← The Best SEO Podcast: Defining the Future of Search with LLM Visibility™
The Best SEO Podcast: Defining the Future of Search with LLM Visibility™
How To Use AI Without Getting Deindexed With Jon Gillham
00:46:02
Loading summary
Transcript47 lines
- [00:00]
A
This is the Unknown Secrets of Internet Marketing. Your insider guide to the strategies top marketers use to crush the competition. Ready to unlock your business full potential? Let's get started. Howdy. Welcome back to another fun filled episode of the Unknown Secrets of Internet Marketing which I am trying to drop that piece of it. Okay, so best SEO podcast. The best SEO podcast. We have all those handles and Internet marketing secrets. We've decided to drop that. So I need to change the bumpers but you can find us anywhere@fest seopodcast. This is the Unknown Secrets of Internet Marketing. We've been running for 12 years straight and we talk about everything, digital marketing and SEO and well, AI because AI has taken over and I thought it would be good as we're continuing to have these AI discussions to hold up, pump the brakes and say, okay, plagiarism, AI content generation, how are we ranking? What's going on? I even remember a publicly traded client we have early on was like 92% human generated rejected. I was like, well he wrote it. You know he wrote IT and the LLMs are trained on human writing. So a lot of it's going to be human written. Like I knew this person written the article like he knew nothing about AI. I was like there's no way that he wrote this. And so that started an education process, AI governance process to understand how we need to frame these things, how we need to look at these things. And I wanted to bring on John Gilham from Originality AI. He's got a AI checker, plagiarism checker, fact checker. Because well, there's an issue with, we were talking to the pre interview, the, the integrity of the Internet as a whole as LLMs are referencing LLM sources. You said you just completed a study on, on how many AI overviews are referencing AI generated content.
- [02:02]
B
Yeah, thanks man. Thanks for having me. Yeah, we, yeah, we did, just did a study. I think it's a sort of, we find it infinitely fascinating to sort of understand how AI generated content is sort of proliferating across the, across the Internet. And so we look at sort of studies on where, where that's happening, where it isn't happening. We saw with this study, we looked at hundreds of thousands of a overviews and then the websites that they had cited and then ran those websites through our AI detector. Our detector, highly accurate but not perfect but across a large data set is, is very telling and can be relied on for, for understanding. And we saw sort of 15, 15 at times 20% of your money, your life searches citing AI generated piece of content. And so it sort of, yeah, definitely raises the question of sort of the snake eating its own tail as some sort of visualize it. But if AI is rooting itself in AI, there's a whole world of problems that can come from that.
- [03:04]
A
Yeah, the degradation of data as AI feeds itself. I would love to even go down the rabbit hole later on synthetic data and how that works. That's something that we're starting to get into and test out some different tools on. But like I remember Elon Musk, right, bought Twitter and then, you know, for freedom or whatever. Like I think it's great move, but I think he, I mean it's clear and I believe this to be true. He bought it to train Grok, right. And he also was tweeting about how many bots there were on Twitter and he had to get all of that synthetic data or that AI generated data out of there because it needs to be trained on real humans. That's also why Google did the deal with Reddit because, you know, real humans need, are providing the inputs. And I think that that's probably why ChatGPT or OpenAI made it free for so many people because they, they need the engagement with real humans to, to train, to train these models.
- [04:08]
B
Yeah, Certainly anytime that LLMs have, have attempted to do sort of a very serious effort on training on synthetic data, it has not, not gone well. And so the majority of LLMs are being trained on sort of all, all data that has all, all human created data. And it's sort of one of those sort of sort of AI paradigm shifts where you know, for the rest of humanity, all known human text data sets have, have been created. The AI has, has sort of infiltrated itself into so many things. Whether even, even when you don't think it has, if you're accepting Grammarly edits, then there's a little bit of AI getting added into that, that human text. And, and so for sort of the rest of humanity, any human text, any human data set will, will have some amount of AI in it compared to sort of pre2020.
