<img height="1" width="1" style="display:none" src="https://www.facebook.com/tr?id=2006193252832260&amp;ev=PageView&amp;noscript=1">

30 Min Read

[The Marketing AI Show: Episode 3] Can AI Create Content?

Featured Image

Wondering how to get started with AI? Take our on-demand Piloting AI for Marketers Series.

Learn More

The Marketing AI Show—the podcast that helps businesses grow smarter by making artificial intelligence approachable and actionable—has officially dropped!

You can now listen to the first four episodes of the podcast on your favorite podcast app. Keep reading for more on what to expect in episode three.

Episode 3: Jeff Coyle, Co-Founder & Chief Strategy Officer, MarketMuse

Jeff is a data-driven search engine marketing executive with more than 18 years of experience in the search industry managing products and website networks. He is the co-founder and chief strategy officer for MarketMuse, where he is focused on helping content marketers, search engine marketers, agencies and e-commerce managers build topical authority, improve content quality and turn semantic research into actionable insights.

Before joining as co-founder at MarketMuse, Coyle owned and operated his own inbound marketing consultancy and managed the Traffic, Search and Engagement team for TechTarget, a leader in B2B technology publishing and lead generation.

In this episode, Paul and Jeff address this burning question that drove our show host, Paul Roetzer, to start exploring AI years ago: Can AI create content? Back in 2015, the answer was generally no. But, today the answer is yes. In this interview, Paul and Jeff discuss:

  • How AI can level up your content marketing. 
  • How major advancements in natural language generation (NLG) are transforming the field. 
  • Plus, major MarketMuse announcements for 2021. 


[Video] Watch the Full Interview 


Read the Full Transcript

Disclaimer: This transcription was written by AI, thanks to Descript

Paul Roetzer: Welcome to The Marketing AI Show. I'm joined today by Jeff Coyle, the co-founder and chief strategy officer, which is a new title we'll have to get into in a moment of MarketMuse. Welcome Jeff.

[00:00:13] Jeff Coyle: Thanks, Paul. It's great to be here. I'm looking forward to the discussion.

[00:00:17] Paul Roetzer: Yeah. Chief strategy officer, you were up until I looked on LinkedIn this morning, the chief product officer. When did that change happen?

[00:00:24] Jeff Coyle: On this recent Thursday?

[00:00:27] Paul Roetzer : Is that right?

[00:00:28] Jeff Coyle: Made the Change. So you are, I think you got it the day after. Really focused on kind of innovation and new business initiatives, with the  new role.

[00:00:39] Kind of previously I was managing the product data science and engineering and marketing side of the business. But with the promotion of Chuck Frydenborg to CEO, and my co-founder Aki below is focused on, financing and also partnerships. And I'm taking a bit of a [00:01:00] verizon scanning new business initiatives focus and it's, it's extremely exciting.

[00:01:04] We have some major announcements coming in 2021 that I'll get into a few of today. But not all of them. And those will be my main areas of focus. So very exciting, very exciting.

[00:01:17] Paul Roetzer: Oh, congrats. I know I have been seeing the changes happening from an org structure standpoint. So I will get into asking you a bunch of questions about that today, but we'll definitely have to talk about that afterwards, but why don't you tell us what MarketMuse does, like tell us a little bit about the company, the, the solutions you guys offer?

[00:01:33] Jeff Coyle: Yeah, sure. So we are, and as I, as I wouldn't be here, if we weren't, we're an AI enabled content intelligence solution business, and we focus on content intelligence in that we want to figure out what it means to be about things using natural language processing, using artificial intelligence platform.

[00:01:53]And we apply that to page level analysis to site level analysis. To inform content plan. So what content should I create? What content should I update to have the biggest impact on my business? And the use one unique differentiator for MarketMuse is we take it all the way down to implementation.

[00:02:10] So it's not just about giving you an idea. We'll walk you through a single source of truth of a content brief. We'll start as we're going to talk about today, building content drafts for you and take it as far as you want to go. And so. Well, we're really focused on is making sure that people have, you know, data-driven insights that guide their content. But also they're able to justify the investments they make in content. And I think that those are the two main areas of opportunity for, you know, for content marketing, especially with regards to artificial intelligence.

[00:02:44] Paul Roetzer: And what's your origin story in terms of AI? I mean, what, you know, career path, how did you end up co-founding a company powered by AI,?

[00:02:55] Jeff Coyle: The short version of it. My background is in computer science and the two main [00:03:00] areas that I researched when I was in school. I'm a quite a long time ago back when, you know, search engine meant internal enterprise search engine. But it was kind of. And enterprise search as well as usability theory.

[00:03:14]So I've always worked in the search engine space of whether it be building ad servers and search engine products for lead gen or in search engine optimization and content strategy. And after going through an acquisition of my business from which I really originally worked from 2007.

