Everyday AI Podcast – An AI and ChatGPT Podcast - 3 AI Lies Most People Believed In 2025 (But You Shouldn’t)

Starting point is 00:00:00 This is the Everyday AI Show, the Everyday Podcast where we simplify AI and bring its power to your fingertips. Listen daily for practical advice to boost your career, business, and everyday life. Meet Firefly AI Assistant, now live in Adobe Firefly, the All In One Creative AI Studio. Just describe what you want to create and the assistant handles the rest, orchestrating multi-step workflows across Photoshop, Premiere Express, and more in one conversational interface. You direct the outcome. The assistant accelerates execution. You've been lied to about AI a lot in 2025.

Starting point is 00:00:50 It's not your fault. There's so much money to be made in these muddy AI waters. Even reputable organizations have squatted to all-time lows to basically lie to you and try to make a dollar. Why? Well, because with a single viral study, markets can move in billions. of dollars can be made. So on today's show, I'm going to break down the three biggest AI lies you were told this year that you may have believed.

Starting point is 00:01:20 And I'm going to slap a few of these silly studies around with facts and stats and let you decide. Were these three organizations who created these viral AI studies telling you the truth or intentionally trying to mislead enterprise leaders in order to cash blank checks? Let's dig in and find out. I'm excited for today's episode. I hope you are too. Time for some hot take Tuesday, y'all, on Everyday AI. What's going on? My name's Stuart Moulson. Welcome to Everyday AI. If you're new here, we do this every single day, Monday through

Starting point is 00:01:50 Friday at least. It's a daily unscripted, unedited, unedited, live streamed podcast and free daily newsletter helping everyday business leaders like you and me make sense of all this AI because it's coming at us nonstop and what can you even believe anymore? All right. And I help you decipher it and grow your company and your career with the actual facts and stats that you need to know. So if that's what you're trying to do, awesome. It starts here. But to take it to the next level, make sure to go to our website at your everyday AI.com. Sign up for the free daily newsletter.

Starting point is 00:02:21 We're going to be recapping today's show, but also on that very website, you can go listen to for free hundreds of hours. Yeah, 660 plus episodes from some of the brightest minds in AI interviews with people from all over the place, all on our website for free. All right. And also, if you want today's AI news, make sure to check out today's newsletter. So, yeah, you've been told a lot of lies about AI in 2025.

Starting point is 00:02:50 And aside from the money factor, because literally billions of dollars can be made with a single AI study or a single article that maybe is false or is cherry picking certain facts to make a certain trillion dollar company. look a certain way or to say, hey, this AI startup's product is better than the other. It can literally move markets. It's not your fault, though, because not only are the AI waters muddy because no one, or I would say hardly anyone is an expert in the generative AI era, right?

Starting point is 00:03:26 So unless you were an early employee in, you know, 2015 or 2014 at Google or Open AI, hardly anyone has any real longstanding expert. in AI. So when you see all these studies from what looks like reputable organizations, you just kind of blindly believe them. And so does the media. And so do people on social media. Right. So essentially these viral headlines just get to trumpeted and, you know, regurgitated. And everyone just believes them. But it's not the truth. I think, unfortunately, AI has become very polarized. It's become binary, right? It is hit or miss. It is. is doom and gloom or miracles.

Starting point is 00:04:10 And there's often nothing in between. So on today's show, we're going to audit the $100 million conflict of interest behind the viral open AI is collapsing narrative. We're to deconstruct the faulty science claiming that 57% of the internet is now AI slop and expose the hidden definitions that label successful innovation in AI as 95% failure. Yeah, you know which one that is MIT. All right. So here are the three myths that we're going to be tackling today. So myth number one, that Anthropic has overtaken Open AI in the enterprise. Yeah, that was a study from Menlo Ventures.