- [05:07]
A
I've never really looked at it that way. That, that is really philosophical actually. Right. Like that, that, I mean we saw it with Grammarly and some of these other tools, you know, SEO Surfer. What is it? Surfer? Yeah, and like you can't get away from data that doesn't have autocorrect or you know, help helps you write it. So there's some level of infiltration in all information going forward. Right. That is, that Is like wild to think about in certain areas. It certainly accelerate more. I mean what to set the table even more. I mean school systems, right are dealing with this non stop. Like hey, like did you write this paper or not? And that goes even back to the question of when I was in school. It's like I'm always going to have a calculator. If I'm ever going to do any math, I'm always going to have a calculator. Why do I need to learn a long division? That was my argument for a long time. I think school's changing. Like you're always going to have an LLM as a co pilot to you know, or a sidekick to what, what it is you're doing and you have now the most intelligent PhD level in every category at your fingertips with you at all times. I feel like school should not fight that but embrace that and teach people how to think in systems and you know, but I don't know the rules around the plagiarism, the fact checking. I know I mentioned to you, I've seen in a couple of conferences someone created a city and got that cited but it was a fictitious city and then I've seen it with fault publications on like sport sporting scores or something like that. It gets sucked up and ingested so quickly and then it gets propagated and then there's reference points and now you're proliferating fake data. I think in the election cycles you're going to have some, some it's going to get pretty bad. It's already started to get there. I mean just to set the table more and when they'll zoom down on you know how to better rank a site using maybe AI generated content. What are the rules around this? Like what are the guardrails? What are the rules?
- [07:26]
B
Yeah, so, so in, in academy I can, I can talk sort of academia and marketing like so in the, in the world of academia I mean as, as you can expect slower slower to adapt resistance. Some people being super progressive about saying you know, LLMs allowed. You're going to use that exactly as you said calculator example. Others saying this isn't how brains are formed where brains are still soft and they need to go through this lifting of the weights of sort of writing and thinking through that process to, to get to a fully, to be capable of using these, these tools to their full potential. You don't throw somebody who, with, with who has never been a race car driver in a 300 horsepower charger and say good luck and, and is and sort of, that's the other, other analogy. And so it's, it's. I, I think, you know, I've got young kids, sort of 12, 12, 10.
- [08:17]
A
Eight.
- [08:19]
B
Relab rats for, for like they're, they're lab rats for, for sort of. We, as, as, you know, I was. For the Internet. And, and, and it, it's sort of what, what does that mean for education? That's a tricky question. I think if, if you're studying at a higher ed institution and you're an LLM can get a better Merck than you can, you should maybe rethink whether or not the thing that you are studying is producing a lot of value in the world. And so that, I'd say that's sort.
- [08:47]
A
Of a, that's the big question, right? That's the big question that everybody. And then, you know, you made me think of that MIT study that came out that showed that people are relying on LLMs for thinking, not, not leverage, but thinking, and their, their, their brains are shrinking or whatever or some, some, some capacity of that. So, so I, I think any really sharp blade or whatever analogy you're using, there's two sides of it and you, and you have to be cognizant. So I think that's a great weighted point there, John.
- [09:19]
B
Yeah, yeah. And then, and then sort of on the, on the marketing side. So if we're talking about like, hey, you're in an organization, you function as a marketer, your job is to produce content that gets you users, whatever, whatever your objectives are. I think the most important piece that I think is often being missed right now is alignment around the proper usage of AI, where it is allowed to be used, where it is not allowed to be used, and the controls put in place to be able to manage that. What we're seeing an extreme example, but we're seeing interns come in with no guidance around AI usage, spinning up an API, spamming the site with, with AI generated content, and sites getting, getting absolutely crushed by, by Google. That's sort of like the extreme example, but the sort of, the risk owner, the business, like the business owner who, who is sort of accepting this risk, is doing so with, with sort of no knowledge and awareness of the risk that they're accepting. And so that sort of step around alignment on where it can and can't be used, whether that's in the editorial guidelines or kind of whatever, the AI usage policy that gets gridded, that's sort of that first step that I think is that you were alluding to, but that is often missed and there's some significant consequences when some people are just turned loose.