[00:03:32]I worked for quite a bit of time with a team that had a very large content organization, wonderful excellent editorial team and group of writers. And throughout that experience really learned every possible workflow for trying to use data, to inform those content decisions. And late in that run at that company at that publisher.

[00:03:56] I had all these documented painful processes. [00:04:00] And I started to research people who were really innovating and that's the language processing. And I came across my now co-founder of five years who had built the original technology for MarketMuse. And it was really focused on evaluating a concept and saying if I were an expert, How would I cover this specific topic?

[00:04:23] And so the original technology was, you know, that was the market as the core.  And I brought to the table a lot of these workflows and we had scientists and you know, developers who could automate them with an emerging technology and artificial intelligence. So it was through having the experience and, you know, content strategy planning and, and search engine optimization and an understanding of how those search engines work, that I was able to inform the teams to automate these really painful processes. And that's really been the journey for five [00:05:00] years. It's, you know, I had the background in the technology but these emerging concepts, these emerging spaces just are, you know, they're multiplying by, you know, 10, 20 times the innovation every year. And so you're just playing, you're trying to stay, you know, current constantly with anything related to, natural language. So. Yeah.

[00:05:21] Paul Roetzer: And you guys have been on my radar since the early days. So I, you know, I think I've shared with you before my path for down AI started in 2011 when IBM Watson won on jeopardy. And I didn't know what AI was, I did not have acute computer science background. Like you, I was a journalist. That was my training was in the journalism school. And so not too long after that, a few years later, I actually attended SXSW and there was a panel with the Associated Press and automated insights, and they told the story of how the associated press was using what they were calling natural language generation, which is in a form to create hundreds of earnings [00:06:00] reports every quarter and turned into thousands using this tool. And I was like, Oh my gosh.

[00:06:04] Like I literally sat in that session. Like, can AI create content like, can it write blog posts? And I think that was 2015 would have been March of 2015, probably right around when you guys were creating MarketMuse. And that was my, like, I had already been studying AI for a few years, but that became the driver to me.

[00:06:22] It's like, well, that's what we do. We figure out what to create for clients to help them grow their business. And then we create it. Can is AI going to take that from us? So why don't you walk me back to like 2015, were you guys thinking about that same fundamental use case of like, can AI create content?

[00:06:42] Jeff Coyle:  That's really an awesome question. The  vision for generation probably didn't happen. Their vision. That generation could be a reality in the type of content that we were looking to influence and the types of updates that we were looking at influence probably didn't happen until a year after that.

[00:06:59] Okay. But the original in that 2015 timeframe the, the inspiration was turning data-driven insights into a format that a journalist or an editor editor, or a writer or a subject matter expert would accept. And they would be excited about receiving. And that is when the content brief was formed. So the original content briefs we built were manual.

[00:07:32] We actually took data from multiple MarketMuse, proprietary data sources, and we, you know, compile, they, you know, content brief and we deliver its customers by hand. Now all of that is automated and scale. But that was the first mantra into building something that. Inspires a writer to, you know, be able to put their best foot forward and take their expertise and amplify it and not bog them down and things like keyword research [00:08:00] and not bogging them down and things like, you know, hitting publish and hoping something does well, which is very common still to this day and not an understanding of, of content efficiencies.

[00:08:10]But not until the, I would say for me personally it was hearing at the Intelligent Content Conference hearing about, I think it was 2016. It could have been late 16 or early 17. It was hearing about Heliograph in person.

[00:08:29] I

[00:08:30] Paul Roetzer: actually, I moderated that panel. That was crazy. Yeah.

[00:08:34] Jeff Coyle: So I had read about it and I read about it. And what was for me inspiring was when they talked about covering the Olympics. And they had said prior to Heliograph, we were only able to cover a small percentage of the events. But after Heliograph, we cover all of them. And the one thing I remember was the there's some Iraq or Iran or error in [00:09:00] the way that the Olympic events are set up and he explicitly calls out the templates and how they have to.

[00:09:05] Like put exceptions and rule-based exceptions in the templates. And one of them was you did, you know, judo had two bronze medals because of the way the event set up. If I remember that specifically, and I remember going on to my laptop immediately after leaving that session and researching judo rules and awards in market views, and it immediately told me that the two bronze medals.

[00:09:32] Would be a special differentiating factor. And I was like, Oh, wait a second. Okay. I can do something here too. That's fast. So we could, we would build the brief and the brief would naturally have surface that that was a point of differentiation. And why it's something, it's a point of differentiation in the topic model.

[00:09:50] Yeah, but that got me thinking that, wow, if we can produce those same, they didn't have that value. They had to manually set that rule. [00:10:00] We don't. We had already automated things. So it told me that there was space there. So we started working and I can get into some of the details of the progression, but we started researching summarization.