Starting point is 00:04:48 Doesn't make sense. Myth number two, the dead internet, right, is 57% of the web really AI generated. That was from graphite, spoiler alert. There's no way to prove it. And myth number three, not that you needed to hear me talk about this again, but y'all, I still still. I still am seeing these posts on social media and even new articles referencing this all the time. The 95% failure rate of Gen AI pilots study from MIT. So yeah, if 95% of AI pilots are failing, why are the smartest and biggest companies in the world spending hundreds of billions of dollars? I don't know. So this, I think, and I think it's important to talk about.

Starting point is 00:05:37 this here, you know, as we are in December in the last month of the year, because it's important to reflect on the conversation, the overall conversation of AI. And so many times, right, it's, it's predicated or based around a couple of things, right? Usually new model releases, benchmarks. That's one thing. And then the other is just the narrative. And like I said, it is binary. It is either, you know, AI is the best thing ever or AI is absolutely terrible. And usually there's some weaponized studies that go either way, right? And they all, start with real data. The data gets massaged and cherry picked, obviously, and, you know, try to put into the best light possible. So we're going to break down some of those three biggest

Starting point is 00:06:20 AI studies or research papers that have dominated a lot of the conversation in 2025, but they're wrong. Right. So I think it's important because we need to find the middle ground, right? In AI, there's there's, there's, there's so. few things that are binary, right? Even as myself, someone that talks about AI every single day, I don't think AI can do everything, right? But I also know that AI is very capable. So I think it's important to have a balanced look,

Starting point is 00:06:50 but, you know, laced with straight hard facts. Because it's, it's every day. It's, hey, AI changes everything to, oh, AI is a scam. AI is a bubble, right? And the truth is actually just much messier. It is nuanced. And, well, unfortunately, if you write a start, that is factual and you try to give it a headline that is balanced, it's not going to get clicks,

Starting point is 00:07:14 right? It's not going to drive the narrative. I'm a former journalist, right? There's a, there's a story or a saying that if it bleeds, it leads, right? You don't find the, the median of your study's findings. You find the outlier. And then you slap it with the sexiest headline possible, right? That you're like, hey, this is going to be the headline that sells this newspaper. This is going to be the headline that gets us clicks, that gets us ad revenue, that puts us as a thought leader in this space, right, that gives us sales. So, let's look at lie number one or not truth number one, right? We'll say that. Menlo Ventures, their report that claimed that Anthropic is beating Open AI in the enterprise. Oh my gosh,

Starting point is 00:08:01 people believe this. Smart people. Literally, I saw dozens of smart. are people that I once respected, you know, tweet this out or put this out on LinkedIn or write long, medium articles about this. And I'm like, some of these articles were almost as long as this study. Yet they just glossed over the fact that no one should even like look at this almost as bad as the MIT study, which I'll obviously get to. But this is laughable, y'all. Anyways, Menlo Ventures, right, a venture fund. In July, they had their mid-year, L.O. market update and it said Anthropic overtook Open AI in enterprise market share, right? I said that Anthropic now holds a 32% and Open AI is 25%.

Starting point is 00:08:47 And, you know, people started talking about this flippinging, right? Because before Open AI was in the lead and now Anthropics in the lead, oh my gosh, and everyone just literally copy, paste, repeat of this staff. I swear humans don't like to use their brains anymore. You know, it's the social medification of our brains. our brains have literally become rot because all people care about they they look at a graph and a headline and then they'll just repeat it use your brains right this this study is garbage and apologies to people in garbage i have friends in garbage who are brilliant um so to call

Starting point is 00:09:24 this study garbage is doing a disservice to those people uh so let's look at this uh the the big um kind of graphic that caught everyone's attention right it's it's it's labeled here. So live stream audience, you'll see this on my screen podcast audience. Nothing super visual in today's show, but you can always go watch the video version on our website, your everyday AI.com. So this graphic showed in 2023, obviously, you know, the enterprise, large language model API market share. You know, so in 20203, Anthropic was at 50% or sorry, Open AI was at just over 50%. And Anthropic was at like 12%. Right. And then in 2025, according to Menlo Ventures, it flipped, right? So what was the exact stat? It was now Anthropic holds 32% versus OpenAI is 25%. So the headlines all read, Anthropic overtakes Open AI in Enterprise. But no one read it. All right. Also, did no one care to bother to see like, oh, when did these companies launch? Right. Open AI. I

Starting point is 00:10:34 obviously launched in 2022, Anthropic in mid-20203. So, yeah, of course, by default, Anthropics market share is going to be much smaller when their product wasn't even out the entire year. Anyways, facts would hurt our brains, apparently. But here's the important part. This Menlo report is not neutral.

