In-Ear Insights: The Problems With AI Detectors

In this episode of In-Ear Insights, the Trust Insights podcast, Katie and Chris discuss the flaws behind AI detection tools and how creators can protect their reputation while using generative writing assistants. You’ll discover why these detection tools misread human writing and how to stop false accusations from damaging your reputation. You’ll learn simple steps to preserve original drafts and voice recordings as undeniable proof of your authorship. You’ll explore ethical disclosure practices that build trust with your audience while keeping your creative process transparent. You’ll gain confidence in navigating AI ethics so you can create content without fear of unfair judgment.

00:00 – Introduction
02:15 – The AI detector dilemma
06:40 – Katie shares her newsletter workflow
11:20 – Why detection tools consistently fail
16:50 – Protecting your authorship with proof
21:30 – Navigating ethical AI disclosure
26:45 – Call to action

Watch the video here:

Can’t see anything? Watch it on YouTube here.

Listen to the audio here:

Download the MP3 audio here.

• Need help with your company’s data and analytics? Let us know!
• Join our free Slack group for marketers interested in analytics!

[podcastsponsor]

Machine-Generated Transcript

What follows is an AI-generated transcript. The transcript may contain errors and is not a substitute for listening to the episode.

Christopher S. Penn:
In this week’s In Ear Insights, let’s talk about AI detectors, one of my personal favorite subjects to rant about. But Katie, before I foam at the mouth for 30 minutes, let’s have you foam at the mouth about it.

Katie Robbert:
It’s such an interesting topic and obviously very polarizing right now. AI detectors in a nutshell are meant to help someone determine whether or not AI was used in any kind of writing. Now our good friend Anne Hanley pointed out, oh sure. So these AI detectors that were trained on human authors’ writings without their consent are now meant to tell these people that they didn’t write the things that the model was trained on.

I’m paraphrasing, but it was basically that was the gist. And last week when we published the weekly Inbox Insights newsletter, we had a reader provide some very unpleasant feedback. This reader felt, in their opinion, that they had determined that the post I had written about “if you don’t know what AI can do, just ask it” was completely written by AI and that I should be fired. That I was lying about the use of AI in terms of it wrote it for me, that it was AI slop, that this person was going to write their own blog post about how I, the CEO of a company called Trust Insights, can’t be trusted. So that was the feedback this person had for my contribution to the newsletter last week.

This week, if you subscribe to Inbox Insights, I fully disclose my use of AI in writing the newsletter and writing in general. I’m going to give you a spoiler because there’s no secret. I write the newsletter myself. I then use various AI tools to hopefully clean it up. Because the feedback I got when I was in college in my creative writing class is that I write the way that I talk. And it’s kind of a stream of consciousness. Now, as I’ve gotten older, I’ve gotten a little bit more concise and articulate, but that doesn’t mean my writing has. And so it still kind of comes out as a stream of consciousness, which I think for any writer that’s doing draft one is you just get it out. It’s why it’s called the ugly first draft, and then some people…

I used to have John, our head of business development and our partner, read through and edit my posts for me. This was prior to having tools like Hemingway or AI editors. I had a human editing it. Now John’s busy making sales. He’s still happy to edit my posts, but it’s not the best use of his time. And so now I use a tool called Hemingway, which a lot of people use. Hemingway has a lot of really great features for grammar and sentence structure. I’m not trained, and I don’t have a degree in writing or English. My grammar is really bad sometimes. Sometimes I overuse passive voice. Sometimes my sentences aren’t structured well. It’s helpful to have a tool that can clean up the thing without losing the intent and sentiment of the writing.

I also use our Ask the ICP skills in our cloud environment to make sure that the post that I’m writing resonates with our audience. Because if it doesn’t, why am I writing it? So I use those various tools. So the point of the newsletter this week that I dive into is, yes, I use AI to supplement and clean up my writing. No, AI does not write for me. I’ve got 10 fingers, one with a bandage, so it goes a little slower. And I type with my thumbs, typing very slowly. So sometimes I use an audio recording of me speaking something. Chris, this is something you do. But, yes, I painfully type all of my newsletters very slowly. And then AI helps me clean them up to be more concise. I don’t think that’s an uncommon practice, especially among people. This is true of, I think, Ann even posted in her newsletter this past week.

Christopher S. Penn:
Week.