[00:10:12] We started researching abstractive summarization. We started researching other types of, kind of technology that I was in its infancy. And we actually took two or three swings at that before thinking, wow, things are too expensive. The technology is just not that and that, and we spent, we took some cycles against that our head of data science and, and now chief technology officer and then our current head of data science.

[00:10:39]And we have a number of scientists still working on these challenges said, you know, these things, we gotta put them on the shelf for a few months and see what happens. And then You know, about a year later through two of our scientific advisors and the research that that same team was doing, we found that the ideas that we had proposed for the, this specific platform, which I know we'll talk about were possible, and not only possible, but affordable.

[00:11:11] To invest in and we projected what it would be. And immediately got a return in value. We're like, Whoa, in a year, and we're not using, by the way, we're not using other people's models, we're doing it all ourselves. And we always said, we're going to do this as much of this as we can ourselves. Because that's really the only way we can, you know justify the, the innovation and the investment.

[00:11:35]And so really that inspiration for it came from those templates and rule-based exception-based and algae that we didn't want to build, because we knew if we could build open long form content with natural language generation, we could always back ourselves up. We could always back pedal and do those templates, scalable solutions too, but [00:12:00] we needed a free hand to do it. Our, my original co-founder one of our original founders was a research scientist who had been innovating in the field as well. So I would be remissed to not mention him. And he had given us some of the early inspiration to consider this too. And he had done.

[00:12:21] Some of the original science on this and we took it and, and, and really made it work. But funny thing is we make it work because of the original technology, the core, if we didn't have the, the automation of the content brief, we couldn't do explicitly what we're doing with first graph. Okay.

[00:12:39] Paul Roetzer: So let's, if we go back for the listeners or viewers, go back to that spring of 2015, when I was asking the question, can AI generate content? The answer and correct me if I'm wrong at the time was yes, it can be trained on structured data to create a narrative. Right. But can it start freeform from an idea and generate something? The answer was no, [00:13:00] not even close,

[00:13:00] Jeff Coyle: not anything worth reading. Yeah. I mean, the models were so they degraded so badly.

[00:13:07]They were, you know, just not anything that could be applied spurt and certainly not applied for marketers. It was very template driven. You had a few companies in the SEO space who were innovating in this area using, you know, randomizer hours and databases to create kind of varying content.

[00:13:32] And then you had companies like the one you mentioned with Automated Insights and Narrative Science. Starting to look at possible applications in places like finance where the content isn't very highly customized over large data sets, almost like product. We are basically in content products, not generating content.

[00:13:54] Paul Roetzer: So. Again, go back to the original question. Can AI create content? Can I write my blog post? Can it write articles? [00:14:00] The answer was no, but I think we can agree today. The answer is, yes in some forms it can create content from an idea from a topic what's changed. What has enabled us to go from that five years ago, where we might've looked out and said, I don't see it happening.

[00:14:19] How have we today arrived at the point where marketers need to seriously be looking at using natural language generation, again, an application of AI to generate content?

[00:14:30] Jeff Coyle:  Yeah. I mean, so much, so much has changed in the, just from a technology space, the ability to build precisely defined models that can generate content that progression just in the last year.

[00:14:52] So there was probably two or three years were good was possible.  And now great is possible and [00:15:00] there's a couple of different ways that we can make it even more you to find is great.

[00:15:04] Is it that if I read it, I won't know a machine wrote it. Is that great? Well, there's two, that's a great question. I think there's, there's two schools of thought on that.

[00:15:14]What my inspiration is as a co-founder of, of market muse is ridding the world of low-quality content. Right? I don't want there to be a market for someone to create content that's low quality. I want the bar to be set by the technology that it's better than that low quality content. I want to be able to build high quality content.

[00:15:38] The challenge is that this worked for. For marketers is it's not mature. Right? So we are not used to getting content from a machine. We read it and we say, no, this isn't what I thought it would be. Okay. We always judge it. Where we're having a lot of success [00:16:00] is where are, you're getting some repetitions with this and you're saying, Oh, wow.

[00:16:04] I see how going from concept to content brief that I've validated, by the way, I want to build this content. We have this structure that is, this is the article I would like to build. Right. And so I'm validating that I have the vision in my mind of what this article could be. And then receiving a first version of that.

[00:16:26] If I have those few steps, I'm less likely to immediately judge it and go, wait a second. There's something that isn't right. I'm looking at it as a, as a tool I'm looking at as part of my workforce. So the actual, immediate to publishing of this article may never get there for me, for me, you know, in the short run, because I'm a marketer and I need to put my own finishing touches on it.

[00:16:52] I need to improve the production value. But what I see it as. It is as yet another solution that [00:17:00] accelerates my ability to achieve my goal. So just like Washington post went from 10% of the events, Olympic events to publishing a hundred percent of the Olympic events. Well, if it were just up to me or you Paul, and you want to write an article about NLG this month, You might only be able to knock out one really good one.

