Podcast: Play in new window | Download
Now that we’re becoming more experienced at using AI tools, how do we make sure all of the content out there isn’t just AI generated? Madelaine Lambert of Originality.ai teaches us how to become “AI sleuths” to differentiate how certain content was created….and possibly even how to outsmart the AI detection software.
Can Your AI Content Be Detected Episode Summary
- Madeline Lambert discusses her career transition from founding a content marketing agency to working in the AI detection industry at ai. She explains how she and her business partner decided to pivot their business and offer AI solutions for content marketers, aiming to address the potential challenges brought by generative AI tools like Jasper and GPT-3.
- The discussion revolved around the challenges of dealing with false positives in AI content detection, and the tool that allows content creators to submit links for review, providing a way to determine if the content was AI-generated or not.
- The use of AI in content creation, the benefits of AI as a helpful tool, and the importance of maintaining a human touch and not relying on AI to replace genuine storytelling and emotional intelligence in content.
- The impact of AI on the writing industry. While AI has benefited a select group of writers with specific expertise, the recent Google update negatively affected sites that used AI-generated content, leading to penalties and disapproval from Google AdSense.
- The correlation between Google AdSense denying websites with AI-generated content and the ability of Google to detect AI usage, and the possibility of an ongoing cat and mouse game between AI detection tools and AI generators, and how imperfections and casual conversational style can help differentiate human-created content from AI-generated content.
- How businesses in the content marketing and publishing industries sought the help of originality AI to detect AI-generated content and prevent the publication of misleading or dangerous information.
Can Your AI Be Detected Episode Transcript
Rich: My guest today is a seasoned entrepreneur and marketing professional who founded a successful content marketing agency before shifting her career to the AI detection industry. As the director of marketing and sales at originality.ai, she leverages her marketing skills to promote cutting edge AI solutions built for content marketers just like you.
Today we’re going to be diving into generative AI and everything AI with Madeleine Lambert. Madeleine, welcome to the show.
Madeleine: Hi, thank you so much for having me.
Rich: So it’s an interesting career shift from founding a content marketing agency to running marketing for an AI detection company. How did that all come about?
Madeleine: Yeah, well, good question. So basically, my business partner and I had been working in the content marketing agency for quite some time. We had founded it in 2016 and had grown it and had really figured out our audience. And then, we started seeing tools like generative AI tools starting to come out like Jasper and other tools that do similar things. And so we kind of came to a bit of a fork in the road and we said, hey, there’s going to be a wave of problems that come with this. Do we want to do a business shift and get in front of it and, come up with new products, new offerings? Or do we want to make an exit and then really get in front of the problems that we anticipate?
And so we decided to do that. We were kind of tapped out on our content marketing side and or agency side. And we were ready to start over again with something else. And so this seemed like a good pivot.
Rich: So you did this with your partner?
Madeleine: No, not quite. So I was very involved with the content marketing agency. And then we actually sold it the week that my daughter was born. So it was a busy week. And so I was away on maternity leave, in Canada we get great leave, we get like 18 months. So I was sort of out of office, my business partner had been talking about this for a while and had started basically building this AI tool on models like Jasper, like the GPT 3 models. So I had kind of gotten in front of it. And by the time that I got back from that leave, he had sort of launched it and then hired me on as his director of marketing and sales.
Rich: Wow. Exciting times. And for my American listeners, they are probably thinking, did she say 18 days? Because that seems like a lot for a maternity leave. But, you know, it was months, people. It was months.
Alright, I want to get into some of the ways that we can drive our marketing forward with AI. But first, I’m curious about originality.ai and what exactly is it that you do there? What does the company do?
Madeleine: So basically, we equip agencies, writers, contractors, anybody who’s really involved with the content creation process, with a tool that will help either detect whether generative AI was used to help create the piece of content, or if you’re a writer, it will help sort of prove that your work was original if you are aiming to deliver original work, if that’s sort of your mandate.
Rich: So it’s interesting, because I’ve heard stories about students and professionals getting their content tagged as AI generated when it wasn’t, or at least they said it wasn’t. How do people avoid or deal with false positives? Do they need a tool like yours? If they don’t have a tool like that, what are their options?