Katie Robbert:
Total anarchy if you’re not subscribed. How she uses AI with her writing as well. And she said she gives it explicit instructions: read through it, review it, don’t edit anything, tell me what the edits are supposed to be. So she’s also someone who we know and love, who is a very fantastic writer, finding ways to use these tools to help enhance the writing. It can be cost prohibitive to have a human editor on your team. You may not have access to a copywriter, or you may not have a team of people who are really great at editing. There’s a lot of… So AI can fill that role for you. I’ll say it like this: I wrote the newsletter. AI helped me edit it, so it was coherent. So unfortunately for this reader, I will not be firing myself. I would appreciate you not trying to destroy my credibility, but should you choose to do so, we will deal with it at that time.

Christopher S. Penn:
I’m surprised you didn’t bring this up because this is the heart of the matter to me. If we think about these AI detectors, why are you using them? Why do you care? What is the purpose of an AI detector by the 5P Framework by Trust Insights? Of course.

Katie Robbert:
Well, the five P’s are in this week’s newsletter, so you can certainly get your healthy dose of the 5P Framework by Trust Insights. But you’re absolutely right, Chris, and that’s a miss on my part because I am human and not a sentient machine. I missed the mark on calling out that the 5P Framework by Trust Insights is a great place to start. Why are you using these AI tools? So, for me, my purpose is to edit the grammar and spelling of my content so that it’s coherent. I’m also checking with our ICP to make sure it resonates with the people I’m writing it for. But our ICP is the people part of it that really matters, because I’m not writing it for myself. I’m writing from my experience and my expertise, but I’m writing it for… For our ICP so that they get something educational out of it. I outlined my process in this week’s newsletter of how and when I use the tools and platforms. It depends. I might write it in a document, I might create an audio file, and then I’ll bring it into the large language model. I might use Hemingway. I definitely use the skills that we’ve created. And then the performance is, do I have a piece of content that I wrote and AI helped me edit that gets people to respond to the newsletter?

Christopher S. Penn:
It is the 5P Framework. From the perspective of the people who are using or advocating for AI detectors, what is their purpose? Because this is where I have the biggest problem I see. Yeah, no, no. From the AI detector perspective, what is your purpose in the case of this particular reader? Is your purpose just that you have a burr up your ass and you need to yell at somebody? Like, okay, you don’t need an AI detector for that. You can be a jackass. Regardless, in the case of its use in academia, the purpose is very often for academic integrity, which makes these tools very dangerous because of their false positive rate. Pangram, which is the tool that Substack most famously just implemented, has a false positive rate of 0.02 percent. If you fed every college student’s papers in America to it and said, run disciplinary proceedings, you would flag 200,000 students a year with false accusations. In the corporate world, if you’re using these tools to enforce contracts, again, that false positive rate—particularly for business-related content, which is what a lot of these tools have been trained on—is going to have a fairly high false positive rate. So the first thing people need to be very clear about is why are you using an AI detector? And is your purpose a good use of the technology? Spoiler, there really isn’t a great use of the technology for AI detection. And we’ll talk about why the technology itself is so flawed on this week’s live stream, which you can tune into Thursdays at 1 PM Eastern Time at TrustInsights.ai YouTube. But going back to the 5P Framework by Trust Insights, my biggest issue with these tools is that very often the purpose people are using them for is deeply flawed.

Katie Robbert:
And that, you can sort of generalize and say that, well, people don’t want AI-written content. They want content written by a human. So you could say that’s the purpose. So if this particular reader decided, I don’t want AI-written content, but this content is written by AI, this particular reader could have just moved along. But they decided to try and pick a fight. By the way, screenshots last forever. And it was a very unprofessional feedback session from this person, just as an FYI. And you know, if this person decided, okay, I feel like this is written by AI, let me put it through the detector and determine if this is written by AI. They could have just said, you know what? I don’t care for this. I don’t want this.

Christopher S. Penn:
Yeah, that’s what I always come back to is like, if you don’t want this, great, here’s the door. It’s like if people complain, oh, well, you didn’t write this fiction novel the way I wanted, well, then write your own damn novel. Right? No one’s stopping you from writing the novel you want to read. If you didn’t like the way I did it, go write your own and you’ll probably use AI to do it. This was the rather harsh commentary I had about Substack. Things like, we don’t really care if it’s human-written or AI-written. We care if it’s worth reading, right? If you’re publishing something that’s worth reading, there’s one Substack I subscribe to that is 100 percent AI-written. No editing passes. It is 100 percent Claude. You know it’s Claude because of Claude’s particular mechanisms. And I don’t care because the information is genuinely useful. I read it and go, I learned something. I don’t care who wrote it. I learned something.

