EP #39 How to Stay Visible When Everyone’s Filtering You Out
•AI at Work, AI in Business
Audio episode
Listen to this episode
LinkedIn is losing patience with AI slop and it’s not subtle about it anymore. There’s 20-25% more AI-generated content on the platform right now, engagement is down 17%, and LinkedIn just added a complaint button that sends a negative signal straight to the algorithm when users flag a post. If your team has an automated content flow, this episode is for you.
Kevin Williams and Matt Graham dig into what the AI content arms race actually looks like from the inside: the outbound personalization that stopped working, the media briefs that CMOs clearly never read before sending, the watermarking technology Anthropic just deployed to comply with the EU AI Act, and why the hyper-personalized cold email had a window that is now closed.
The conversation also covers what’s still working targeting the right person at the right moment, keeping a human in the review loop, and building the kind of differentiated content that doesn’t get mistaken for something an agent generated in 30 seconds.
Key topics covered:
✅ LinkedIn’s new AI complaint functionality and what it means for automated content
✅ The 20-25% content increase and 17% engagement drop and why it’s getting worse
✅ Anthropic’s watermarking rollout and the cat-and-mouse arms race with humanizer tools
✅ Why hyper-personalized outbound backfired and what actually converts now
✅ The CMO problem: AI-generated briefs that no human reviewed before going out
✅ How to stay visible when every channel is filling up with noise
Kevin Williams · Passage 3I am doing well in the dog days of summer. I understand that you managed to get out of the office for a little while.
Matt Graham · Passage 4Little bit, yeah, I had a little vacation, but you know, my phone's blown up the whole time with all these AI news. So it's hard to hard to disc disconnect a little bit. Too much excitement going on in the world.
Kevin Williams · Passage 5S so so so it so it admit it, did you have like agents running and were you like sneaking away from the fam to keep going, or were you a were you a good dad and and resisted the temptation?
Matt Graham · Passage 6the agents are always running, man. I'm I'm I'm a twenty-four seven kind of guy. With there's sort of like a expectation in our household that yeah, you know, I'm gonna hang out during vacation, but like I need to have like some time during your working. You know, like we we reach out to a lot of founders and CEOs and people the sales team will tell me, slow, it's 'cause it's summertime. I'm like, CEOs are always working, man. Trust me. There's always like
Matt Graham · Passage 8some portion of their attention you can grab even in the summertime. So it's just a reality of running your own business. I'm sure a lot of people listening can relate to that.
Kevin Williams · Passage 9So but but are you running things off of your phone? I guess for from a listener perspective, like what does the level of automation look like? And you know, I I think you guys are a Slack organization, so are you mainly like approving and triggering things or are you actually like outrunning agents or sort of what does your air quotes disconnected, semi disconnected world look like?
Matt Graham · Passage 10Yeah, it's a great question. So like Codecs on my phone, Claude Code on my phone. So I have everything through my phone if I need it, but I generally try to like when I'm with my family, put my phone down, be present, and then use my laptop because I'm just so much more efficient on my laptop. Like I can do everything I need to do, and that's my work time, and this is like my family time. So even though I have everything hooked up and I generally can do anything I need with my agents from my phone, like it it's just, you know, a personal preference to just do it from my laptop for speed, efficiency, and also like presence with my family. But for those of you who are are like kind of tied to your laptop, you you can actually get a lot done from your phone. My a lot of my teammates, you know, when I spend time with them in person and they, you know, we're on a retreat, they get an email, a client's asking for something, they're like, okay, let me just like knock this out quickly on my phone. You know, something simple, nothing too crazy.
Kevin Williams · Passage 11I would say that the other thing that's starting to happen more and more is a shift to cloud environments. so I was traveling extensively last week, and you can now use cloud functionality with both Codex and Claude. And I was boarding my plane and I was sitting there in the lobby, like trying to kick off some things that I wanted to run, and I shifted it into the cloud. And it does have some limitations, like generally it doesn't have access to the same files and things like that. But I got on a three hour flight and my little agents were crawling over the code for all three hours. And by the time I landed, like there was the the PR that I needed in the code to like keep going what I was doing. Then I hopped in the Uber and I like kicked it off. And I wish I could I wish it was a Waymo, but you know, you can't take Waymo's from airports yet. Like if I want to do like the the robotic continuity, it would have been a Waymo. But yeah, and then I'm I'm in the Uber and I'm kicking off the next step and whatever. And I realized I did about six or seven hours of vibe coding like just from my phone. And that's pretty remarkable to be honest.