Starting point is 00:10:54 It's promotional. We need to look at who Menlo Ventures is. Well, they're a venture fund, and they have invested billions, billions, into Anthropic. Adobe just introduced an entirely new way to create, bringing the power and precision of its creative suite into one conversational experience.

Starting point is 00:11:20 Meet Firefly AI Assistant, now live in the Adobe Firefly app, the all-in-one Creative AI Studio. Powered by Adobe's Creative Agent, Firefly AI Assistant lets you start with your vision, just describe what you want, and shape the outcome as it takes form with the Assistant. The assistant orchestrates multi-step workflows, drawing on 60-plus pro-grade tools across Adobe Creative Cloud apps, including Photoshop, Illustrator, Premiere, Lightroom Express, and more to help bring your ideas to life.

Starting point is 00:11:51 You can also get started with creative skills, a growing library of pre-built workflows for common creative tasks, like batch editing photos, creating mood boards, portrait retouching, and creating social variations. Every step the assistant takes is visible, so you can refine, redirect or take over at any time. You stay in the driver's seat as the creative director. Adobe Firefly AI assistant now in public beta. See it today at firefly.adobie.com. Let me get the exact math here.

Starting point is 00:12:26 Let me fact check myself here in real time. So okay, no, sorry. Over reportedly over 1.2 billion dollars. So billion plus. All right. But regardless, Memo Ventures is a huge, financial supporter in Anthropics. So they obviously have a lot to game. And Menlo Ventures has publicly declared that Anthropic is the single largest investment in the firm's history. And they led Anthropics

Starting point is 00:12:53 series C, D, and E funding round, which was billions of dollars. So this obviously is just an investor talking up their own book of business. And not only that, but Menlo and Anthropic actually co-created the anthology fund specifically to fuel anthropic ecosystem adoption. So startups in this fund receive $25,000 in free Anthropic credits and $100,000 in AWS credits. And guess what it's really easy to use on AWS and especially in 2023 and 2024? Oh, Anthropic. So essentially, they're paying companies in this anthology fund to adopt Claude. And guess who they surveyed?

Starting point is 00:13:37 You'll never guess. People in their fun, in their joint fun. All right. Yeah. So this entire claim came from a survey, a single survey where they talked to 150, yeah, 150, 150 technical decision makers. But they were pulled from Menlo's own venture capital network and portfolio ecosystem. So, yeah, is selection bias so obvious?

Starting point is 00:14:06 It's embarrassing that anyone even. reported on this or talked about it because this is clearly marketing, all right, bad marketing, by the way. But people fell for it. And here's a couple important details, right? Aside from the fact selection bias at its finest or at its worse and the fact that it was 150 people, well, it only looked at API usage, right? It sounds super comprehensive, but it's not because it completely ignores Microsoft

Starting point is 00:14:38 co-pilot, which guess what? Up until four months ago, was built almost entirely on Open AI's models. So they really overlooked the entire enterprise. At the time when this study came out, you know, Microsoft co-pilot had

Starting point is 00:14:54 the largest slice of the enterprise. So they ignore the fact that OpenAIs models power Microsoft co-pilot. Obviously, they use Anthropic in some instances now. And they just cherry-picked API calls only. Oh, and the fact that chat tbd enterprise is one of the fastest growing software products ever and they have more than a million

Starting point is 00:15:16 business users which anthropic obviously does not so again uh they rigged the result of this before it was even produced uh so let's just go ahead and say how this is disguised as market research but it's just marketing so menlo invest uh more than a billion dollars in anthropic it then creates a one million dollar fund to accelerate Claude adoption. They give startups a bunch of money to use Claude. And then they survey those same people. And then they announce, oh, Claude is winning.