[00:17:19] Right. But you know, you should probably write about 20. If you're using MarketMuse, it's going to tell you the other topics you to be covering, right. To be able to build that cluster of content. Well, it may be the situation where you can focus on the one as a creation effort, but act as an editor on those others wanting.

[00:17:37] And knock out that entire cluster this month. And so that's the type of workflow that I think is possible today. Why? I don't think I know it because I'm probably that customers that that is happening for, but the going straight to publish, I think for marketers is always going. That is going to be a maturity challenge throughout.

[00:17:57] And it's going to be one of these things where people are going to have to [00:18:00] get over the judgment and recognize this as a, yet another source of data that they can craft and use, if it's going to be a bit for them, I think editorial teams are going to struggle as well. In the same way. It's because it's a, it is a shock.

[00:18:19] It's like getting hit in the face with a glass of water. When you read your first generated article that doesn't really sound terrible. Sounds pretty close. Like I would have done this differently. I would have added a section about this. I would have touched up. I would probably, ah, this seems a little like it's a little fluff, right?

[00:18:37] A little fluffy. That's the kind of feedback that I get constantly when I'm having these

[00:18:42] Paul Roetzer: and he's going to have to give it to an intern or to the like entry level writer. Like it's. Yeah. Yeah. If you don't think it was the, again, the name and we'll talk a little more of a specific offers, first draft, but like it's not finished product it's it's meant to be.

[00:18:55] So, I mean, for, for people listening who think this whole idea is abstract, that a machine could [00:19:00] actually write a narrative, a thousand word narrative. What are some tools that people may be using right now that they don't even realize are being powered by this kind of underlying technology they're helping with their editing or actually composing pieces of what they write.

[00:19:14] What are some popular tools that come to mind for you?

[00:19:17] Oh, gosh,

[00:19:17] Jeff Coyle:  there's, you know, there's so many that, but you know, just, you know, Google Smart Composer is a great example. Everyone is probably using it in your Gmail, or if you use it as your Google admin, it's part of that experience. It's finishing sentences, you know, you're Grammarly there's computer vision solutions like Microsoft Azure, there's computer vision. There's summarizers in browser extensions like S and AI Hemingway is another popular one for grammar checking. But there's also things you're reading, you know, there's a great percentage of product descriptions that are being built now on [00:20:00] e-commerce sites even on you know, popular ones that you probably are getting things shipped for the holiday season, and those descriptions are being written out of databases. Most of the  financial sites you read  are having consistent posts that are coming from some sort of hybrid of templatizing and AI. So you're reading articles and not realizing it, or you're really reading it and going, Oh gosh, well, this is useful information. I can kind of tell it's not written beautifully by an artist. Right.  But all of those things are, are active and it's, it's, it's only becoming more More prominent in your day-to-day work.

[00:20:42] You know, the question I always ask is like, how many software solutions are you interacting with each day? How many of them are processing text or processing some information right now? It's almost all of them, almost all technology that you [00:21:00] commonly interact with. Is analyzing a data with natural language generation or natural language processing.

[00:21:06]And now what we're seeing is there's, there's more use cases popping up and you know, in our case, it's we want you to be able to put your best foot forward. We want you to never write content that isn't as good as, or better than your competitors every time. And this is yet another way we can allow you to publish everything to tell the story about your business or your blog or otherwise.

[00:21:32] And I think that that's a really special use case you know, you look at like a narrative science or an automated insights in there, you know, it was, Hey, let's get your data out of databases and onto, onto the web and narrative with a very different goal. And I think that we will back pedal into that use case as kind of a hybrid of editorial text.

[00:21:55] In addition to what we're doing. And we already know we can do it. We just took a [00:22:00] different approach to the market because we think this is the this is the market breaker for natural language processing. It's being able to write content and turn it to write like Paul. And that's, that's what we can do.

[00:22:14] Paul Roetzer: I took your state of natural language generation course that you created for our AI Academy for Marketers. And thank you. It's fantastic. In that course, you talked a number of times about an innovation called Grover. And you talked about how important, like a milestone that was and what Grover enabled.

[00:22:32] Can you explain to us, like what, what is Grover and how did maybe that play into the thing people may have heard more about, which is GPT-3. So maybe like we'll, we'll get into how you guys are applying similar tech, but the couple of things in the last, even 19, or I guess since early 2019, that you identified as some really important advances. And when you did that course, GPT-3 hadn't [00:23:00] happened yet. It was an actual GPA. Yeah. So why don't you just explain to us just a couple of those advancements, and then I want to talk a little bit about what you guys are doing and where this may lead us as an industry.

[00:23:11] Jeff Coyle: Yeah, sure. So Grover school if you, if you look for it, it'll be under Rowan  Zeller's site.

[00:23:18] He was one of the innovators and the goal was effectively. We have a major societal problem in fake news. We can. Yeah. Have you heard that? There's a, and there's also the potential for it, right?