Matt Graham · Passage 12Super remarkable. I mean it really shows you how How d development is becoming commoditized and the ease of doing it, the barrier to doing it. Like you don't need to be at this giant machine. Like when I first got OpenClaw, I hooked it up to the cloud as well. I didn't just run it locally. so it's it's really incredible what you can do on the go now. And think about what that means for labor too. Like people that may not have access to laptops. You know, I've traveled the whole world, I've been to a lot of poor areas in the world. And You know, laptops are something you may take for granted. There's a lot of people out there who will work for very low prices, five, three dollars a day. just from the and they all have phones, but they'll never have laptops, that kind of thing. So
Kevin Williams · Passage 15So there's also a the okay, part of this is technology. Part of this is just advancing in the the way that the large language models are working and the way that the apps are working. But admittedly, I am not as functional on the phone. I am not curating the outputs as much as I would be if I was at my laptop. And it actually suggests that perhaps I'm over curating on the laptop. That if I'm sort of sort of like blindly accepting as I'm in there in the Uber, do you think I'm actually diving into like Like individual code lines? Absolutely not. I'm looking at it. I'm like, yeah, that's kind of smells right. And I'm moving forward. And I think the reality is that I think I'm doing more than I actually am in those cycles of review. And that we're probably getting closer to that idea of, you know, I always like to call it baking cookies, of having what that concept is and basically putting it in the oven and not. having an interaction with it until it's gone through a whole cycle of review or testing. So you know, I sort of joke that the, that, that the what this is going to look like in six months, a year, or 18 months is essentially you, you, you spend an hour doing product design, product conceptualization, use the LLM to sort of bounce ideas back and forth. And then you basically say, go do it, you pivot in your chair And if you're using the most expensive, fastest model, by the time your chair pivots around, that first version is is almost done for you. So I'm feeling way more comfortable with that, which is which is interesting. Like I I'm I think it's like a level of trust in the outputs that I feel is going up. And part of that is expertise and practice and sort of having a nose for what's good and what's bad. Part of it is infrastructure, is, is I have built the right underpinnings to the way that I approach building things that gives me continuity in in between apps. So I'm not having to refigure things out. But also the models are just they just continue to get better. And I think we're gonna look back in this period and be like, you did what? You were spending like all day like looking at this stuff as opposed to. doing the intellectual work surrounding what the output's supposed to look like.
Matt Graham · Passage 16Exactly. I think that's why, you know, like I like my laptop because I still there's more things I need to click, more control I can have. But the models are get I mean, we talked last week about long horizon tasks, about how a lot of these firms are buying company data so that they can train their agents. And I think we're starting to see the effects of that. Like it's getting so good that you can have really crappy prompts, really crappy or or lower touch points with your agents to accomplish the same level of work. versus like six months ago, like you pretty much I mean you could do it from your phone, like with OpenCloud, but it was better to be at a laptop and have like the full functionality of your laptop to get your work done. but Kevin, in two years, I think somebody in a foreign country with Codex on their their old phone is gonna be pretty competitive in the development world.
Kevin Williams · Passage 17Yeah, and they'll probably going to be using codecs with so the latest and greatest model in in GPT is five point six soul. And in a couple of years, like the the soul of today is going to be not to to use their nomenclature. It would be Luna or it'd be Terra of these like lower end models. And you know, fact is we've said this a lot, but you don't necessarily need the latest and greatest model. and there's some interesting data that's floating around about that with anthropic fable in particular.
Kevin Williams · Passage 19that's so Fable is the is the high-end anthropic model. It's sort of it was related to to Mythos, which is the one that caused the kerfuffle. And I've used a lot of Fable for the security side.
Kevin Williams · Passage 21it's also really expensive to use and couple of different things are happening. there's a bunch of data out there that's saying so RAMP, which is the you guys probably use RAMP. It's a it's a finance management, you
Kevin Williams · Passage 23know, platform is great. But they get, they have transactional data from basically everyone. it tends to be tech forward. So it's a self-selecting population, just just just to be clear. And they can gauge different spend levels, even on a model level to some degree through some sampling methodologies they use. And they're finding that most people are not using Fable. And there's there's one like technical reason that I think that well, not technical, but policy reason that I think is getting missed in the media, which is that Fable still has a data retention policy. So even though if you're if you're using Fable within a corporate dev environment, even though your terms of service say that your data can't be used for training in the models. Fable is being retained for 30 days behind the scenes. And nobody is entirely sure what Anthropic is doing with it. It's it's obviously like a liability type issue. But if you're a big corporate entity, you can't have Anthropic just holding on to your data in that respect. So it's banned in a lot of organizations and they can't use it. So that's one. But it is also crazy expensive to use. It's like It'll light your tokens on fire to use it for any period of time. So internally, we generally are using OPIS models instead, which are great. So being able to bias to the right model for the right task at the right cost level is really important. And I do think that it's a bit of a fallacy to assume that people are going to always want to use the latest and greatest, smartest and best model if it has a cost disadvantage to it.
Matt Graham · Passage 24Yeah, the one super quick side quest here. I saw the most amazing video on YouTube about how Fable was used to solve these incredibly difficult math problems that haven't been solvable for a hundred years. So in mathematics, you know, there's tons and tons of things that we know are true, but we're not able to mathematically prove them. Right. You run lots of experience maybe in physics and you always get the same result, but mathematically you don't know the proof that like makes that true every time. And so they threw Fable at some of these math problems that have been unsolvable for a hundred years and it actually created the theorem or the proof behind them. I don't know if you saw that, Kevin, but that's
Matt Graham · Passage 26what you should be using Fable for, guys. Okay, if you're a PhD mathematician trying to solve a hundred year old math problem, use Fable. First of all, it's just crazy that it actually worked. and that's
Matt Graham · Passage 28like the level of capability these things have, which I think you know, we're not gonna just have a step function and all of a d sudden wake up one day with A GI. It's gonna be a slow, slippery slope where we all of a sudden realize like, Whoa, okay, I guess it's here and it's been happening over the last couple of years.