- [10:50]
A
Yeah, I think that executive teams and owners need to understand this technology, to understand how people are using it. I, I see a bifurcation many times of legacy businesses that, you know, even digital marketing, even components of digital marketing, they don't completely understand Data governance. Right. And with LLMs, if you're connecting it up internally, maybe to a Google Drive, it can pull everything. I've heard horror stories of HR materials getting accessed. And if you have an intern that understands how to use this stuff, sort of, and you're giving them the race car and no rules and they're driving it all over the place, it's going to break some stuff, it's going to destroy some stuff. So I think data governance, AI governance is a big topic as well as like ethical AI. Like, I think there's some real societal issues of this technology seeping into everything and everybody has access to it. Even. There's some real, even personal horror stories of guardrails that need to be put on AIs. What, what are the, what are the big frameworks or the, the big bumpers that, that just, you think everybody needs to know. From an education standpoint, that's maybe creating content online. Let's start with the first principle basics here.
- [12:22]
B
Yeah, yeah. So I think it's important to see that like, not all, not all AI content is spam. All spam in 2025 is, is AI generated.
- [12:35]
A
And that, that, that's a point that Google came out and I felt like they were waving the white flag and said, hey, that's when they added to eat expertise, authority, trust. They added that experience component to try to say, prove it right, because you can just be the expert. But they said if it's useful content, AI content is, is okay. So if you're, you know, working on the content and you're linking it and you're citing references and you're, you're adding, you know, images like that content is okay. Sorry, I kind of cut you off.
- [13:09]
B
Yeah, no, so like, so like back to like that first principle. The basics of like, okay, not all a content is necessarily spam, but if you just leave it to its own devices and somebody without controls, it'll turn into spam pretty quickly because it's optimizing for the wrong metric. And so that's sort of one important guideline, another sort of like important guardrail. And I think you kind of alluded to it. But if you're, if you're competing on just words right now, then you're, you're facing a challenge. And so if you're sort of in the business of trying to put out 750 words on topic X, that's you're now competing with, with LLMs that can produce sort of, you know, near top level intelligence words at near zero dollars at basically the cost of electricity. And, and that's a, you're, you're going to lose that battle over time. And so you need to think about adding value beyond just the words. And so that's sort of another sort of like talk about bumpers. Like if you, if you're viewing, if AI left to its own devices, somebody left to their own devices going wild, they will end up producing spam. Google is on it has an ex, has an existential threat to their business. If their search results are overrun by nothing but AI spam, why would anybody go to Google? They would just go to the LLM that had the most knowledge of them, which might end up being Google, but it's not going to be you. 10, 10, you know, 10 blue links a click away. And so there's sort of the existential threat factor that if Google search results are overrun by AI, then Google search results as we know it will die, which maybe they are now already. And so that's kind of given all of those, what I would say is sort of the guardrails to understand when it comes to Google doesn't hate AI, but hates AI being the sort of overrunning the search results and having no extra value being added.
- [15:12]
A
Okay, so the first thing you said, we talk about a concept called human in the loop. Like you need to have somebody in there looking at it, checking it, making sure it's optimizing for the right things. On the second piece, I absolutely think Google's being overrun and they need to really do a lot better job with AI mode. I think it's pretty bad as of this moment I used it. It's very clunky. Gemini is okay, but ChatGPT is crushing it because of the positive reinforcement or however they've weighted it. And I think that maybe you can speak to this with these Google updates. Is there a threshold of a site that's like, okay, if it hits 20% or 25% AI generated content, it's getting flagged or it's getting deprioritized in the search rankings. Have you seen anything or know any data related to that?