[00:23:34] Paul Roetzer: So the propaganda, so, so fake news doesn't necessarily mean one like one side of the political spectrum.

[00:23:39] It's no, we're talking about actual propaganda of like things that are created out of nothing that never happened in our real life.

[00:23:46] Jeff Coyle:  Yeah. This is not politically driven. This is literally I'm telling somebody, writing machine written content that has no basis in any data has, has no training of any kind is it's [00:24:00] just, you know, mimicking the style of a corporate document. So it has nothing to do with politics.

[00:24:08]And so the the, the cool innovation there was, you could actually train a model that mimics a writing style. Can I and so they had created, and then they had a few, you know, tools that you could test it out and like start writing a paragraph or give it a topic. And it would just shoot out a random, you know, blob of text.

[00:24:34] And this was, this was interesting because, but this, the goal of that team was to develop a strategy, to respond, to attacks or instances where that type of thing might permeate again, you know, in society. I think if you're a brand, too, people can create fake things about your company or your leaders or whatever it may be.

[00:24:59] Paul Roetzer: So this is [00:25:00] very relevant. The idea that someone could in theory create completely fake stuff about you and do it. At an infinite scale, almost like just create as much as I want about this brand is, and the ability to identify that and be able to react like this is critical stuff that most brands don't even know as possible, even watching on not only that is other people are summarizing even Google in some of their more recent implementations are reading texts and trying to DeepMind for example is able to.

[00:25:33]Jeff Coyle: If you look at scoring models in that specific component of what they're doing, it's, you're looking at snippets. And you're trying to identify the things that signal that this page is extraordinary, extraordinary. And then you're seeing answers appear in search results and you're seeing right answers, but you're also seeing wrong answers.

[00:25:55]You know, the, the new joke is the the SERP dirt. It's where you actually [00:26:00] are right in the question. And the answer is so wildly wrong, right? And it's because there's a lot of content out there that is being digested differently than one would intend. So there's like two stages of this and Grover was just a cool example like that this can happen.

[00:26:17]It's much more like the implementation of the tool was to show it and like, wow. People it's, you know, much more of a Of an exciting kind of observation that it was possible. And yeah, basically putting yourself in the situation where the system, you know, for, for, for building content is possible.

[00:26:39] And then, you know what, but what is the ideal model though? It's that you can generate texts that actually is trained for writing style. It meets the specifications that you want. And it retains its coherency throughout the document. That wasn't what that was focused on. It was really focused on [00:27:00] identifying whether something was machine-driven or not.

[00:27:03] Paul Roetzer: Gotcha.

[00:27:03] Jeff Coyle: And and now with the innovations that we'll discuss, you know, we we've, we've gotten closer to part two, so we can actually identify, you can actually identify still whether something was or whether something wasn't wasn't. Less and less over time.

[00:27:20] Paul Roetzer: So, and then I, we could spend an entire episode just talking

[00:27:23] Jeff Coyle: about GPT-3 and compare,

[00:27:25] Paul Roetzer: but give a kind of a, you know, a 32nd overview, like what is GPT three and why might it matter to our audience?

[00:27:32]Jeff Coyle: It is, there's, there's so much too that you can read online that I will, you know, we'll put it in the show notes too. We can have some of the resources in there, get some details. Well, the way to really think about it is that there are massive it's that massive scale? It's a general purpose language model.

[00:27:56] It can be, it's kind of like the Play-Doh it's it's [00:28:00] unformed for a specific use case. But for marketers, it lacks structure and provides pretty shallow topical coverage with its standard output of something that you can train. It's really not thinking about your workflows yet. It's a,it's a date.

[00:28:19] It's a, it's a source. It's a general purpose thing, but it is magnificent. I mean, these just the scale is almost difficult to think about. And, and that's really what to think about from a From a content creation. It has infinite use cases that people right now are just figuring out box customer service.

[00:28:40] Creation of original pieces. Like it's exactly. The, you know, so what, what has to happen is someone needs to take either build their own technology or start to think about ways to make it not suffer from what it suffers from out of the box. Okay. Not just crossed which you know, it is licensed exclusively. [00:29:00] I hope by Open AI through Microsoft. Right?

[00:29:02] Paul Roetzer: So you can't go by, like, if you're listening to this, you can't go search like GPT-3, just buy it and start building things. That's not how it works.

[00:29:09] Jeff Coyle:  Licensing it through Open AI as an exclusive agreement for through Microsoft. So you can buy credits and use it.

[00:29:17]But you're not going to be able to like, go in and say, Hey, write, go wright. Wright my blog posts for me a lot. But you know, even when you are training it to build content, you're going to get chunks of like, if you probably read the article on Guardian, how that was written, I think having that was a match up of paragraphs from 10 generations. So they picked and choose and they folded it in to make it read like the author wanted. So if you're, if you're just asking it to go run free, it's not [00:30:00] going to come out that way right now, right?