Kevin Williams · Passage 29So a couple of a couple of notes on that. Number one is that it's interesting that it was actually a broad-based model. It wasn't a specific model. There was definitely
Kevin Williams · Passage 31this period of time where organizations, the Frontier Labs, were leaning into very specific models to solve problems and they were developed legal medicine, and the
Kevin Williams · Passage 33and there's still like fine-tuning going on in that respect. But in general, they found that the bigger the model is just the better it is at all tasks. So unless you can build a giant model that is medicine specific or legal specific, there's really no point in doing so as opposed to just tuning it to those results. But in this case, this was just like off the shelf stuff. And I believe that the cost was like $2,200 or something like that in tokens, which is a decent amount, but it was like, you know, six hours of runtime and twenty two hundred dollars to like come up with a theorem. And, you know, you ask there, there, I I I wasn't well prepared for this topic, but I I know that there was a mathematician that spent like seven years of his career on like trying to solve one of these. And on the one hand, it's gotta be like, man, I spent all of this time and effort working on this. But on the other, now he gets to focus on the next thing. And he doesn't have to worry about this. Like this is this is settled.
Kevin Williams · Passage 35And now we're going to move on to whatever the next frontier of math is. I wouldn't necessarily want to be a budding theoretical mathematician at the moment, but being somebody senior who can really make some hay out of the results of this would be pretty exciting.
Matt Graham · Passage 36Yeah, totally. And a funny thing going back to our prior conversation. I don't know if it's true, but this is what I saw on YouTube. One of the mathematicians was literally on a jog when he, you know, prompt he prompted Fable to actually solve one of these problems. And, you know, he'd jog for a while and it'd come back to him. He's like, no, do this, do that. And that's it. I, you know, I I don't know how much of that is like, did he start at his desk? Did he go on a jog and do a few things on the jog and then get back to his desk? But still pretty crazy that he's on his phone for at least part of that, you know, cracking of a hundred-year-old math problem.
Kevin Williams · Passage 37So it's even worse than that. He's not a mathematician. The guy who did this was actually he was an anthropic staff member and with a bent towards math and was like, I'll give this a try and see how it goes, and then go off for a run and then come back and have the answer. yeah. So I that like that's the level that we're starting to get to. And yeah, this show doesn't really talk about like AGI and such because that it just gets to be such a fraught conversation. But like when Coding becomes a solved problem or math becomes a solved problem. Like what doors does that open up? On the end, the coding side goes back to what I was talking about about just, you know, turning in your chair and having a a solution out the other day, other end. That's super exciting as an innovator because you don't have all of these levels of friction. And people are just creating things like left, right, and center. And it's it's absolutely amazing. And then from the the math perspective, like what again, what new frontiers of of science does that open up?
Matt Graham · Passage 38Yeah, I think it's always interesting because if when AI first came out, I was like, this is not thinking, it's just probabilities. You know, it's not going to be able to come up with the concept of relativity if it didn't already exist, or the concept of gravity, or these absolutely brilliant insights. And this really challenges that thinking that I have that hmm, I mean, yes, it's a probability machine, but you know, it is cre it is discovering things that humans haven't been able to discover so far.
Kevin Williams · Passage 39So there was some major research in cancer and in the last week as well. noting that that wasn't generative AI. this it's very easy in shows like this to miss the fact that that the world of artificial intelligence is much bigger than generative AI. We tend to talk about generative AI, large language models. But two different areas that are really important for people to remember on the one side is machine learning that has been in place for you know 20 years is now being amplified. By code that is generated by generative AI that is solving really, really challenging problems in medical research and other places. And then the other field that I think is is getting undercovered is our physics models or world models. And that's sort of what's happening next in in development, that that language models are dealing with combinations and correlations between words and concepts. Okay, fine. I think we kind of get that at this point. But when you think about something about like how an autonomous car works or how a humanoid robot works, you can use a large language model to tell it that the physics of a parabolic arc are going to be this, but it's basically words that are describing it and it has to interpret that and turn it into a mechanical function. A world model takes that and instead applies the physics of whatever that action is on top of it. Such that the the the robot or the autonomous vehicle or whatever it is essentially knows what that next action is. And like I don't feel like enough people are paying attention to this, but billions of dollars are being spent on world models right now. So it's just, you know, machine learning is advancing, generative is advancing, these worldwater models are advancing, you've got The the the Chinese just finished their humanoid robot games. I'm gonna get the name of that wrong. And one of the robots broke Usain Bolt's hundred meter record. Like
Matt Graham · Passage 44Yeah, I th I think the the div I mean I you know, I I don't know for sure what will happen, but my hunch is that we'll see slower progress in the real world stuff. Like when I just think about how long it took for self-driving, and I think the amount of data that you have to calculate, like every single object on the road and the physics behind every single object that's already in motion and where it's gonna be, it's just so much more complex to me and so much more multimodal that I think I mean we'll definitely have progress, but I just think it will be way, way slower than what we see with code and generative text. And I think like really this key unlock that you described before of okay, do we have to tr put a billion dollars into training Harvey for law and a billion dollars into the best accounting AI and a billion dollars into the best medical AI? I think that's not actually what we need. We just need the best generative model, like you know, how humans think, like Da Vinci, like he's a great artist, great scientist, great architect. And that like diversity is actually playing out, like you just said before. that will really make these models very powerful for agentic. And I really think agentic AI is where the you killer app is or where the biggest unlock of value commercially is for the world. For right now. I think the physic physical world, yes,
Kevin Williams · Passage 47I I I I don't know, man. I I I'd kinda like to go to Beijing next year and see this this like world humanoid thing because you know, think I like to think beyond just like technology and commercials, like like China has a real demographic problem in that
Kevin Williams · Passage 49due to its own policies, it actually has negative population growth. And it's recognizing that if it's going to keep can continue to be an industrial powerhouse, it has to fix this problem. So they're throwing tons and tons of resources at it. And this is the second year of this this like Olympics thing. And last year, if you you may have caught some of this, they had like soccer players that were like ambling around and they'd fall over and sort of like squirm on the ground and just in one year.