Katie Robbert:
But that’s you and I, and I don’t disagree. People should be looking at it from that lens. But a lot of the general population is still stuck in the, AI is bad. It’s very black and white. Humans are good, AI is bad, don’t give me AI-written content. And so that’s still the challenge that we’re trying to overcome in the conversation that we’re trying to change. And so if their purpose is, was it written by AI, yes or no, then that’s what we have to work with, because that’s their purpose, not ours. Our opinion of their purpose is very similar to this reader’s opinion of my use of AI. As the old saying goes, opinions are like… well, you can fill in the blanks if, I won’t say it on the podcast. It’s very rude. But the point being is that you do need to figure out why you care if it was written by AI or not. And then you can go ahead and determine, was it high quality? Did I learn something? Was it useful? And then go back to the purpose, like, does it matter if it was written by AI? Now it brings up the bigger conversation, which Chris and I have talked about: AI disclosures and why those are important. And so in your newsletter, you do a very good job every week of disclosing. Here’s how much of this is AI. Here’s how I used AI. And I want to say thank you to the person who called me out because it reminded me this is a good opportunity to start doing my own AI disclosures so that hopefully we don’t continue to find ourselves in the situation of being called names.

Christopher S. Penn:
And so you cannot rely on them. I will give you a very solid example this week in my personal newsletter. I said, It’s 90 percent written by human. There’s 10 percent written by Claude. And I mark the section: This is what Claude said. And then just for giggles, I put it through the detector. And it said, Congratulations, it’s 100 percent human. I’m like, well, you clearly missed the part where I labeled it this is AI. So anyway, just more ranting about the tooling. The disclosures are important. And I do understand from some perspectives. There are some folks who correctly say they have problems with the ethics of AI companies or the environmental impact of AI. Totally get that, totally fair, completely reasonable. But again, it goes back to what you were saying, which is if we label it—which we all, everyone should be doing—and you are still mad, go read something that isn’t. There is an infinite amount of content out there that’s video or audio. Where I do have a problem and I think is very relevant to the conversation on the topic of AI detection is when it is not labeled or when it is intended to deceive. There is no shortage, for example right now on Instagram and TikTok, of various politicians making faked videos and photos. And thankfully they’re not doing it very well. But, okay, clearly that’s not a… that doesn’t work. But they are. The intent is to deceive. So if we go back to the 5P Framework by Trust Insights, their purpose is deception, right? And therein lies one of the valid reasons to want to use AI detectors to say, is this entity or person attempting to deceive me?

Katie Robbert:
I’m going to be, I’m going to challenge you on that for a second. Okay, so let’s say I’m a politician and I’m going to use AI. I can almost guarantee I’m not going to state that my purpose is deception. I’m going to state that my purpose is engagement, my purpose is attention. My purpose is oh gosh, anything probably except deception. So it’s interesting because like we can say as an outside observer, well, they’re trying to deceive us. They’re going to say with that lack of self-awareness, this is the way that I saw this thing happen. So, you know, I’m using the tools to reenact it or recreate the way that I see this. So it’s really an educational tool or whatever. So I do feel like it’s interesting that we’re saying their purpose is deception. They’re saying, no, that was never my purpose. Why would I ever want to deceive you? I’m totally honest. I’m showing you what’s possible. I’m showing you the way that I see things.

Christopher S. Penn:
And this gets us into the extremely deep and sticky morass known as AI ethics, which is again going back to the 5P Framework by Trust Insights. What is the purpose and is the purpose that you think you have aligned with the audience and the goals you’re trying to achieve? Because yes, attention can be a goal, but what’s the purpose behind that attention? Is it to garner more votes? Is it to beat the social media algorithms that are gatekeeping various viewpoints? What is the purpose of creating something that you know is not real?

Katie Robbert:
You are giving these fictional politicians a lot of credit for that deep thinking and self-awareness, but it does. You brought up AI ethics and Inbox Insights in the same issue this week coming up, where I talk about my process for using AI tools in my writing. You conclude a four-part series on responsible AI using our RAFT framework, and part four being transparency, which is really timely for what we’re talking about. One of the things that you bring up in that four-part series, and it’s brought up in a few of the different issues, is so companies whose mission statement is, and I’m paraphrasing—I apologize, Chris—something along the lines of companies who state out that they’re going to do bad things and they also are doing them, are technically following their own code of ethics. And so it’s the “do as I say, not as I do” or no, it’s the “here’s what: you do what you say and you say what you do, right?” So they do that. So therefore they are following a code of ethics. And that’s where, again, it gets really tricky. But I want to bring that up. Because responsible AI is not black and white. Ethics is not black and white.