- [16:10]
B
Yeah, so we have some data. So we're tracking the amount of search results that have AI Content in in them. And we're seeing that rise at a much slower rate than most other online platforms. So we're seeing medium is at like 50% of content at times has been suspected of being a generated LinkedIn. Over 50% of long form posts are were in their last study were likely a generated Google is likely 20% is kind of where it's, it's staying. And there has been times where that has declined which lines up with Google taking significant action. And so March 2024 Google did a manual, I called it like a Psyops update where they did a manual de indexation on a ton of sites. The vast majority of those sites had the majority of their content being AI generated. And so what what we are seeing is it's hard to say like at this point at this percentage you're at risk. Well, this percentage you're safe. I would say what we are seeing is that you know, Google calls it scaled content abuse. When, when a large number of pages are getting published that is something that is easy for Google to identify. It looks like at that point there is sort of a second level check on the site done and then those sites are getting, are getting nuked. And so if you are scaling content rapidly, you are very much at risk of having a penalty in some capacity applied to your site. If you're scaling AI content, you're definitely at risk of both getting it flagged and then getting punished. How sites have been using it successfully. It's again really up to if, if you. I think my, I'd say my sense is if you stay off the radar of the scaled content abuse and your helpfulness stays extremely high and that it sort of shows up in the, in the user metrics that, that they rely on, then you're, you're probably kind of in that, that okay, maybe in the maybe okay camp.
- [18:23]
A
Yeah. And algorithms are deciding whether you fall into these categories or not. So there's definitely weighted thresholds on all this stuff. How in your world are. How is plagiarism being viewed today? Because. And also hallucinations by these LLMs because they're just predicting what is most likely the next word. And I've found it to be. It's like oh, link to this content and like oh, that page is not built. Maybe we need to build that page. That could be a good page.
- [19:00]
B
I haven't heard thought of that insight, but that's a great insight. Clearly the ll wants this thing to exist so that it can cite it.
- [19:07]
A
Yeah. Yeah. Well, so, so we're Working around a term like there's a lot of these different terms like chatgpt, SEO, Gio ao, like I, I, I don't think the industry's decided on it. We've really gone with LLM visibility and said we're building around LLM visibility. That's ultimately what this is, this is what we're trying to do. And we built like a custom framework to do that and a strategy on how to make that happen. And we're even working on and working with some partners on some tools and indexes and things and you know, there's a lot of noise. But I think back to your point, which I do want to talk about plagiarism, but Google is dying. I think Google's trying, this is how I see it. I see Google's trying to take on Amazon with like buy right now, right? We have the search traffic buy right now. And then ChatGPT is becoming like taking over Google from a search standpoint at scale. Right? Like they're just winning the at scale. So I feel like the business models are shifting. Even Google announced the big launch with YouTube and I think that that's one of the battles is AI generated. People are coming and are here. Actually they are totally here. But the vast majority of people on YouTube are actual people and that's really rich content that can be used a lot of different ways. And so Google and ads and everything of what I'm seeing is being pushed to YouTube and to like Buy now. I don't know that that's at a high level what I'm seeing.
- [20:51]
B
Yeah, I mean I think I, I think what, what's certainly evident and I'd say, well, you know, what's, what's gonna, if you know, rubber crystal ball, what's going to happen with, with Google we're going to see, you know, sort of in like basic metric stuff, but it's like we're going to see reduced clicks, we're going to see increased conversion from those, those clicks that we get. The, the, the funnel will be collapsed. I think we're seeing that already where people are going to be in their, their LLM of, of choice to do the research, gain the knowledge and whatever they're wanting to do and then their transaction, they'd move to the web for, for transaction. Maybe that'll eventually get to you, you know, to the, the one click situation of hey, plan my trip, go book it all.
- [21:37]
A
Yeah. Oh, they age in economy here. You're talking that. Yeah, yeah, yeah, I agree.