Madeleine: Yeah, it’s tough. So unfortunately, the big backlash that AI content detection faces is that it’s always going to have a false positive rate. No matter how you skin the cat, there’s always going to be a very small percentage of false positives. And that sucks when you’re a writer. And we’re very, very, very cognizant of that. And we have been on the content marketing agency side of things so we’re extremely cognizant of that and compassionate to that problem, and we worked very hard to reduce that false positive rate as much as possible.
But the reality is, it’s AI, it’s a predictive tool, it’s always going to be there. And so to sort of get around that using a tool like ours, we’ve actually created a plugin specifically for content creators. And this plugin allows you to sort of submit a link to whoever you’re submitting the content to, whether it’s a teacher or like an employer or a client. You can submit that link and it will give that person the ability to see the creation of that piece of content. And so you’re going to be very, very easily see AI generated or not because you can literally see the person writing. It takes a screen capture of the person writing and it will do a timestamp of how long this piece of content took. And we all know that if you’re spending four hours on a piece of content, it’s probably original. And if you’re doing like 20,000 edits to it, which this plugin will track that, then it’s probably original. So that’s kind of how we’ve helped solve for our false positive rate. Because it does suck.
Rich: Yeah. So without giving away any trade secrets, I’m just curious. Do you know of any AI ‘tells’ that your software picks up on? Like the first couple times I uses ChatGPT, it’s magic. And then after a while, it kind of starts to sound the same. And obviously you can do prompts to make it sound differently. But are there some tells that are common right now that you see out there that we should be aware of?
Madeleine: Yeah, it’s funny. We get this question all the time. It’s like, oh, well, what is it that your AI detector is picking up? And I wish we could tell you because it’s AI. It’s trained. It’s basically trained itself, right? So we wish that we had more insight on what exactly it is that it’s flagging.
But the general understanding is that generative AI will produce content that has similar sentence lengths, and it will pick up the similarity in word sequences and stuff like that, that are more common with generative AI.
Rich: Yeah, I noticed. I didn’t notice it right away. I did a podcast interview many months ago with ChatGPT, where I posed the questions, got it back, and then used a different piece of AI to do the voices back and forth. And it wasn’t until I’d been using it for quite some time I realized some of the ‘tells’ of AI.
And this was really basic stuff. Just the repeating of the question and then the posing of certain phrases that it would use multiple times throughout the interview. Which admittedly, I have my own ‘tells’ as well, but you know, it was just kind of some funny stuff.
Madeleine: Yeah, for sure. And the tonality is very similar across the board, right? Like it seems to be very neutral, and sort of doesn’t have much personality, and it’s safe.
Rich: It’s safe. So moving away from the AI detection piece of it, and just kind of thinking about how AI can help or hurt us as content creators. How do you recommend that we, as s content creators, as marketers, copywriters, brand ambassadors, what have you, how do we show that there is a human behind the words that we put out there?
Madeleine: Yeah. Yeah. So here’s the thing. There’s such a misconception about us because we are an AI detection software. And people think that we must be anti AI, but we love AI. We use it all the time. I use it. I’m the director of marketing for this company. And I use generative AI every day to help leverage my content creation. I think the distinguishing factor is I think AI is such a great, helpful tool. Where I think people go wrong with it is I think people get lazy, and I think that people aim to replace rather than leverage.
And so I think that if you’re using these tools in a way that’s going to make your content better, make your process faster, and sort of leverage what you already have and amplify it to make yourself better and to make yourself smarter, than that’s great. Where I think that people get it wrong a lot of the time is to create shortcuts and replace themselves with something that’s not as good.
Rich: So, many people – copywriters especially – say that AI doesn’t have emotional intelligence, it can’t tell personal stories. But the flip of this is, there are these large language models and they’re filled with other people’s personal stories, and certainly a lot of content around writing with emotional intelligence. Can’t we just tell AI to create content, or create content with emotional intelligence and fake it?
Madeleine: Well, I mean, I’m sure you could find some creative prompts that would help with that. But if you’ve got a brand voice and if you have a track record of having that consistent brand voice, it’s going to be really hard to replicate.
And so I don’t know. I think that if you leverage it properly and you’re able to allow it to help you create and then hunker down and edit really, really well to have it sort of reflect your brand voice, then all the power to you. But I think people, again, people just need to be careful and not aim to replace themselves with these things.