[00:30:02] Yeah. So it suffers from degradation. It's still suffers from, you know, repetition. It's not doing things like checking plagiarism. It's certainly not doing a deep dive and learning about a topic. Right. And so that's where this isn't specifically tuned for marketers. There are it, it has the potential to be, and the next versions of it, you know, we'll get closer and closer to that, but will it ever the solve the kinds of problems that marketers have sports specific use cases, just keep an eye on the solutions being built with the models. With the, with the technology. I saw one that was writing  very, very short form product descriptions you know, two sentence product descriptions that look pretty deadly. I don't have those yet.

[00:30:49] That's very sexy. But I haven't seen anything to compare with what we've been building. Yeah.

[00:30:56] Paul Roetzer: So tell us about first draft. Like, what is that, what is what you guys have been building [00:31:00] and how is it maybe a little different than?

[00:31:02] Jeff Coyle: Yeah, no, for sure. For sure. So, you know, not to get too much into the. Exhaustive technology of it, but we w we, we take the model of our goal is to, obviously we don't want the poor quality content to exist.

[00:31:16] We're also looking to give people objective ways of measuring the quality and comprehensiveness of content. That's always been in the model. That's our technology. We can actually read an article and tell you how comprehensive it is on a particular topic. We can tell you whether it was written by an expert.

[00:31:32] So we wanted to use that innovation and that technological advancement that we had and also our content briefs, which basically are the outline and tell the story for a single source of truth for a writer. So we wanted to take all of those things to be able to build content and build that, you know, those specific brands.

[00:31:55]And so what we do, you know, in a high level is we're. Able [00:32:00] to build a base model that is  the person, the model of the world. That's the wisest the topic you care about. Yeah. Okay. It is brilliant. It knows everything about this thing. How do you get bees out of your garage? I mean, it can be something really specific or it can be something like AI.

[00:32:20]And then we want to train it against. That you know, that deep, the details that you care about the personalized stuff. So how you wrote all your other articles about this, how you write generally, how all of your authors write as a collective hive, you know, it can be done on any level and we're able to generate the content that hits all those specifications and would be quality would qualify against the content brief. And when you, when we way that we've done that it, you know, does the unique, some of the unique aspects are [00:33:00] that it has memory modeling. I can't get into all the details, but it has memory modeling so that it stays in, it stays in the moment, right? It doesn't lose track of what I say.

[00:33:12] Paul Roetzer: AI doesn't have memory per se, like the humans do it. 

[00:33:16] Jeff Coyle: and so we have con we understand context. We understand the investment, we understand what it means to be an expert. We understand comprehensiveness. We don't lose our place in line as we're going, we're not repeating. And so we've built this technology to write the article that you ordered and now.

[00:33:36] We keep innovating that and it keeps going. We keep focusing on making those better. That first draft, the goal is to have a higher hit rate for someone looking at it and immediately getting over that first tendency to judge and, and transitioning to Whoa, this sped me up. It's not perfect, but [00:34:00] I might be able to get this out.

[00:34:02] In an hour or two. Right. And wow. What, and I didn't have to do anything, but look at the brief, maybe build the brief, depending on what, what, you know, marketing needs plan I'm working on under I build the brief order. It, yeah, this is the one I wanted. Shoot me the draft. It appears. And then we're creating novel user interfaces to work at those drafts to emulate the way we've.

[00:34:28] You've worked with, you know, on user research of how people work with this type of thing. So what you'll see in January when this goes into production is you can actually bring the draft into a text editor and tomb in multiple formats. You can actually bring it in on the side and drag pieces of it.

[00:34:48]And you can take sections. You can reject sections, you can do all kinds of different things, too. Speed up that process of building something you're [00:35:00] proud of, which is the goal. Yeah. And I think that that's a very different goal than what people are naturally thinking with NLG. And it's, but it's always been what we've.

[00:35:13]Strive for, you know, I want there, I want you, I want Paul, I want you to put, be able to put out more content that you're proud of with the same resources or less and that, you know, that we are so close to that. And it's going to happen in, you know, In the first half of next year. And, and it's, it's, it's exciting. It's really exciting.

[00:35:38] Paul Roetzer :  To kind of wrap on this section. 12 months from now, 24 months from now, I mean, 2021, 2022. What does, how is content marketing different how's content creation, different for the industry? And, you know, give me like a minute or two and just how you see, because the advances are moving so fast. That is hard. It's hard for us. And I mean, you're in it more than even me, [00:36:00] but I mean, sometimes I will step back and be like, I can't, I can't believe that is, it was February of 2019. Then we had like an advancement in the fall. And then all of a sudden GPT-3  it's like. Man, this fundamental question I asked five years ago about can AI create content?

[00:36:14] It's just, I'm having trouble keeping up. So what does this mean? Like what does it mean to marketers in the industry of where we're going to go with content creation and language center?