- [21:42]
B
And so, you know, I think I think we're going to probably step our way there and I think sort of what the world will look like over the next couple of years maybe, maybe you know, hard to make predictions that age well in AI, but it's will be people continuing to exist in the LLMs for the, for that knowledge gaining. As a result, most websites seeing a reduction in clicks, especially if it's superficial layer information and then a increase in quality of the clicks that they get from a standpoint of conversions and that yes, LLM visibility and seeding LLMs with the knowledge that drives them towards your solution as being the optimal solution for that problem is I think definitely the name of the game.
- [22:30]
A
Yeah. And I know that we're SEO podcasts but man, I also see to the link the, the agent economy. I see crypto, right. I see the money of the Internet coming involved in this to be able to transact like I think there's just so much transformation that's happening and to your point, predictions don't age well. Who knows where, where things are, are going to go. And, and I like, I like, yeah, a lot of it is about seeding the lms and I talk about right now is the opportunity, right, right now is the opportunity where everybody's trying to kind of figure out what's happening. Google, I feel like AI overviews was like a stopgap measure to keep people from moving to the LLMs so quickly. But as people use the LLMs now, a study did just come out, I can't remember who put it out. I'll have to have them on the podcast. But basically it showed, and this is early, that data from the LLMs were just as good, maybe slightly better, but not as much better as I thought it would be than Google itself. So I thought that was interesting. I however think that the customer journey is about brand management and if people are using last clack glass, click attribution, who knows, right? They, you know, things come in on an ad but they've seen it on Facebook. They've, you know, they, they've saw it on Reddit. They Maybe used multiple LLMs to make decisions. It's hard to say what's happening and how people are making this decision. It's, it's really about how is your brand showing up, how visible are you in these LLMs. But you know, I'm dealing with John entity issues. So you can see my name here is Matthew Matt. So I'm showing up in the knowledge graph as two separate people because my, my title's changed, my name's changed. We've changed the name of the company, we changed the name of the podcast and there's also other people out there with my name. So there's some ambiguity around who is me or what. And it's, it's becoming a greater and greater issue. So I'm trying to disassociate as well as I'm trying to overlap these tiles because my, my actual name is Matthew Bertram and then Matt Bertram and Matt Bertram Live are just aliases actually, or nicknames or whatever you want to call it. And so understanding how all these things are connected kind of go back to that concept of seating the lms, but speaking to the LMS in a way that they understand what you're trying to tell them because they're doing a really great job of having to sort through all this information and give you the best possible answer to what you're looking for. So this is really important piece of the future, I think.
- [25:38]
B
Yeah, no, it's for sure. We're, we have a content optimization tool within originality and we are sort of adding a feature very shortly that is all about LLM visibility optimization and tuning that content to what has been known to date around studies around stat placed high in the findings. Quote from Expert sort of what are the features that LLMs want to. Want to ingest?
- [26:15]
A
Yeah. No. All right, let's circle back to plagiarism. Okay. And, and start, you can start very big picture and then tighten it all the way down to like what the marketers want to hear.
- [26:27]
B
Yeah. So I mean plagiarism been especially in academia monster business. Everyone's been been through it since sort of the birth of the Internet. There's been been plag plagiarism checkers. What we have seen sort of both both in. In that world and in sort of the digital marketing world, the amount of plagiarism that's happening has been on a massive decline. Because why would somebody plagiarize when they can just copy and paste and get something out of chat gbt? And so whether it's been plagiarized and then rewritten by AI, there, there's certainly some. Some risk of of all of those things happening. But what if the Google trend is really fascinating. We, we actually should mention this early because it's kind of funny. We launched Originality AI three days before ChatGPT launched and when we launched there was zero search volume for AI detector. Now it's afforded 8 million search volume keyword a month depending on. Depending on when it's happening whereas plagiarism is like sub 1 million. And so if you look at the Google trends between the two, it's quite fascinating to see this sort of plagiarism checker seasonality been there forever, growing year over year chat GBT comes out, a detector takes two years but now skyrockets above above plagiarism checker. And plagiarism checker is starting to decline. And so I think it's still something that is worthwhile checking because there's legal risk associated with direct plagiarism as a marketer. And so it makes sense to sort of include that in your QA QC process but the sort of prevalence of it is significantly declining.