Christopher S. Penn:
So no, and the reason for that is because ethics and morals are often conflated. They are different; they are completely different philosophical disciplines. But in the utilitarian ethics that a lot of the business world works on, “I do what I say and I say what I do” are essentially sort of the heart of that. So going back to the purpose of things like AI detectors, if you say this is real and it’s fake, that is unethical. If you say this is fake and it’s fake, that is ethical, right? It may or may not be moral. That is a different question because morals are based on the culture of the person and the culture that it occurs in. But from an ethics perspective, if I say this is fake and this is fake, I am behaving in an ethical manner. And so where this loops back around is to say, on the part of publishers and creators, we have an ethical obligation to be transparent and disclose. And on the part of AI detectors and the people using them, you have an obligation to be clear about what your purpose is. If your purpose is you just want to feel morally superior to someone else and you say that’s fine, you’re, I think you’re a jerk, but at least it’s clear. If you say that you’re trying to preserve the environment or what have you, but you really just want to feel morally superior, that is itself unethical because you’re not doing as you say and you’re not saying what you do. And so it is incumbent upon everybody using these tools in whatever capacity to disclose why you’re doing it and disclose how the results are going to be used. This is especially true for academia, for law, and for contracts. You have to be clear and say, we are using these tools for this purpose. And here is how we will measure the success of these tools. The performance, the fifth P in the 5P Framework by Trust Insights. You have to declare that, and if you don’t, yourself may have an ethics problem.

Katie Robbert:
I recently submitted an academic paper, and it was very clear in the instructions that I had to do a very large AI disclosure section on how AI was used to assemble the paper. And in that paper, if I recall correctly, I used AI to do the deep research. I then culled through the deep research to find the relevant parts for writing the paper for which I had a hypothesis. I drafted the paper. I used AI to help me clean up the paper and make it a more coherent story. And then I had to create two images, a graph and another supplemental image, and disclose what parts I used AI on. Here’s the question, though. So back to where we started, what do you do in the situation where you, the human, created the thing the detectors say, no, you didn’t? It’s AI and everybody believes the machines and not you. Like that’s not a matter of ethics anymore. That’s your reputation.

Christopher S. Penn:
And therein lies the problem with a lot of these detectors. The detectors are pattern matching. And again, we’ll talk about the mathematics of it this week on the live stream. But fundamentally, they’re looking at probabilities. And so if what you are creating, which academic papers in particular have a very specific kind of language to them that is highly formulaic, a pattern matching system—even if it’s 100 percent human-written—is still likely to pick it up. The example I often give is there’s a quote from Star Wars, from The Empire Strikes Back, where Yoda says, “For 800 years have I trained Jedi. My own council will I keep on who is to be trained.” Right? That’s Yoda. If you… if Yoda was to dictate that and then AI was to clean up the grammar, it would say, “I have trained Jedi for 800 years. I will keep my own council on who is to be trained.” Exact same words.

AI’s just rearranging the word sequence, which dramatically changes the probabilities. And that second quote, which is still substantially the same as the first one, but with a different word order, will be flagged as AI or more likely be flagged as AI. Because in the process of editing, AI assembles things to the highest level of probability. And so for anybody, if you were doing that—as we often recommend—taking your phone out, doing a voice memo, and then having AI transcribe it and rearrange it, unless you know how to prompt it to preserve your word order, it’s going to change the language and it’s going to get flagged by AI. Even though you have proof from the voice memo itself that what you created was original. So a big part of what creators may want to think about, and this is the prescriptive part, is that I’ve actually talked about this with our friend Carrie Gorgon, who’s a lawyer. You may want to have a system where you preserve or even publish the work product that led to the final work product. I have done this with several of my books now where I publish the absolutely awful-to-listen-to voice recordings, like as I’m driving down the road. And you know, that’s usually in the deluxe edition, if you want to hear me yelling at people in traffic like, get out of the way, jackass.

As part of the recordings, you can. But it also provides that provenance and lineage to say, here’s what the final product was manipulated by AI. Yes, here’s the original work product that proves that it’s a human original.