Matt Graham · Passage 50They look silly. You're like, look at that and you're like, no way they're gonna make any progress on this. But then you see
Kevin Williams · Passage 55the pace of change is is enormous. And the other piece with with humanoid Robo Robots, we're we're not a robotic show, but maybe we'll become a robotic show. Maybe heck, maybe we are robots. Who knows? is that the humanoid form can slot into humanoid tasks, right? So you can build, we've had like arm-based robotics for a long time, but they require very specific designs of factories and applications. but a humanoid form can basically walk right into whatever the the application is and start doing its thing. you know, one of the topics that we wanted to cover today is just a state of the state around AI slop and AI generations and what that means for or for people and for organizations as they lean into using more and more AI generated output and You know, I know Matt Matt is a killer on LinkedIn and you know i LinkedIn has really, really been affected by this. It seems like every comment, every post is pretty much AI generated at this point, right?
Matt Graham · Passage 56Yeah, absolutely. Like I I it's almost an unbearable platform at times because it's like it's just so much AI slop. The comments are not engaging at all. You get almost nothing out of reading the comments. I think the content itself is getting better, but the the comments are just unbearable sometimes. It is like, ugh, what am I doing on here?
Kevin Williams · Passage 57So related to this is just two weeks ago, Anthropic released an actual watermark concept within their own platforms. And this was actually required by the EU AI Act. So the EU AI Act passed a few years ago in a very European way. It is much more conservative than the rest of the world, and it requires disclosure. So the way that Anthropic is complying with this is through introducing algorithmic watermarks into their output. And it is already live and it is already out there.
Kevin Williams · Passage 61So so so how watermarks work. so you know, it's not like an artifact that's built into the text. like, there's some like hidden Unicode or something like that that's in the text. instead, it's actually designed probabilistically as far as the word formation that comes out of an LLM. And there are certain patterns that LLMs use. And you know, there's there's sort of the obvious like, it has an dash and yada yada yada. But very, very subtle differences from from standard basically human writing that humans generally won't pick up on, such that the watermark can carry through copying and pasting. So again, you can't just like strip it out because it's it's not like Unicode in there. the pattern is really, really hard to disrupt.
Matt Graham · Passage 62But but I wonder, like, are we gonna see a flood of tools that try to un-AI your AI? You know, instead of just having a human, you know, like look, I think AI is helping people be more communicative, more concise, ha get their content fine-tuned so it's actually better. So I don't think AI generated content is bad if a human's in the loop and editing it and actually fixing it up as a co-pilot. What's really bad is like you just have your CMO who puts their content out there and so like says their agent, Go ahead, make a month worth of content, blast it out. I don't need to review this. And I think we have examples.
Kevin Williams · Passage 63Well, but that's exactly what's happening. Humans are lazy. So, like my own anecdote about the CMOs is that I was working with a organization that does ad buying. And he was telling me that his clients are CMOs of pretty large organizations. And an ad buyer works on a media plan and a budget and whatever. And what's happening to him is he's receiving media plans from his client that they obviously haven't read. And his job is to execute on the media plan. His job is not to like question the media plan. I mean, obviously he would if it was like really a miss. And then he'll do exactly what he was told to do. And the CMO is then surprised that he did exactly what he was told to do. And it becomes glaringly obvious that the brief was basically phoned in through GPT or Claude and no human took any any time to look at it. And then they just shipped it out the other way. And like, you know, what what What what is your job at that point? If you're just like pressing a button and that button is shuffling data to somebody else who's pressing a button and then doing something, you kind of deserve to get replaced at that point.
Matt Graham · Passage 64Definitely. I we have some some teammates who are super users of AI and they just copy and paste, copy and paste. They don't like you know, like when you're done with a meeting, just clean it up a little bit 'cause the AI didn't really get the action items right. It didn't really get like everything perfect. Or it just generated like three paragraph likes way too much stuff that you don't even need to read, right? It didn't make it concise enough. So I'm always one of those guys that just edits anytime I'm gener using AI. It's helping me go faster, but you gotta edit it. Otherwise you're just ruining everyone else's time. It's super noticeable with candidates. Like candidates who, like, you know, we're hiring for a social media manager to, you know, and like they literally just gave us the test task as AI generator. I don't think they actually looked at any of the test tasks and like did the research themselves. Cause there's all these little things that you can tell it's this is garbage. And I wonder with with schools too, like You know, like you people aren't learning to think anymore, Kevin, right?