- [28:17]
A
Let's talk about this sensor around that. What are all the major laws or guideposts that people need to follow if they're trying to build a program and incorporate AI like is there references that they should look at? Is there laws they need to be aware about? What would be the best kind of governance policy around this?
- [28:43]
B
Yes, so it depends on the use case and so there's certainly fair use allowed of other people's content depending on that use case and and depending on the proper citation of that of that content. And so what are the best practices around ensuring you don't get your business in trouble with with plagiarism? Is running a plagiarism check identifying the sources that get identified and then doing a manual review to see is that truly copied or is that just your own words written written the same five same five words that can be used and by, by multiple people multiple times. And so I I would say it's some some organizations so the the rules around it for for most marketers would be around sort of like fair use. 5%, 10%, 15% thresholds are are not uncommon as sort of for for companies to to sort of require and then ensuring that anything is properly cited. You are you are certainly on solid footing if you're sort of ensuring that there's 5% or less plagiarism in a piece of work and that any time that there is copying or other sources identified, citing, reviewing if those should be cited or not, that that gives you a pretty solid footing that you are not going to get yourself into any any issues.
- [30:13]
A
No, I like that I started to think images images there's a whole business out there of you've used this image for fair. You know it's not fair use and you know they're trying to get money and people are doing that and now just like content you have AI generated images which are completely new images. And I was listening to one of the Google podcasts, Google Webmaster Podcast, and it was, it was saying that that's totally okay. It was like, that's completely new. That's totally okay. And then the proliferation of content that they're having to index and that's why the unindexing is starting to happen. Because you, the standards across the board, if everybody's, you know, it has moved up, right? Like the, the standard has, has moved up. I would love to hear from you maybe on the horror side of things as well as like the awesome side of things, some case studies on how people have used your tools and what are some of the success stories around that?
- [31:19]
B
Yeah, yeah, sounds good. So I'd say kind of a horror story. And we unfortunately hear it too often, but someone typically a website owner will come to us and say, hey, our site just got de indexed. We weren't using AI and it's like, okay, like maybe let's run. We have a website scanning feature. We feel terrible. Here's, here's a bunch of credits that has happened to your, your business. Go, go run a website scan and see what the, what the tool says. And then inevitably there's a point in time so you can see this graph of the website's content. And inevitably there's a point in time where whether something, something changed in that, in that editorial process and all the content went from call it one post a week to eight posts a week and they were all AI generated. And then the, the site, the site tanked. Those, those ones suck because there's, there's real pain associated with those businesses laying off employees, livelihoods lost because a website got tanked because somebody on that team was taking on risks that the risk owner, the business owner didn't, didn't understand. And so that, that's been a, been sort of a pretty common use case that, that we have, we have seen where it's, it's definitely a, a horror story when we see it.
- [32:48]
A
Have you seen a threshold? I feel like Google's taking a lot longer to index stuff. I don't know the sandbox terminology as well as the, the, the delisting or unindexing of the content. Did you, have you seen like a, a point where they're, they're indexing it but they're not really ranking it. Like there's a lag time and maybe they're, they're running it through some of these systems or also is there like a grandfather Period that I feel like older content does better. There's some equity accrual or link equity accrual that, that potentially happens over time from what I've read in the, the, the trade, the, the, the patents or whatever. So there's a lot of tools now that you can pull out from the patents like what's happening. But I wonder if there's a way to, to know where that cutoff is or where that threshold is because new content's taking a lot longer to get indexed. And so when we're working with clients and you know, they usually come when they're in a bad situation, you know, to turn it around. We want it to happen quickly, but it's happening a lot slower than we would like to see it. Like Google doesn't turn on a dime anymore.