Rich: I’m starting to think about, like, there’s different types of content. Obviously, there’s different types of content. But there’s the content that’s really critical to your business and your brand, and then there’s the transitory content or the filler content. So I might spend a lot of time on a very important blog post, but then I would use AI to create the social post, tweak them a little bit, but just kind of that’s not as important to me, so I can outsource that fairly easily to AI.
I think it’s putting that emotional intelligence, and telling those personal stories or a case study that you ran, into your content that really helps separate it from AI. Because the thing is, I read a lot of people’s writings, and AI writes better than most people I know. I mean, that’s the sad truth of it. There are some amazing writers out there like Mike Kim and Ann Hanley, and others. But there is this sort of, it brings the bottom up to the middle, but then there’s still the people who are going to be able to tell better stories than AI can.
Madeleine: Right. And I do think that AI has caused a world of hurt for the writing industry. But I do think that the writing industry has gotten so large and so overpopulated with mediocrity that it has helped sort of extract the really, really great writers. And unfortunately, the writers that have sort of fallen short have been able to be replaced pretty easily with generative AI.
And so I do think that it’s benefited writers, some writers, like a very elite cohort of writers. And it’s funny, as a client I have a couple of really, really great writers, and I’m willing to actually pay them way more now than I was before. So I do think that it has benefited some types of writers that have very, very specific expertise in things.
Rich: Many of us, I assume many of my listeners, have begun to use generative AI in our content creation. Whether it’s ideation, developing frameworks for a post, generating social posts, or full-on article creation. Which I rail against, but I know there are people out there doing it. One thing that isn’t talked about a lot that I’m seeing is the impact of generative AI on SEO, or perhaps the impact of SEO on generative AI. What are your thoughts around SEO when it comes to this AI generated content?
Madeleine: Yeah. So I think that it’s definitely a gray area right now. I think that it’s so new and Google doesn’t really know what to do yet. But I mean, have you seen the Google update, the most recent one?
Rich: Which one is that?
Madeleine: The most recent big one.
Rich: I may have. I mean, I’ve seen the search experiments that they’re running where they’re pulling in from BARD and then using citations for some of the results. I don’t know if that’s the one you’re talking about though.
Madeleine: No. So I haven’t had like a… full disclosure, I don’t know what I’m talking about. No, I’m just joking. Full disclosure, I haven’t had the chance to dig in too closely, but we are running some studies currently to understand – it appears – that this Google update has negatively impacted sites that use a lot of content that was AI generated.
And so we are digging into that right now. We’re trying to understand whether that’s a claim that we can make. But it appears that that it is a trend that we’re seeing right now. So we are seeing penalties. We are also, and I’m not sure if we’ve touched on this before in our pre interview, but we had basically a client come to us and say, “Hey, I just bought a site. I bought it for a lot of money, and it was monetized through all sorts of different ad platforms. And so bought the site, made the transfer. When I went to reapply for like Google AdSense, they denied me.”
And this client was like, man, our traffic was like 30,000 users a month. Like, what the heck? This is crazy. Our content is ranking. I don’t understand why we would get disapproved by Google ad sets. And anybody who has a site knows how easy it is typically to get approved. Just throw a couple logs up there, have them indexed, and Bob’s your uncle.
So we did some digging, and we scanned this person’s site for plagiarism, and then we just scanned the whole site for AI generated content. And there were like a handful of articles in the top 10 ranking articles. I think there were like three or four of them that were AI, flagged as 100% AI generated.
So what we did was we actually unpublished them, just to see. And we said, reapply in a few days. And so this client reapplied, and sure enough they got accepted right away. So that, to us, shows a pretty strong correlation between Google AdSense denying your website because of AI generated content.
Rich: So it’s interesting, Google seemed fine with the AI generated content in terms of ranking those pages, but not so much that they wanted to run ads against it. Probably because they don’t want to be supporting millions of websites that just have AI generated content up there, or obviously AI generated content, at least.
Madeleine: Right. Which indicates to us that Google has the ability to detect whether AI was used to create content.
Rich: You think it’ll be like an arms race, like tools like yours are going to be able to detect ChatGPT 3.5 and 4, but then maybe 5 it’s going to be harder to. And then you create the next iteration of your software that catches up on 5, and then the next iteration… like, is it going to be that kind of cat and mouse game going forward?