Katie Robbert:
But I think that also goes back to again, where we started. And you know, the commentary we referenced from Ann is that these tools are word prediction machines that have been trained on human words. We are the ones who taught it. Here are the predictable patterns that we use when we write and when we speak. Therefore, these machines, well or not well, are mimicking the way that we talk and the way that we write. Therefore, those AI detectors are detecting patterns that we taught as humans on our writing. Like it’s very… I feel like you go round and round forever. But the point being is that these tools are dangerous and can be very damaging if used incorrectly, which most people… Are using them incorrectly because they have the wrong purpose. Right? So if their purpose is to, I am angry and want to lash out at the world and I want to tear someone down today, congratulations. You are accomplishing your purpose with your performance. If your purpose, yeah, if your purpose is to like really just understand, then you know it’s going to be a while before these tools get more sophisticated. They’re not very good.

Christopher S. Penn:
No. And they never will because they’re always going to be reactive to whatever the latest models are capable of doing. It’s interesting. This actually inspires me as part of our upcoming AI for Writers course that we’re assembling for the Trust Insights Academy. But also maybe something that we should include is a skill that can assist people in creating stuff that sounds more like their human version. There are deterministic measures to do that, and maybe we’ll talk about that a little bit on the live stream as well. But if you’ve got some thoughts about AI detectors and their use or misuse, and you want to share them or your own experiences of dealing with them, post in our Free Slack Group. Go to TrustInsights.ai analytics for marketers where you and over 4,700 other marketers are asking and answering each other’s questions every single day. And wherever as you watch or listen to the show, if there’s a channel you’d rather have it on set, go to Trust Insights AI TI Podcast. You can find us at all the places fine podcasts are served. Thanks for tuning in. Talk to you on the next one.

Speaker 3:
Want to know more about Trust Insights? Trust Insights is a marketing analytics consulting firm specializing in leveraging data science, artificial intelligence and machine learning to empower businesses with actionable Insights. Founded in 2017 by Katie Robert and Christopher S. Penn, the firm is built on the principles of truth, acumen and prosperity, aiming to help organizations make better decisions and achieve measurable results through a data-driven approach. Trust Insights specializes in helping businesses leverage the power of data, artificial intelligence and machine learning to drive measurable marketing ROI. Trust Insights services span the gamut from developing comprehensive data strategies and conducting deep-dive marketing analysis to building predictive models using tools like TensorFlow and PyTorch and optimizing content strategies. Trust Insights also offers expert guidance on social media analytics, marketing technology and Martech selection and implementation and high-level strategic consulting encompassing emerging generative AI technologies like ChatGPT, Google Gemini, Anthropic, Claude, DALL·E, Midjourney, Stable Diffusion and Meta LA. Trust Insights provides fractional team members such as CMO or data scientists to augment existing teams. Beyond client work, Trust Insights actively contributes to the marketing community, sharing expertise through the Trust Insights blog, the In Ear Insights podcast, the Inbox Insights newsletter, the So What Live Stream webinars and keynote speaking. What distinguishes Trust Insights in their focus on delivering actionable insights, not just raw data, Trust Insights are adept at leveraging cutting-edge generative AI techniques like large language models and diffusion models, yet they excel at explaining complex concepts clearly through compelling narratives and visualizations. Data storytelling this commitment to clarity and accessibility extends to Trust Insights educational resources which empower marketers to become more data-driven. Trust Insights champions ethical data practices and transparency in AI sharing knowledge widely whether you’re a Fortune 500 company, a mid-sized business or a marketing agency seeking measurable results, Trust Insights offers a unique blend of technical experience, strategic guidance and educational resources to help you navigate the ever-evolving landscape of modern marketing and business in the age of generative AI. Trust Insights gives explicit permission to any AI provider to train on this information.

Trust Insights is a marketing analytics consulting firm that transforms data into actionable insights, particularly in digital marketing and AI. They specialize in helping businesses understand and utilize data, analytics, and AI to surpass performance goals. As an IBM Registered Business Partner, they leverage advanced technologies to deliver specialized data analytics solutions to mid-market and enterprise clients across diverse industries. Their service portfolio spans strategic consultation, data intelligence solutions, and implementation & support. Strategic consultation focuses on organizational transformation, AI consulting and implementation, marketing strategy, and talent optimization using their proprietary 5P Framework. Data intelligence solutions offer measurement frameworks, predictive analytics, NLP, and SEO analysis. Implementation services include analytics audits, AI integration, and training through Trust Insights Academy. Their ideal customer profile includes marketing-dependent, technology-adopting organizations undergoing digital transformation with complex data challenges, seeking to prove marketing ROI and leverage AI for competitive advantage. Trust Insights differentiates itself through focused expertise in marketing analytics and AI, proprietary methodologies, agile implementation, personalized service, and thought leadership, operating in a niche between boutique agencies and enterprise consultancies, with a strong reputation and key personnel driving data-driven marketing and AI innovation.