- [34:13]
B
Yeah, I mean, I think, I think core, I mean it, it seems like there's more. My sense is that there's less movement between core updates than what potentially used to be the case in Google. And so like movements happen when those core updates are happening in between those core updates. There's not as much movement in, in the results as, as there used to be. I believe that like that's certainly what we sense and I believe that's kind of what the data. Data supports. And so I think that lends to sort of the sense that new content getting published doesn't takes longer to, to drive results because it's sort of waiting for that. Especially if a site is sort of in that, in sort of a down. A downward trajectory. It takes that to the next core update to, to sort of see that. Okay, we've addressed the eat issues or, or whatever. Whatever it might be.
- [35:10]
A
Yeah. Yeah. There's a lot of conversations I think at Google around trust, like levels of trust and you know, thresholds and you know, there's. It depends. Right. That's a horrible answer. I know.
- [35:23]
B
Yeah, I'll always answer.
- [35:26]
A
What are some really positive use cases that you've seen with originality? AI to help. Help save something or saw something and caught it before. Maybe they, they got penalized by, by Google or something.
- [35:42]
B
Yeah, so we, we. I mean the, the common use case for originality is that there is an somebody functioning as an editor within a marketing company, within a website writing team. They function as an editor. They have their team submit content to them and they run a QA QC process on, on that content. Some of the things that commonly happen are writers swear up and down that they didn't use any AI runs through the tool says that it was AI. And then they share. And we have a free Chrome extension for these situations where sort of the writer says, I didn't use AI. The tool says, you did. You likely did. The tools provide sort of a probability, not an absolute judgment. And then it says, then the editor takes that document, the Google document, to our free Chrome extension, which then visualizes the entire writing process. And at that point the editor can see that the writer just copied and pasted the entire text in the document. In the document. A couple formatting changes and then got 100% AI score shows that to the writer. And the writer says, yep, you're right, I apologize, I lied. I did use. I did use AI.
- [36:59]
A
So go back to that one piece. So if someone's cutting and pasting something in a document, there are markers or tokens. What are, what are like the little dashes, right, that, that come out. Because AI can't. It's easier to put together a thought without putting together a full sentence. I think it's just. And so I think they put that. I felt like it was like a watermark for a long time. Um, like what. What are the fingerprints? What are the telltale signs that something's AI generated?
- [37:32]
B
Yeah, it's. It's an unsettling answer that I'm going to provide. But. And, and it's sort of similar to like ask. Ask John Mueller or anyone at Google, you know, why. Why is this ranking above this? And it'll be a bunch of sort of general platitudes that, that. But the reality is they don't know it's an AI. It's a black box. They can sort of understand its behavior on a large scale. And similarly, we can understand our detector's behavior on a large scale, but on any individual piece of content. Why did this, why was this identified as AI or not? That's very challenging to do because the AI detector itself is a black box. We have a feature coming out called Deep Scan, which looks at a. Understanding, better understanding how the text could be sort of adjusted if it was incorrectly identified as, as AI. So that people can sort of. Writers can sleep at night knowing that their work is going to pass an AI detector. And so it's, it's a unsettling answer. There, There are some things that can lead to it. So if it is AI content, it is more likely to get identified as AI content. AI content. If it is highly formulaic, highly formulaic, very structured writing, or oddly formatted, those can lead to so highly like, highly formulaic, highly structured writing. Can lend itself to looking more like AI. Oddly, formatted text will reduce the AI detectors accuracy and then therefore will result in more times it being identified as AI.
- [39:19]
A
So specifically your tool, you were saying that like it was like cut and paste it. Like is the tool looking at the number of drafts or how long the document. Is that what it's looking at, how long the document took to generate the document?