[00:36:24] Jeff Coyle: Well, I think the first thing you'll see, which has already started to happen is  the market for low-quality human written content disappears. Okay. And there's a lot of that. And so the, the, the industries, the content farms, the outsourced, you know, the, the non-native translation spinning services go, and all those groups are already having to change their business models and actually focused on content and managed services, right?

[00:36:53] Paul Roetzer: That's the race to the bottom of 3 cents a word is done. I don't have any complaints from me. It's fine.

[00:37:00] Jeff Coyle:  I mean, it is gone. I mean, it, it, it will be gone in, in 12 months because it's not, it's not effective. First of all. The other hybrid to that concept is those same resources will have technology that lets them right at a higher level, but the outcome is.

[00:37:17] The bar for quality content quality goes up. That is the first prediction, the second, the second one is a little bit controversial.  That I'll say is the various search engines and directories that exist are really going to have to change the way they think about what they do.

[00:37:39]And the controversial part of it is that many, a lot of consolidation is happening already. And the, you know, one entity doesn't own one site, you know, one entity owns 50 sites, owns a hundred sites, owns many brands. And when those businesses [00:38:00] are content powerhouses, editorially, and they've already produced, high-quality comprehensive content.

[00:38:07] When those businesses become enabled with this type of technology, it could become the situation where when somebody is researching a particular topic, they have, you know, 20, 30, 50 entities on the web and they can monopolize a finding a findability, you know, Workflow finding workflow. So I think that what, what is going to have to happen in it probably won't be a year, but it'll be two years or three years is the figuring out the real estate dynamics and true ownership will be a top priority.  And it's already starting to happen. There is [00:39:00] I watch the search results aggressively at scale.  and it's not, it's no longer about who's winning. Who's losing. It's not, that is not what people should be watching right now, but people shouldn't be watching right now is acquisitions.

[00:39:18]And.  Ownership amongst entities as well as partnerships. That's where the story is going to be in tune. It's how, no matter what you do, if you're looking for an X, your money goes to Y entity and that's, that's a problem. Yeah. It's controversially. That's a whole, that's a whole other episode. That's why that's something that natural language generation brings front and center. Yeah. And I think it's something that should have, should be already being dealt with. But first natural generation brings that to the top of the priority list for the biggest businesses in the [00:40:00] world right now.

[00:40:00] Paul Roetzer: Wow. Well, we're going to wrap up this episode like we always do with our rapid fire questions for Jeff

[00:40:06]You ready for rapid fire again? First one voice assistant. You use the most Alexa, Google Assistant, Siri, Cortana, or I don't use them

[00:40:18] Jeff Coyle:  very rarely when playing with the kids, Alexa. Okay. I turned the lights off in my house. We haven't turned off except yeah.

[00:40:26] Paul Roetzer:  When, when it is to Apple to that, I, I use our Apple home pod more. Oh, well I guess technically that's Surrey something. Alright. More valuable in 10 years now. I don't know if you have like bias because you have one, but computer science degree or a liberal arts degree. Oh, a computer science degree. Yeah. Ironically, like we we've asked that question of like some expert in specs or spotlight.

[00:40:47] It's like 50, 50. And Mark Cuban is on record numerous times saying he thinks liberal arts degree.

[00:40:53] Jeff Coyle: I think that computer science degrees enable you to think abstractly. I think that the, [00:41:00] the market is with, for kids. It's so ubiquitous that they can control computers and phones and devices, but they're, they're not thinking about computer science.

[00:41:09] And when they get the ones that are really enabled and then they actually learn computer science, their mission, their mission, I'm probably with you in that camp. All right. No net effect over the next decade. More jobs eliminated by AI, more jobs created by AI, or it's not really going to have a meaningful impact on our economy.

[00:41:29] Oh, great. I'll just in marketing. Big, big picture. Creative. Absolutely. Yeah. Creative. Yeah.

[00:41:35] Paul Roetzer: What does an AI agent win first or at least share with a human Nobel Peace Prize, Oscar Pulitzer, or it's not going to win anything. Be a writer. It's going to share it with a writer. It's going to be an acting it's going to be,

[00:41:48] Jeff Coyle: I would say. Because there's so many Oscars and obscure fields. I think it will be an Oscar like item for [00:42:00] something like voiceover excellence or something like that. Yeah. Some random category. What are the technical Oscars? But they showed like the day before though, that'll be,

[00:42:12] Paul Roetzer: I saw a great article the other day making a very strong argument that a Deep Mind, Alpha Fold could be the first Nobel Peace Prize that is partially given to. How machine and it makes, it made a ton of sense, like the way they position it is. Like, I think that might, I might not be able to ask that question again in a few months, a technical, Emmy or Oscar.

[00:42:35] I'll be, I'll put that one on there. All right. So then the last one, favorite AI marketing tech that you use that isn't MarketMuse, is there, do you guys have a go-to tool that you're using like a Grammarly or something like that? That is a huge part of what you're doing?