Starting point is 00:15:50 Come on. This is trying to make the market not measure it. All right. Talk about revenue, right? Open AI has more revenue than Anthropic. They have more customers. They have more users. So no, there is no flippinging.

Starting point is 00:16:04 No, Infraffic did not overtake Open AI. in any measurable overall scope of enterprise usage. It's not even close. So just FYI, if you believe this, if you saw this, again, it's not your fault. But if you saw anyone talking about this, go ahead, unfollow them, delete them, right? Stop. You need to stop paying attention to people that don't want to read and don't want to understand facts. Because independent data and anyone with the brain knows that open AI is still leading overall enterprise adoption.

Starting point is 00:16:36 So enterprise from, sorry, data from Ramp from October showed that 93% of organizations who have a vendor in the Gen AI category use Open AI and that Open AI maintains the highest adoption among enterprise companies at 95%. So the truth is surveying 150 people from your own network has absolutely nothing to do with actual real enterprise AI usage. When billions of people use generative AI, you don't talk to 150 people in your business. backyard who you know how they're going to answer because you're paying them to use that product. Menlo's just promoting their fund and not conducting real research. All right. Line number two. Here's one.

Starting point is 00:17:19 I don't think I've actually talked about a lot on this show. The graphite study that claimed 57% of the internet is now a eye slop. All right. So this was a study in October that analyzed 65,000 URLs via the Common Crawl database. And they claimed, graphite claimed, that over half of these new articles are AI generated. So these were, you know, quote unquote, new articles from Common Call. And then the headlines all said that the internet is now sloped or, you know, data internet theory is officially here and there's more AI content on the web,

Starting point is 00:17:55 but then human written content. So let me just say this first. I don't necessarily disagree, right, with the underlying premise here. from the graphite study, right? I'm someone that's been a writer, my whole life, right? Not my whole life, but I mean, getting paid to write for 24 years, right? So, wow, saying that out loud is, that hurts to say, right? But I've been getting paid to write for almost a quarter century.

Starting point is 00:18:25 So I know how much the internet has shifted to just becoming a lot of slop now. And even before chat GPT, there was plenty of, of generative AI content going back to 2020. And then before that, you know, there's spin tax and all these other, you know, just there's pre AI slop and now it's AI slop. So don't be wrong. I agree with the premise that probably more than half of what's written now is AI slot.

Starting point is 00:18:54 However, the methodology is garbage. There's no way to decide this. Right. So here we had this, this graphic that kind of went viral that showed, uh, over time, right, since the early 2020, the percentage of human written content was, you know, near 100. AI written content was, you know, in the single digit percentage. And then now today, AI content has overtaken or surpass human content.

Starting point is 00:19:27 So here's the problem. They used Surfer SEO's AI content detector to classify these 65,000 articles. And if the detector flagged more than 50% of an article as AI, they label the whole thing as AI generated. So this entire half the internet AI claim depends on one commercial detector being accurate. Guess what? It's not. No, but this isn't anything against Surfer. Right? Surfer AI. It's actually a fantastic product. I've been using it for many, many years before I even started everyday AI. So Surfer has a great product. But any company that has an AI detection, it, it, it, it, it, it, there's no such thing. There's literally no such thing as AI content detectors for text,

Starting point is 00:20:11 right? Because there's so many different things that you can do with AI content, right? You can put it in a quote unquote humanizer, right? No, I'm a dork. So before Chad GPT came out, I literally every single AI content detector at the time, because before Chad GPT, they had their GBT technology to dozens of different companies. And so pre-chat GBT, there were even AI content text detectors. And at the time, I spent, I don't know, 30 hours going through and looking at human written articles versus AI written articles and putting them through these detectors. Yeah, they're garbage. You know, one AI content detector could say, hey, this piece is 99% AI and another one could say it's 1%. All right, they, all these AI content detectors do. You know, one,

Starting point is 00:21:01 do is they measure statistical patterns in word choice, not actual authorship. So it's a pseudoscience, not in actual sites. And the error rates are obviously massive. So here's an example right before, you know, starting the show. I just grabbed, I don't know, first part of the Bible, right? Genesis from the Bible, copy and pasted some of that. And yeah, 86% right, 86% AI generated. So yeah. Hey, biblical writers, you should have used your brain in divine intervention and not God GP.