Kevin Williams · Passage 65Yeah, I had a I I I had a a really lengthy conversation this weekend with the head of AI for a regional university. And she's in charge of thinking through what this means from a from an operational perspective from the university, but also what it means from a human perspective. And, you know, already you've got this whole I know, we're gonna sound like old, but you have this whole generation of kids. who basically have grown up online. A lot of them like we had formative years through COVID. And then the kids who are graduating basically this year came into university with GPT already running. So they've spent like three and a half, four years like deeply embedded in just using it. And you know, it's it's basically a travesty that they're they're graduating having abdicated from most of their their their work. they've lost a lot of human-centric sort of activity. And it was it was heartening to hear that that that she had a very solid focus on not not liberal arts per se, but like core human activities around communication and leadership and presence and speaking and et cetera. Because you know, as we were saying earlier with the humanoids, like it's gonna take a while for for those things to be a thing. But the human element sitting on top of the technologies will still have a ton of value.
Matt Graham · Passage 66Yeah, I'm just trying to teach my kids, be curious. You know, like it I think where this breaks down is people don't engage with what they're they're generating. Like it's okay to use AI and if you're engaging with the content that's produced and then using some critical thinking to modify it or get a better output. And then change it. But if you're just if you're just abdicating all that thinking, that's when you're in trouble for your career and your future. And the people that have this curiosity or just natural interest in what they're doing, they're just gonna be so much more superpowered than the person who's just sort of like, it generated the answer. Like, this is good enough. Like, whatever. I didn't even need to read it.
Kevin Williams · Passage 67So meanwhile, like on the slot battle side, like LinkedIn and the social channels, I have been very vocal about the fact that for a long time, Google, LinkedIn, Meta, they don't necessarily care that it's AI generated as long as it's good and it gets engagement. Well, that is actually starting to change. And there's some data out of
Matt Graham · Passage 70Is that a good thing, Kevin? Because I'm if it got engagement and it actually produced value, even if it was produced by AI, I don't mind. Like
Kevin Williams · Passage 73pieces so some of it is, some of it isn't. So yes, if you if you create a really engaging piece of content, particularly on on consumer social like meta, like that can definitely be rewarded through clicks. But through like business channel communication, et cetera, the data is that you have something like 20, 25% more content that's being generated. And like 17% less engagement on that content. And that's not just because there's more, it's just because it's not very good. And if you're LinkedIn and you know, your your primary purse, per purpose is to connect different professionals and allow them neck networking opportunities. How that has generally worked is through the promotion and the creation and the curation of good content that allows people to share their voice, etc. And you know, Substack. the the the blog the paid blog channel is also very similar and both have introduced basically ai complainting complaint functionality built right in like you can click a button on link it linked in now and say this is ai swap and sort of like a thumbs down and it's sending a negative signal back to the algorithm. So for those of you who have laboriously built like content flows that auto-publish and do things. I think that the banhammer is probably heading our way.
Matt Graham · Passage 74But I I guess I j I'm a little worried because I think AI does help people communicate better and generally help people produce better content. Again, if they're interested, if they're trying, like same thing with a student in school who uses AI is going to get a better output. If I use AI to help me write my LinkedIn content, I'm going to get a better output. But you know, what happens if they have a little gauge that says, Matt, is this is like 40% AI? You know, and I they I don't know if we're gonna start seeing interfaces where this is fully AI or zero AI or is just like partially AI or any of that stuff. But, you know, I don't know if that's I don't know if that's a good good thing. Like I don't know if you really need, you know, I I think they're AI transparency is important, but I don't want to be penalized for how having AI help me write better content than what I could generally even do myself. You know what I mean? I think
Kevin Williams · Passage 77I think the consumption matters every ba bit as much. So okay, cool. Like you use all the AI you want. If if it's not getting flagged, or if people are engaging with it, fine, but increasingly people are going to be using their agents to engage with the content. And then you have this like weird agent to agent like death loop thing going on because people are just deluged with mediocre content. So Those who can get an audience, and I think Substack is probably a good place for this, for the writers who are still writing and have really novel things to say, people will gravitate to those sources. They'll gravitate to long form like Atlantic type articles or whatever they are if they want to get that. But but things that are on the peripheral will just be harder to either get attention, even if it is authentic. Like there's just too much noise that's out there.
Kevin Williams · Passage 79And I I I think we're we're we're probably going to see some changes in consumption that way that will be detrimental to the way that we've gone about content for the last ten years.
Matt Graham · Passage 80Yeah, I think in general, like if you if we let this continue unchecked, we would have produced ten times more content in the world than what humans could have possibly could produce themselves. And it all mostly would be garbage. So I in general, I think this is good. I
Matt Graham · Passage 82just hope the pendulum doesn't swing so far to the other side that people that are legitimately using AI and part of their workflow to get better content improved and the communication. Like they, you know, you go on Substack and you see something that says forty percent AI generated and you don't even read it because you're just like so adverse to any AI, you know. But my friend texted me the other day, he's like, Look at this article. I was really pleasantly surprised, because the quality was so good in it, but it was still like forty percent AI written, according to whatever algorithm. I also wonder, like, are we gonna have a bunch of like, you know with shadow tools out there that try to like undo the AI, you know? Like what's that gonna look like? Okay, I got this perfect content
Matt Graham · Passage 84and now I need to undo the AI to get the forty percent to twenty percent.