Madeleine: Oh, it always has been since the get go of this thing, right? They’re just getting better and better every single day. And we actually have an entire, it’s called a ‘red team’. So we have a ‘red team’ made up of machine learning engineers, and their whole role is to try to beat our AI detector. And when they beat it, we’re able to then train our model against that to make it better and better and more robust.
And so their whole job is to get ahead of it so that we’re playing cat and mouse, but figuring it out first, and not allowing somebody else or another competitor or another tool to figure it out before us. So we have five or six people whose full-time job it is to do that.
Rich: You mentioned earlier, I think I heard you say basically that your software, one of the things that it can do is you can kind of be creating your content, and your software it sounds like will monitor my writing so that it can say, “Yeah, Rich spent four, six, eight hours on this blog post or tons of edits. This was obviously done by a human being. Stamp of approval.” Do I have that part right?
Madeleine: Yeah. And so that would indicate if you ran this through originality and it flags you as potentially using AI to generate that piece of content, then you have proof that it wasn’t, right? And so that’s kind of why we built that tool.
Rich: And I love that. But for those of us who maybe haven’t used the tool and wrote a piece of content that for whatever reason gets flagged, not necessarily by your tool but by some other tool, what tips do you have for content creators to make content that would pass muster? Like what do humans bring to the table that AI can’t yet, that would really help humanize our content and help us out in this world that we live in today?
Madeleine: Spelling mistakes. That’s honestly, and I don’t advocate for that poor grammar and spelling. But that is kind of what differentiates us at this point.
Rich: Imperfections.
Madeleine: Imperfections, exactly. But more casual and conversational type dialogue within your content is what we’re seeing as more human. But yeah, it sucks when you’re writing a scientific paper, and obviously you’re not going to write that in a casual way, and you’re going to have very scientific language. That’s where AI detectors can fall short and flag you.
It’s going to be very uncommon for a review article or a review blog post to get identified or classified as AI written if it was human. Those are pretty clear use cases that we’ve nailed down very accurately. The gray areas are more of the scientific stuff, the medical stuff, the stuff that has a little bit more technical language around.
Rich: And outside of the content creators that may want to be able to show their work, show that they’re human to their editors or to their clients, what type of businesses come to Originality AI? What are they looking for in terms of this AI detection?
Madeleine: Yeah, so I would say our biggest clients are big content marketing agencies. And I think we did that very purposely. We speak their language, we’ve been in their seat before for many years. We were able to speak to that industry very, very clearly and understand the pain point from a very personal and intimate standpoint. But other companies that have come to us recently are large publishing houses.
I think we’ve probably all heard of instances where books are being published and they’re coming out later that they’re actually AI generated and that they might be giving out dangerous advice. One is coming very clearly to mind to me right now. There was actually an eBook, it was I think mushroom cookbook. A foraging book. We made a comment on that actually with The Guardian. And basically it was this foraging book that gave really, really dangerous advice to people about cooking with mushrooms. And those have real societal consequences, right?
And so big publishing houses are really scared to inadvertently publish something that’s AI generated. But so the whole process around what actually ends up getting published is changing, and there’s a lot more safeguarding that process at this point. So hopefully we’ll see less of that. But yeah, we’ve seen a lot of issues in recent times with that stuff.
Rich: All right. For people who want to learn more about you, Madeleine, or learn more about Originality AI, where can we send them?
Madeleine: Yeah, you can send them to our website, originality.ai. We’ve also got a really great LinkedIn page where we share all sorts of case studies that we’re working on. And yeah, we’re also on Twitter, on Facebook, all sorts of places.
Rich: Awesome. And we’ll have all of those links in the show notes. Madeleine, thank you so much for coming by and sharing your story with us today.
Madeleine: Thank you so much for having me.
Show Notes:
Madeleine Lambert from Originality.ai is an expert in accurately identifying plagiarism and AI detection in content. Check them out online, especially their Linkedin page, where they share case studies and other fascinating info.
Rich Brooks is the President of flyte new media, a web design & digital marketing agency in Portland, Maine, and founder of the Agents of Change. He’s passionate about helping small businesses grow online and has put his 25+ years of experience into the book, The Lead Machine: The Small Business Guide to Digital Marketing.