Starting point is 00:21:41 All right. Yeah. So yeah, these AI content detectors are all garbage. There's no such thing. Right. Even right. Like you can make an argument like, oh, anyone that includes the, you know, follow up response from chat GPT. That's a telltale sign. Is it maybe? Yeah. But I could write that as a human to try to throw you off, couldn't I? Yeah, I could. So a lot of people point back to the time when even Open AI had their own AI content detector. And this, unfortunately, I wish Open AI never did this. And I said this at the time before Open AI ever took it down. I'm like, this is dumb. It does not work. Open AI needs to take this out.

Starting point is 00:22:20 They eventually did. And the reason why, well, it's because it only got it right 26% of the time, worse than a coin flip. So detectors cannot detect. So if the company that built chat GPT can't detect AI text, why should we trust any content detector? Well, we shouldn't. And it's actually bad. AI content detectors are bad because they target, unfortunately, non-native English speakers.

Starting point is 00:22:45 So detectors flag simple, structured sentences as robotic because that's how non-native speakers write. So a Stanford study found that 61% of ESL or English as a second language student essays were falsely flagged as AI generated. And a Vanderbilt University actually disabled turn it into AI detector to avoid hundreds of false accusations per year. Hey, if you are working at a college university listening to this and you are still using one of these in the year 2025, stop. Tell someone, right?

Starting point is 00:23:19 I will personally get on the phone with your president of the university and tell them that this is stupid, right? I won't even use better words than that. I will say, this is dumb. You are robbing, you know, whether it's for entry essays, homework, whatever. Yes, I know that it's a problem, right? Students are just copying and pacing everything from chatypte. They're not writing a single word, probably, a single word of anything that they're turning in,

Starting point is 00:23:45 especially if it's a hybrid or an online class where you just turn in a paper. They're not writing it. AI is, but you can't use these AI content detectors, not a thing. So here's the truth about line number two. Well, there's no concrete way to prove text is AI generated. Yes, on the video visual side, there's synth ID, there's all of these things, but as quickly as any embedded watermark in whether we're talking about in text, in video, in photos, etc, as soon as anything pops up, there's already minutes or hours later.

Starting point is 00:24:18 There's already multiple, you know, watermark removers, right? I remember when Sora came out or Sora 2 came out. and it had these watermarks, literally that same night, there was already multiple services that had watermark removers. Text is very easy to manipulate, change things, right? But again, there's no embeddable, you know, invisible watermark in text like there is in video and photos. All right. Line number three, I'm not going to go long on this one, because I've already given it plenty of air time. But the MIT Nanda study that claims 95% of AI pilots deliver zero ROI. Let's be honest, many, many, many people fell for this one and you shouldn't have.

Starting point is 00:25:04 So this was an August 2025 report from MIT. Yes, that MIT that stooped so low, they would have won a limbo contest, claimed that only 5% of AI pilots reach production with measurable, business impact. Yeah, we don't. Even when MIT comes out with a study now, we just don't share it in our newsletter because I don't view MIT as a credible source of AI information anymore, which is crazy to say. And that statistic obviously went mega viral, right?

Starting point is 00:25:36 When you can slap something on this says 95% of AI pilots fail, you know, and then you get into this whole, you know, AI is a scam or AI is a miracle, this binary, nothing in between, right? It went mega viral. And this actually spooked markets. Yes, markets moved billion to dollars because of this piece of marketing masquerading as a study. And this became the go-to proof for skeptics claiming enterprise AI is a bubble that's about to burst. FYI for all the people, you know, cold emailing, trying to get a guest on the show or just trying to sell me some AI product.