[00:42:51]Jeff Coyle: I've got a number of them.

[00:42:53] But I would say if I have to do that, that to say one, two so hard, [00:43:00] I have a pilot I'll rattle off a few. Okay. now I'll just limit it to one. MadKudu is one where I really, really think  . Yeah. I gotta check that out. I feel really strongly that the premium versions of the model training that they're doing is onto something very special for predictive lead scoring.

[00:43:26] Paul Roetzer: Oh, you have talked to me about them before. I think it was years ago.

[00:43:30] Jeff Coyle: We use them, they are embedded in our workflows and I'm an advocate of theirs, but yes.  I think that, I think that it's still a work in progress. But already what we've been able to do with them is, has been just innovative. But I think what we're gonna start seeing is product led growth.

[00:43:49] Enabled predictive lead scoring capabilities for self-service workflows for opportunity identification that I [00:44:00] think they may be the first mover there. And I think that that's, you know, that's the most exciting field because if someone cracks that nut. They're against the unicorn. Oh yeah. It's awesome.

[00:44:12]Paul Roetzer:  All right. Well, I could, I mean, we could talk, we have talked for hours about this topic so we could keep her home, but any final thoughts for our audience in terms of how they can better understand and apply AI? Like just where do they start or any final tips you've got.

[00:44:29]Jeff Coyle: It's the end of the year, take stock of your existing content inventory.

[00:44:34] Understand how much, how much content you've created, or how many content items you updated and the outcomes that those motions delivered. And if you're not happy with those percentages, it could be that your research process is bogged down by dated practices. It could be that your prioritization, like what you create is.

[00:44:58] You know, being done by the [00:45:00] highest paid person in the room or brainstorming, or, you know, some other data point that isn't legitimate. It could also be that you haven't adopted a source of truth for your writers. And you're thinking you're getting one thing and they're writing something else. And think through that content efficiency no matter which area of that workflow you think is inefficient.

[00:45:25]You can fix it and you can fix it in weeks, not months, you can put yourself in a situation where you don't make any more bad decisions on Content. And this is the time when I always asked teams to look at the mirror and say, how much did we create? What did it produce? Am I happy with that?  And what we talk about a MarketMuse is making that a reality, but it's also for Marketing AI Institute.

[00:45:55] There's a solution that has that for email. There's a solution that [00:46:00] has that reality for lead scoring or your BDR program or for, you know, so whatever, I think this is really a good time to just take stock in all of your marketing channels and determining which ones are not as efficient as you want them to be.

[00:46:13] Paul Roetzer: Yeah. That's a great takeaway. All right. How do people find you? What's the best way to get in touch with you?  LinkedIn guy, a Twitter guy, an email guy.

[00:46:22] Jeff Coyle:  I am a Twitter email, LinkedIn guy. So Twitter, Twitter, Jeffrey underscore Coyle jeff@marketmuse.com among LinkedIn all the time. Any of those, any of those three places.

[00:46:34]And then yeah, come to market muse sign up if you heard about it here, shoot me a note@jeffmarketmuse.com and I'll make sure that everything is as described. 

[00:46:47] Paul Roetzer: All right, Jeff. Well, as always great to catch up, I'm always inspired by our conversations and hopeful for the future of our industry. We'll have to do it again.

[00:46:56] Once the first draft is really humming in the industry and come back and take a look at the impact and when GPT-4 which I'm sure will come on next, it's going to keep moving. So we'll keep it up there.

[00:47:07] Jeff Coyle: They're going to put a supercharger on, you know, knock on wood. It'll be Q1 for first draft in bulk production.

[00:47:14]But if you want some examples, it's out in beta right now. So if anybody wants examples, fire me a note, Jeff@marketmuse.com. And you could have a first draft in your mailbox in a couple of weeks that I personally touched. So yeah. Give me a buzz. If you want to see this in action. It works great.

[00:47:34] Paul Roetzer: All right. Well, thanks again. This has been The Marketing AI Show and we appreciate you joining us today.

Related Posts

[The Marketing AI Show: Episode 1] Salesforce Leader Discusses the Post-AI Consumer

Sandie Young | January 19, 2021

You can now listen to the first four episodes of The Marketing AI Show podcast on your favorite podcast app. Read this blog post for more on what to expect in episode one.

[The Marketing AI Show: Episode 2] Microsoft Exec Covers Conversational AI

Sandie Young | January 20, 2021

You can now listen to the first four episodes of The Marketing AI Show podcast on your favorite podcast app. Read this blog post for more on what to expect in episode two.

[The Marketing AI Show: Episode 7] How to Leverage Conversational AI to Accelerate Revenue

Sandie Young | March 10, 2021

In a new podcast episode from the Marketing AI Show, explore how to use conversational marketing to drive pipeline that closes faster.