- [39:35]
B
Yeah. So sort of best practices are that an editorial team says if they're using sort of the originality tooling is that the editorial team says writers, you must use a Google Document from start to finish. And then that Google document will get a score for AI, the probability of it being AI. And then it will receive a report that shows the length of time, the number of writing sessions, characters per characters over time. And if you see this sort of thousand word document that was worked on for two hours with a bunch of edits and deletes and then one little section is identified as potentially AI and you've worked with that writer for a long period of time, you can be confident that no, this, you can sort of see the writer having written it. It's mostly. So the detector has said this is likely human, but this is some, some uncertainty. You look at the Google Document document and that and that Chrome extension and you can see them putting in two hours of work into that, into that piece of content. You can be fairly confident that is human generated human is in the loop. And this isn't just sort of a copy paste out of chat GPT.
- [40:48]
A
Got it. Perfect. Okay, so is there anything that I'm, my brain's full. So is there anything that we can, that we haven't covered that you think is really important for us to discuss based upon our current conversation?
- [41:06]
B
I think a couple things that I always love to sort of like make sure are sort of understood. Like I think there's a lot of this is something that didn't exist like from a search volume, keyword textures didn't exist, you know, a couple, couple years ago. And there's been a lot of misunderstandings around, around them. They're highly accurate, not perfect. And so sort of on any individual case they, they can have false positive, they can have false negatives where they incorrectly identify something as AI or human. And so that's, that's important to sort of understand. And then the second piece is that originality and most tools provide a classification of AI versus human and then a probability, a confidence score. So it's, I think it was a, the AI detector will say I think it was AI or I think it was human. And here's how confident I am in that prediction. That often gets misunderstood as I'm 70. Like this content shows up as 70% AI, 30% human. And that's, that's not exactly what that means. It's a binary classification, AI or human. And then a confidence score in that prediction.
- [42:17]
A
Got it. Okay, very cool. So what are the biggest takeaways or biggest tips that you can give to marketers today that are jumping in head first into AI because you need to be, but doing it in a responsible way?
- [42:38]
B
Yeah, I think first is ensuring that the risk owner of your company that you're doing marketing for understands the, and agrees upon the, the where the AI is being used and the risks associated with that. So I'd say that's first. Second is ensuring you're in the content that you're producing, ensuring you're adding value beyond just words. If you're competing with words, you are going to, you're, you're competing against sort of infinite free words that, that's hard. And so finding ways to add value beyond words, tools, graphs, primary data, finding a way to add. And then third, it's, I've been sort of doing SEO for long enough. It's always very, very, very tempting to find that shortcut and click a button that, that doesn't, that has never and doesn't exist now. That makes it makes this whole process easy.
- [43:34]
A
I love that. Well, John, how do people get in touch with you, find out more about your work, your studies? They can of course go to the website, but I'll let you kind of share your handles and stuff.
- [43:47]
B
Yeah, so yeah, we publish studies constantly. Heavy focus around sort of how AI is impacting the sort of Internet and how many people are using LLMs.txt. we have a study sort of running, tracking the number of websites that are using that, which is always interesting to see. You can see all of that at Originality AI and sign up for our newsletter and we, we keep sharing. People can get in touch with me Joniginality AI or find me, find me on LinkedIn.
- [44:18]
A
Awesome. Well, John Gilham everybody. Thank you, John, so much for coming on. If you got value out of this podcast, please go to the platform you're listening on and leave a quick review. It would be super helpful. You know, if it has to be AI generated that's fine, but we really need some reviews from real people and you know, if there's things that you would like to hear about or topics or feedback, please please leave that as well. We want to engage with you. This is a exciting industry that is growing. I think it's transforming from just, you know, SEO and a vertical just like traffic is left. 50% of traffic has left the website and has gone everywhere else. LLMs are are coming into it as well. So share this with somebody that you thought was valuable like it tag us Shaiko us Share like follow we really appreciate it. If you want to grow your business with the largest, most powerful tool on the Internet, which I guess is well bigger than the Internet now, maybe soon to be LLMs, reach out to EWR for more revenue in your business. Follow me on LinkedIn. I'm trying to post more. We are launching our LLM visibility certification very soon. John, thank you so much for coming on. Until the next time everybody, my name is Matt Bertram. Bye bye for.