Starting point is 00:26:15 If you lead with the MIT 95% study, I don't read it, right? because I don't know. To me, that just means that you need to do some reading and some thinking. So, yeah, it was called the state of AI in business 2025, right? And here was the telltale sign. This thing was a dumpster fire before it got there. Again, sorry to my people in the trash fields. You couldn't get this report.

Starting point is 00:26:41 You had to fill out an application just to see the report. Yeah, there was a Google form. I didn't get it, but plenty of other people got it and then shared it online. And it was an absolute disaster. But when it came out, it was completely viral. And part of the reason, well, it was the media and social media, right? And that's why this show, the whole point is like, you've got to be careful what you read and what you believe. Because all of these articles that came out, I remember being an overworked journalist.

Starting point is 00:27:08 I was a journalist for, I don't know, full time for like seven years. It's hard, right? You're overworked. You're underpaid. You're always trying to not get scooped by someone else. You got to get the story out faster, faster, faster. Right. So everyone just essentially copy and paste the same headline, right?

Starting point is 00:27:25 95% of AI pilots have no return. So if you want my full hot take on this one, go listen to episode 597, because I'm going to wrap this one up quick. All right. So that 95% failure from MIT came from just 52 interviews. Oh, my gosh. We thought the one from Menlo was bad with 150. This is 52 people.

Starting point is 00:27:50 So that's 52 conversations. All right. And this is clickbait dressed up as academic research. And here's the thing. This was just a vibe study because even those 52 conversations, those were just directionally accurate. So in the paper that hardly no one read in one of the little footnotes there, it talked about the research limitations. And it said, these figures are directionally accurate, all right?

Starting point is 00:28:16 based on individual interviews rather than official company reporting. Sample sizes vary by category. How many you have categories when you only talk to 52 people? Sorry. There's like three people in each category. And success definitions may vary across organizations. In other words, this was a vibe study. It was directional accuracy only.

Starting point is 00:28:37 It wasn't even a true, like, quantifiable study. These were conversations. And then, you know, someone over there is like, yeah. Seems like they've, this failed, right? Yeah, seems like it. No, this is garbage. I need to get a better word. What's a better word than one word calling something garbage?

Starting point is 00:28:56 I don't know, rubbish, but that's just another word for garbage. Dumpster fire, garbage. Why is everything to say something's bad garbage? I need to expand on that in 2026. All right. So, well, one of the reasons why, well, because MIT had a preconceived definition of success that they knew would come out with nothing but failures. And don't worry, it gets better because they turn it into an infomercial because only they can help you.

Starting point is 00:29:21 Just wait. So success for their study, for their criteria, required measurable P&L. Yes, actual measurable movement on a profit and loss statement within six months of deploying a Gen. A.I. Pilot. My gosh, there's no such thing. Like, nothing, no major IT, no major IT pilot has, has, success in six months on a P&L statement. Are you serious? Come on. Yeah, because most enterprise transformations need at least 12 to 18 months to even show financial results, even ones that are

Starting point is 00:30:00 obvious, right? Not even AI. Like, hey, let's, I don't know, get our people computers, right? Takes years. So by definition, any new enterprise software initiative would fail by design out of the box, nothing they can do about it. That's not the worst part. So the pilot successfully prevented a bad deployment would get counted as a failure. So the study just completely misclassified what a failure actually meant. So productivity improvements that don't immediately hit the P&L, failure, right? I don't know. If you can now get something done in half the time, but you can't prove that on the P&L, failure, right?

Starting point is 00:30:37 So early stage products that needed more time to mature counted as a failure, even if they were showing signs of success, saving time, increasing. revenue didn't matter. But here's the thing. Here's the infomercial at the end. And like I see, MIT stooped to an all time low on this one. And when I read this, I kid you not, I literally put my face in my palm when I read this because MIT was literally trying to put themselves with their NANDA product, which stands for networked agents and decentralized AI, right? So essentially, they're building new AI agent protocols. So they published that 95% of current AI fails to essentially position their product as a solution.