Kevin Williams · Passage 85It's already an arms race. Like if i i if if you want to play that game, there's a game that can be played. So you you you do AI processing of your material and then you loop it back through one of the AI like detection tools. So Pangram is probably the one that's the best. I I spent years saying all of these things were junk and they're becoming less junk. and then you keep running it through whatever LLM versus Pangram to improve it to the point where it becomes, you know, only fifteen percent AI generated or whatever it is. However, Pangram stays in business by staying on top of that. So whenever anybody has a humanizer device, they buy it, they figure out what it's doing, and then they train it into their internal model. So this is always gonna be a cat and mouse thing. but I do believe in like academic environments. We were talking about that for a second. Like, you know, this is bad news. And being able to have a reasonable detection met mechanism is great, but changing the way that people actually are interacting with with higher education is probably going to be more important. the other thing when with like, okay, how do you create good AI content? I would say that for me, the primary piece of advice that I have for people is making sure that you're saying things that are that are you. Like even if you're using AI. So for instance, there's a transcript that's being generated behind this podcast right now, right? But Matt and I at least like to think that we're saying the occasional smart and original thing. And using the smart and original thing that we're saying today at the end of August. Is far different than going out to an LLM and saying, hey, write me a blog post about what's going on in AI today, because the the LLM will naturally revert to the mean. Whereas there's a level of human curation as far as the idea and the idea focus that comes from a conversation like this, such that it is differentiated. At least theoretically it's differentiated. And then, yes, using a good prompt layer to it to try and make it sound as much like you as possible. And then lo and behold, you have to do the work and you have to go back and you have to look at it and clarify. What would you add to that
Matt Graham · Passage 88actually our team, our content team, we have we have a Gemini Matt brain and we have a a Claude Map brain. And a few teammates have just tried different tools to download all my posts on LinkedIn, all my my transcripts from all my calls to try to get Matt's tone. Right. Cause they're they're usually I'm dropping them a voice note and they're you're going to AI obviously to draft it. And then we have a couple of review steps after that. But they're using like, you know, all this content and the prompt says, write this content in Matt's voice based off, you know, these three years worth of recorded. voice of Matt. so I I think that helps a lot to get the tonality right. But nothing's gonna replace like the human insights. Like, you know, I d I still don't see AI. I mean it gives me like interesting ideas, but I think my ideas are way more interesting. I think your ideas are way way more interesting. You know, maybe we're a little biased, of course. But I don't think it ever comes up with anything too original or mind blowing. I think that's still uniquely us in the moment, right now, this week talking right now, right?
Kevin Williams · Passage 89So the other ramification of this is like for business people who are listening, is what is how is this reflected in in how your marketing and sales teams are doing communications? And both of our organizations do a lot of work in sales structures and go to market structures to figure this out. and I I I had to laugh. I sent Matt a a s a screenshot earlier this week of of one of those emails that we are all getting, all I'm I can guarantee that whether you're recognizing it or not, you were getting highly personalized emails. And generally they take the same sort of format. They're like, Hey Kevin, I see that you're doing this really cool new AI build kit thing on on your website. That's great. And then they'll try and sweeten it and they'll like scrape LinkedIn, which you're not supposed to do, but you can. And they'll be like, I saw your post on X, Y, and Z and et cetera. Look at how informed I am. Look at how personal this is to you, such that you're going to interact with it. And for a heartbeat, yeah, that stuff like totally worked. But now that it's just like the standard, first of all, it is actually like the standard that if you want to get anybody's attention, you need to be doing this. It can also go seriously awry. And the one that I had sent Matt basically had a mismatch of like totally wrong information, like just a little kernel that was accurate, but their scraper had gone completely awry and their automated SDR, their sales development like re representative app or agent had gone awry and created just a hodgepodge of information that made absolutely no sense. So, you know, I guess Matt, speak a little bit to what are the what is what are the best approaches everybody's gonna do this. So figuring out what
Matt Graham · Passage 92We we do we b we invest a lot in outbound. and so there's a lot of infrastructure you have to do, but so that's like your science part, but then there's there also the art of the message, the content, the sequence. And I remember two years ago, like this was all the rage. This is why Clay t really took off. If for folks out there who don't ha run outbound campaigns, there's a bunch of tools. Instantly, warmly, clay, and they all like have this functionality and we thought there was gonna be this with this incredible boom. in email outbound again because email can can't kind of had died. You know, it was just so much spam. It's like junk mail, everyone's ignoring it. It became too easy to set up this infrastructure and the whole world started doing it. And the the amount of volume just went up so that it just created too much noise and everyone stopped listening. So we thought that there'd be like a renaissance in outbound, like all this ability to customize this stuff and really grab people's attention and make it really specific to you. And what happened? It just, you know, you got a lot of garbage, like Kevin's saying, you know, it would totally mess up the message. It would make the wrong assumptions. It's still possible to do it. But what actually works better is just a plain old message, like you're my ICP in this industry. We built the solution, and this is why you should care about it. That's all that matters. It doesn't need to be about your most recent LinkedIn message. It doesn't need to be about like some news article about your raising a bunch of money and you know saying congrats at the beginning. That stuff is just so detectable by all of us now. And you can just tell it's AI. And then, you know, a lot of times it's getting it wrong completely wrong. And so I talked to a lot of people in the old in the outbound game and industry and they all say the same thing. Dude, don't just just stop worrying about all that customization, that personalization. Just keep it simple. This is who we work. This is our use case. This is what problem it solves and this is why it's valuable for you. That's it.
