness of the patterns of the way they speak. And you can distill that out, as long as you have a good enough prompt, you can distill that out with these tools, and then reuse that. So you might, you might not want to write exactly like you know, another person, but you could certainly coach the tool to say, produce content in this style. And if you know that you’re going to be you have an opportunity to present that content to that person’s audience, maybe as a guest blog post or something. What better way to do that than to give the audience the information you want to share in a tone and in a style and a voice they are accustomed to.

John Wall 37:02
Yeah, match it as it goes. Now. Coming soon, our virtual conference with all the biggest names.

Katie Robbert 37:08
It’s, it’s funny, Chris, that you went that route, because my first thought was, you know, John, you get pitched for guests all the time. And sometimes you’re not quite sure if the topic would be a fit. And so I can almost imagine you training using the content as a training library, and then spinning up some kind of a chat bot to help vet potential guests where they put in their topic and say, This is who I am, this is the topic, and then the chat bot will will come back with additional questions and help you screen in or screen out potential guests sort of taking that off of your plate to have to do the dirty work to say, why would you pitch me this topic? It’s definitely not a fit, like, let the Chatbot do it.

John Wall 37:49
Oh, man, yeah, I’m totally on board with that. Because I can’t do flag words. You know, it can’t be that basic. I can’t just be like, Oh, if you’re a branding guru, you know, that goes in the trash bin, because that still gets used for some great people. But yeah, to be able to have AI be like, oh, yeah, this is a brand new guru. And, you know, formally franchise chicken shacks. You know, this is obviously not somebody that is going to have a story we want to keep So yeah, that’s a really interesting idea of being able to kind of filter out the other automated generated stuff that’s gonna clog up your box, if you don’t come up with a better way to handle it.

Katie Robbert 38:26
Just let the bots fight it out. And then they’ll come back to you with an answer.

John Wall 38:30
Yeah, I just want the three good ones, I could care less about the rest of the fighting.

Christopher Penn 38:34
Yep. Yeah, these tools are capable of repurposing content in all sorts of different ways, including ways that probably no one has ever given any actual thought to one thing I was playing around with earlier today, which I meant to tell you too, about, I forgot. It is taking the LinkedIn profiles of the most valuable customers that we worked with, right, the people that we love working with our clients, and extracting the LinkedIn profiles, putting them in a text document and feeding that to Claude and saying, You are a sales consultant. Your first task is to review all these different LinkedIn profiles. Here’s the we think our typical buyer is helped me understand the buying profile, like what do all these LinkedIn profiles have in common? Because I don’t know. I mean, I can make some guesses. But by having all that knowledge, their backgrounds and education, you know, what they’ve posted about? That’s super valuable for finding connections in texts that we can’t because we just can’t hold that much information in our brains all at once. When you have a 100,000 token context window. 66,000 words, you could put 500 LinkedIn profiles in there and develop a buying profile from it.

Katie Robbert 39:47
You don’t have to give away the answer live, but did it give you an answer? Just a yes or no?

Christopher Penn 39:52
It did give me an answer. And it did not. The answer did not surprise me. That’s fair. Fair. If you look at our client roster stir it says, Hey, this is, so your client roster probably is. But then you will use that to do alignment checks. And this is something that I use. If you are not subscribed to our newsletter, go to trust Today, we sent out a promotion for Google Search Console for a Google Search Console course, I wrote the first draft of that copy. And I was like, okay, cool. And then I wrote out the information, the prompt from this LinkedIn exercise. And I said, I want you to review this sales pitch email against our buyer profile, and make suggestions about what you think I could have done better in the copy, or the sales pitch. And it it changed the text what you saw Katie, when before we sent it was after AI, and it was so much better than what I had written because I was going from my perspective, like, Hey, this is a cool thing. This is all this and the ad is like, no, your your prospects will hear about this?

Katie Robbert 40:55
Well, you know, it’s interesting that you say that exact phrase that it was, from your perspective, that I think that’s the trouble that marketers in general have is not writing it from the user’s perspective, we write it from our perspective naturally, and maybe don’t realize that we we think we’re speaking to the audience, we think we’re telling them what they need to know. But we’re still introducing our own set of biases into the content. So I think that is a really interesting use case.

Christopher Penn 41:26
Here I can I can show you just a little bit from ChatGPT. I said here, this was the original pitch, right? And then it went through and it said, Hey, you got to address some pain points. replace this with this, replace this with this, you know, and so this is this click copy that eventually made it into the final it said, it said instead of just saying, hey, it’s got 12 lessons and said no comprehensive insights and 12 lessons, two hours of focused instruction, a certificate to validate your newfound knowledge, instead of just hey, get a certificate. Like it changed the language. I didn’t even think about writing for example, the closed caption says it we foster inclusive learning with with accommodations, and it says the and more which is your typical sales letters and the confidence to make strategic decisions at the age of AI. Right. So because I gave it a good buyer profile. This is content repurposing, do, I gave it to that content that we had from LinkedIn profiles, turned it to a buyer profile, turn that into a sales pitch alignment tool, and now I have a better pitch that went out. Now, of course, the test will be if anyone actually buys it. At least at least we use the tools to repurpose content and refocus content to the way that the audience hopefully is more receptive receptive to then than what’s coming out of my brain.

Katie Robbert 42:45
Makes sense? Totally makes sense. So to summarize, if you want to use artificial intelligence to repurpose your content for content repurposing purposes, it’s absolutely possible to repurpose with purpose. So basically, there’s a few different ways to approach it, you can build your own process using code and open source tools, that’s a lower cost version, but it increases the amount of dollars spent doing the thing, versus other tools that can sort of do it Start to finish for you. The tools themselves will cost more money. But the labor involved aside from a sort of a learning curve and training will be less overall. Or the third option is to bring on a company like Trust Insights to help you set up that process to help you evaluate what you have, what content, you can repurpose, and build out the process help you select Tools, or even get you 90 or 100% of the way there. And then all you have to do is just do something with the content. So if you want to reach out to us for one of those options, you can find us at trust And you will be paired up immediately with our chief statistician, Mr. John Wall,

John Wall 44:02
right here, just follow the line.

Katie Robbert 44:10
And I think on that note,

Christopher Penn 44:10
on that note, thanks for tuning in, folks, and 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 the Trust Insights podcast at trust AI podcasts, and a weekly email newsletter at trust Got questions about what you saw in today’s episode. Join our free analytics for markers slack group at trust for marketers See you next time.

Need help with your marketing AI and analytics?

You might also enjoy:

Get unique data, analysis, and perspectives on analytics, insights, machine learning, marketing, and AI in the weekly Trust Insights newsletter, INBOX INSIGHTS. Subscribe now for free; new issues every Wednesday!

Click here to subscribe now »

Want to learn more about data, analytics, and insights? Subscribe to In-Ear Insights, the Trust Insights podcast, with new episodes every Wednesday.

One thought on “So What? Whisper and Claude for content repurposing

Leave a Reply

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

Pin It on Pinterest

Share This