Starting point is 00:31:20 Yeah, marketing again. So Nanda at the MIT Media Lab, if you didn't know because no one knows because no one's heard of MIT's Nanda, right? Even me who covers AI every day. I had barely heard of it by the time this study came out. I read about it like once, right? But they offer that solution to businesses for about $250,000. Yeah, covering their expertise tools.

Starting point is 00:31:44 and possible bespoke implementation. Yeah. So don't worry. Literally, MIT in this report, said, one of the reasons is, well,

Starting point is 00:31:56 we need to start thinking agentic. And they put themselves in this NANDA product, you know, in the same paragraph as some of the biggest names in agentic AI, as if like, you know, it's almost like, you know how the bottom shelf is like the cheap,

Starting point is 00:32:12 generic stuff? And then the top shelf is the nice. stuff. It's like they took something from the below the bottom shelf and they snuck it on the top shelf thinking no one would notice. Yeah, we noticed. Like you, sorry, MIT Nanda is not in the same conversation as real enterprise agentic AI, right? It's not. So classic marketing funnel, step long, publish an alarming study claiming that current AI has a 95% failure rate. Step two, you identify the learning gaps in static tools, right? They're saying that's the reason why AI's failing. It's because these static tools. And then, oh, don't worry. Step three, that's our solution. Nanda, because it's a learning capable system in new protocols. That's the solution, and you got to buy it.

Starting point is 00:32:57 So, obviously, other studies looking at AI and ROI tell a completely different story. I did this on the previous show, but yeah, real studies done by real companies that are unbiased, mostly unbiased, right? But with tons of people, people show something completely different. So the Boston Consulting Group AI at Work surveyed 10,000 plus employees, and they said that 75% of employees see value in momentum in AI. A study from McKinsey and company, 1,300 participants said that 92% of companies plan to increase their investment in generative AI.

Starting point is 00:33:36 If you want just strictly R-O-I, well, the EY study of 500 senior U.S. leaders, but an actual study, not just a vibe conversation. 97% of senior leaders who invested in AI reports experiencing positive ROI. And then a snowflake in ESG report of a thousand business and IT leaders said that 92% of early adopters are seeing a positive return on investment of AI. So yeah, any legitimate study with sound methodology, obviously, shows a positive return on investment of AI, right? So here's the truth about line number three.

Starting point is 00:34:16 It's clickbait, right? From a lab looking for customers, right? Looking to make a name for themselves. You know, 52 interviews with a six-month P&L requirement is not rigorous research methodology. It's a joke. No one, which is, again, which is why I know half the people, half of the media people that wrote about this study did not read the whole thing. Right?

Starting point is 00:34:37 They probably read the synopsis, the first page overview, the last page overview, maybe ran it through an AI and said bullet point this, right? Because if you have a brain and you read that, and you're saying, oh, six month has to show success on a P&L. And I did this in the dedicated episode. Even their selection, their selection bias was ridiculous, right? But 52 interviews is what this is based on. Directional accuracy, right? It's like I talked to you for a little bit. And at the end, I'm like, yeah, I decided that you're, you know, your AI probably does. doesn't meet our criteria next. 52, y'all, I already said this.

Starting point is 00:35:15 I can go on to parking lot at a conference or I can go anywhere in an hour and come up with a better study with more sound methodology than this piece of marketing. Yeah, because studies with 20 times, 50 times larger samples show at least 70, 80, 90%

Starting point is 00:35:33 are seeing a positive ROI versus if you look at the MIT study, it's 5%. Right. So here's the pattern, right, across these three AI lies that you've been told. So Menlo used 150 people from their own paid ecosystem while being Anthropics' largest investor ever to say Anthropic has overtaken OpenAI. Graphite used pseudo-scientific detection tools that the FTC has literally sued companies over, and MIT used 52 interviews with an impossible success bar while building a competing technology to sell as the solution

Starting point is 00:36:09 for the 95% value rate. Do you see the trend here? Yeah. Bad methodology. Bad studies disguised as marketing. So here's what I want you to do in 2026 and beyond. Number one, read. Number two, unfollow anyone that's been sharing these AI lies.