Kevin Williams · Passage 93So so the you know the key, I I've been saying this for years, that a marketer's job right now is just to be a better marketer. Like, you know, that that that it it hasn't necessarily changed. and finding the right timing of that message and the right recipient that is within your ICP, your ideal customer profile. is a great use. So things like we
Kevin Williams · Passage 95all know that if you're trying to sell into an organization, somebody who has recently moved into their role, there's this window of opportunity where they have budget and they have freedom that they can shake things up. So how do you find that person at the right time and make sure that you're getting a message to them that is that is totally relevant to what they're trying to solve. It matters less that yeah, okay, they posted on little league championships or or whatever. if they are in need of your product and you put your product to them at the right time, that's that that's where the mojo should be applied.
Matt Graham · Passage 96Yeah, and I see this in companies that have to work across Europe and the US. Like your your ability to target customers in the US is just so much better than in Europe. Like there's so many laws around privacy. And I mean, we could debate that if you want, if that's good or bad, but at the end of the day, like the US has much looser privacy laws which allow companies to solicit and really target way better the types of customers. I personally think that's good for You know, because you know, a lot of people have problems that they're looking to solve and they don't know how to solve it. And so it's helpful to be able to know that you're my ICP and you need this problem solved. Of course it the downsides it leads to lots more a lot of more spam. But Like these tools are getting really good at collecting all this data and really narrowing in on who your ICP is and when they have a problem, especially if if you're selling solutions that are, you know, time based. Like, hey, this we this person had this intense signal. So now we know they're thinking about a solution. You know, like everyone needs a look an accountant, but there's only certain times of in my journey as a CEO that I need to get a new accountant, right? Or change
Matt Graham · Passage 98my accountant. I don't want to get spammed for five years about Hey, I need an accountant. but I do want to get spammed when I'm actually looking for an accountant, right? And these tools are really helping on that side. So it's it's a good point, Kevin.
Kevin Williams · Passage 99Okay. So I so I I I'm just gonna pontificate on that what I think the next generation of this is. as we were talking about
Kevin Williams · Passage 101like agentic workflows and things, you know, I'm I I just took a glance at my own inbox and there are just hundreds of of emails. Like it's really hard for me to just deal with emails, even being like an AI guy and like vetting it. But and then texts. So I I find text spam to be just so painful. Like The the amount of like loan offers, phone calls. I
Matt Graham · Passage 102Phone calls, Kevin. I have a hundred and seventy-nine voicemails. Like I had to change
Matt Graham · Passage 104my voicemail to say don't leave me a voicemail because literally I can't handle them anymore.
Kevin Williams · Passage 105I I won't look and I'll I'll I'll miss things. Yeah. So but it's it's not that far off that even if Apple ever figures out Apple intelligence, like, you know, I I there's a pattern to it. Why for pretty soon I bet I'm going to have like a little personal assistant agent that sits in iOS and I just say I am not interested in loan offers in no way, shape, or form. I am not interested in There's something called commissioned LLC that that the interwebs think that I'm associated with. And I get like 10 messages a day about this other Kevin Williams' like IRS dead or whatever it is.
Kevin Williams · Passage 107Well, you know, there are a lot of Kevin Williams around there, just for the record. It's good to be anonymous, but you know, kid,
Kevin Williams · Passage 109this is this is a Kevin Williams in Florida. He has some IRS problems. I know a lot about this guy because I get like all of this inbound. And but why do you get the inbound? You get the inbound because they have incomplete records that somebody has scraped an IRS database. It's in Florida or whatever, and they've identified that this dude owes a bunch of money to the IRS, but they don't have his contact information. So then they do an enrichment step using clay, using whatever, and they've hoovered up me from a data broker and they're just making a numbers swing at it. Like there are I happen to know there are over 11,000 Kevin Williams in the US. Very common name, 11,000. So they've hoovered up probably 2,000 of us, and all of us are getting like the the the notices on this. But that's that's an easy vetting criteria for something like a bot where it would retain memory. Like, I'm not interested in this. I will tell you if I am ever interested in like fundraising or taking on debt or whatever it is. And until then, I don't want to see any of the phone vi the emails, I don't want to see any of the phone calls, like all of that needs to go away. Okay, that's fine, but now you're going to start seeing those bots that are creeping into boxes. And it's like it's I'm reminded a little bit of the battles over over ad targeting. Like, you know, I'm I I you know, I I I interesting that I I value my personal privacy, but I've been a digital marketer for a long time. So like having access to that data is really important. but I actually tend to allow platforms to target me. If it's in platform, if it's like meta or whatever, because if you don't, the content that you get all turns into like McDonald's and Ford and like health insurance and other stuff. Yeah, and whatever it is. Hey, I'm not that old, but yeah. So but but but it's the
Matt Graham · Passage 110Yeah, like Alzheimer's drugs and the stuff that like, dude, I don't care about any of that stuff. Like, yeah. Come on.