Starting point is 00:36:34 But I want you to ask yourself these three questions. Any viral study? All right. Number one, who funded the research and what do they financially gain from the conclusion? In all three of these cases, right? Graphite, a lot of people don't know. They essentially sell SEO, right? So they come out with this study and, oh my gosh, you know, Google hates AI, right? Because there are some other stats that said, you know, 80 some percent of, you know, what ranks on Google is human written.

Starting point is 00:37:01 So it's like, oh, you know, you need our SEO services, right? But all three of these companies had a very direct financial gain. from cherry-picking bad stats and selling it to you. So number one, who funded this research and what do they financially gain from the conclusion? Number two, does the study have sound methodology that an unbiased person with the brain would agree with? All right. Gosh, it's so sad that the bar is this low, but it is.

Starting point is 00:37:30 This is the truth. And then number three, what are similar studies from unbiased third parties saying about the same subject? Right? because yeah, if you look at studies that have sound methodology, that talk to a lot of people, and if you look at multiple studies, you will start to see the trend. And the trend is, well, it's undeniable. At least when, you know, ending on study number three, AI adoption is here.

Starting point is 00:37:56 And you need to be paying attention to it. And you also need to be paying attention where you get your information from. All right. So I hope today's show was helpful. And, well, I hope you continue to get your. information from everyday AI, the rest of 2025 and 2026 and beyond. We spend a lot of time, try to cut through the nonsense, the BS, the marketing, the height, and cutting it to you straight. So I hope this is helpful, but I think this episode was important because I got so many messages,

Starting point is 00:38:24 emails about these types of studies all the time. And I'm sorry, I suck in emails. I can't get to them all, right, without having an AI answer and I don't want that. But let me tell you this. I feel bad because so many well-intentioned business leaders saw some of these studies. And it probably derailed either some of their AI progress or for those that were unfortunately still kind of sitting on the entire AI implementation fence in 2025. This caused them to sit on it for even longer. You cannot believe these types of studies because they're not real. right you need to use your own brain you need to read right literally i talk to people who made huge multi-million dollar AI decisions on some of these studies and clearly they did not read it

Starting point is 00:39:20 right they did not read past the headline or the one cherry picked graphic we all need to do that all right AI is is here to stay it's driving the conversation whether you want it to or not All right. So please, in 2026, let's all do better. Not everything is binary. Let's actually read. Let's use our brains. Let's put the human, keep the human in AI, right? All right. So I hope this is helpful. If so, please go to Your EverydayAI.com. If you could share this show, repost this because I think more people need to hear this. And then go to our website at Your EverydayaI.com. Sign up for the free daily newsletter. Thanks for tuning in. See you back tomorrow and Every Day for more Everyday AI. Thanks, y'all. Meet Firefly AI Assistant. Now live in Adobe Firefly, the Allman One Creative AI Studio. Just describe what you want to create in your own words, and the assistant handles the rest, orchestrating multi-step workflows across Adobe Creative Cloud apps,

Starting point is 00:40:22 including Photoshop, Premiere Express, and more in one conversational interface. You direct the outcome while the assistant accelerates execution. Stand control with the ability to step in and refine at any time. See it today at firefly.adobie.com. that's a wrap for today's edition of Everyday AI. Thanks for joining us. If you enjoyed this episode, please subscribe and leave us a rating. It helps keep us going.

Starting point is 00:40:53 For a little more AI magic, visit Your EverydayAI.com and sign up to our daily newsletter so you don't get left behind. Go break some barriers and we'll see you next time.

Everyday AI Podcast – An AI and ChatGPT Podcast - 3 AI Lies Most People Believed In 2025 (But You Shouldn’t)

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.