Kevin Williams · Passage 111you know, if you if you if you want to be more targeted, you get more targeted information. If you use agents to basically scan your way out of everything. You're going to end up narrowing your focus a little bit. This is going to happen anyway in inboxes. So I think that this is going to get a lot harder, such that you can say, I am simply not interested in any solicitations because we're all just overwhelmed. You are interested in solicitations when your accountant like decides to quit the business. You are. But you're no longer going to get those. So it's actually could be harder for you to find your way to the right solution at the end of the day. I know that sounds a little tortured, but it's like on the one side, marketers are breaking the system by putting too much slop, too much crap out there. On the other side, consumers, be they business or individual consumers, are likely to overreact. And the winners will be. those who are basically offering really broad based generic services, the insurance companies, the healthcare companies, the dementia drugs, like whatever,
Matt Graham · Passage 114Yeah, I think I think it like when I think of myself as a business owner, what are most of my problems? They're mostly who problems. They're not how problems. Like I don't really need to know how to solve the, you know, top ten problems of my business. I probably need, you know, more I there are more who problems. Who can solve these problems for me? And I I need to have solicitation for that. I need to know what are the solutions out in the market. I need to know who are the vendors. I need to know who are the people in the space. So I want to be solicited for certain things. It's just a volume. is too much. That's what's causing all this backlash. If it if the if the something could make it more expensive to do high volume, let's say LinkedIn, like LinkedIn has really low limits. So you have to be pretty targeted, right? Like I find the quality I mean there's still spam on LinkedIn, obviously, but the quality of what I'm getting spammed on on LinkedIn is far better than the quality on email. Just because anyone could do email is just, you know, you can throw anything in anybody's inbox and it w at a pretty low cost. But LinkedIn, you're limited. And you're gonna get better solicitation there. So I agree with you. As a business owner, I need to find solutions. I need people to be coming after me because a lot of problems I gotta solve and I need help and I need s a vendor or you know, buy something. So it's good. It's good for business. I think it helps business move faster. It's just it's got out of control in some some areas and some channels.
Kevin Williams · Passage 115Yeah. And I think for folks in our audience, this is heading your way. So one, we've given you some pretty practical messages about how putting up better information. And better information is not necessarily personally like like tailored to the nth degree. It's using the available data to find the right people at the right time. You also need to recognize that basically everybody has the same data. That whether you're in instantly or you're in Apollo or you're in Zoom info or whatever, like all of it's kind of a dirty industry. No offense to anybody's in that industry,
Kevin Williams · Passage 117as far as they're buying data from here and they're buying data from there, and it's all been like aggregated and then they're putting different skins on it such that they can offer it. But you know, you know, these opt-out services, fine, but like my Kevin Williams emails have been sold like ten thousand times. And now they're just everywhere. So like that's don't think that you're special unless you actually are special, that you have some sort of like list that's a captive list that's of like, you know, attendees of your events or whatever. That's the sort of stuff that marketers need to be seeking.
Matt Graham · Passage 118Yeah, it's interesting. Like I I don't know last time we talked about how Matt has a security company, Matt has an IT services business. I actually have a lead gen business as well. And we've I've learned a lot about the data brokerage industry and like A lot of dirty, crazy things happening in the data space. It's incredible how much data they have.
Kevin Williams · Passage 119Yeah. so again, marketers need to think ahead and figure out like what is this going to look like. I mean, I think to some degree, I'm I'm working with some very, very high touch type businesses right now where the transaction value is incredibly high, like, you know, upper six figures, multiple seven figures type type. And for them, just doing cold like email generation is never was never gonna So it's doing things like identifying where people are in their ICP and putting themselves physically in front of them. So what conferences are these people going to? Who's going to be there when you walk into the conference, having a r a real dossier on the people who are there and having basically an abbreviated pitch if you manage to work your way in front of these people? that's gonna get annoying for everybody as well because, you know, sort of the idea of everybody having meta glasses and being able to look around a room and see who's who and be able to identify that might be coming as well. But at the very least you can kind of expect that that that people are gonna know a heck of a lot more about you than they did previously.
Matt Graham · Passage 120Way more. They know what websites you've gone to. They have they have pixels on your website. They have cookies on your machines and your devices. I think it's interesting Apple plays a big part in trying to limit this, but literally, guys, there's if you go onto the dark web, you know, part of our security firm goes on the dark web and we get we see almost every American social security numbers out there. Their credit card company knows exactly where you are and what you bought. It is an incredible amount of data. You are tracked everywhere. They know everything there is to know about you. Pretty scary world.
Kevin Williams · Passage 121All right, Matt. So we've we we we dove into the dark side of digital marketing there. But it is a it's a hard time to be a marketer. It really is. We're getting a lot of mixed messages about what's going to work, what isn't going to work. you know, there's so much volume, there's so much noise out there right now that you want to be as genuinely as you possibly genuine as you possibly can. and as yeah, we can go back and forth about the personalization thing, but you do want you want to be authentic. So using these tools in that way I think is is where I want people to focus.
Credit the named speaker and AI at Work by AscendAI. Link to the episode or the passage you quoted.
“EP #39 How to Stay Visible When Everyone’s Filtering You Out.” AI at Work by AscendAI, 2026-08-28. https://ascendlabs.ai/podcast/ep-39-how-to-stay-visible-when-everyones-filtering-you-out/