The #1 Mistake in AI-Generated Content (And How to Fix It)

Sean Weisbrot: What if the future of brand marketing isn't about being the loudest voice, but about being the most discoverable to AI? Today I'm interviewing Curtis Spar, the principle of Bo Spar, a leading marketing and PR agency. Curtis shares how AI is transforming SEO video and corporate communications, and why in a world of algorithms, authenticity might be the only thing that AI can't fake. How is AI changing the way brands market themselves?

Curtis Sparrer: Well, I would say that people are really thinking about all the different steps that AI takes and gobbles up content. And now they're thinking, okay, how can I make sure that my content is more scannable? How can I make sure my content is more understandable by ai? And then how can I make sure that either it's living up to the best standards of answer engine optimization or generative engine optimization.

Sean Weisbrot: What do those terms actually mean in English?

Curtis Sparrer: It means that when you are asking AI like ChatGPT, I need a tech PR agency, it will spit out something based on what it's able to find. And usually what it's searching for is for journalist stories about a company. And so generative engine optimization is a fancy talk for saying, make sure people think well of you, make sure they write about you and make sure they're respected journalists who people care about.

Sean Weisbrot: So can't you just buy that influence then to, to get the AI to want to talk about you.

Curtis Sparrer: That's a good question. Short answer, no. Uh, the reason being is AI prioritizes the content that is earned, uh, by journalists and media outlets that are respected. It also prioritizes media outlets that are very niche and very, uh, in the weeds, if you will, but it doesn't really get excited about media outlets that are obviously paid commercials. And so. It's trying to give you an authoritative answer, and so it's giving those answers based on the authorit of the person and the media outlet.

Sean Weisbrot: I've seen many different platforms, say for $95, we can get you onto Forbes and Business Insider and these other platforms. If those people are actually able to get you onto those platforms, are they not really then able to manipulate that algorithm?

Curtis Sparrer: That's a really good question, and that goes to the various nature of the Pay for Play. Universe. And the idea is, is that they will say, Hey, we pay by performance, you will get X amount of, you know, outlet coverage and Forbes Inc. And usually the outlets that actually take quote unquote contributors as opposed to staff and, Hey, I've been a contributor. For Forbes and for others, and I've been asked to do a story for a couple thousand dollars on X, Y, or Z. Now I refused, but you know, people had to put bread on the table so I could see some people saying, yes, I'll do that. But AI is hip to that kind of trick, and so that's why those sort of, uh, obvious sort of pay to play shenanigans aren't really doing as well.

Sean Weisbrot: Hmm. How can you. Objectively measure that they're doing well versus not doing well.

Curtis Sparrer: There are many ways. GEO takes its cues from SEO. And so search engine optimization has a variety of ways of telling just how authoritative a link is and how meaningful it's, and that's the current way in which we're measuring a lot of the generative engine optimization scores, if you will. And so. That's one way of doing it. Another way of doing it, of course, is, you know, doing it yourself where you're simply, uh, searching, you know, questions that you think people would ask about your service or your company, and then you'll apply it to your competitor and you'll see what your competitor is doing. We're actually working on a tool right now that will help people get a baseline for how they stand against their comp competition.

Sean Weisbrot: So I know that video is something that's been really hyped for helping brands to be able to get the information out and, and even I do this as well, providing content that is video based that can be put into your social media channels or your sales pipeline, et cetera, for the purpose of making it easier for people to find you and understand what you do and to trust you and all of that. I've also found that when I try to share a link from YouTube to Google, uh, AI or chat d bt, they struggle to really find that content and to transcribe it or to analyze it. So how can you make it find you better? How, how can you make that video content actually work in order to, to help that AI to understand do better?

Curtis Sparrer: I find providing transcripts is really useful. Usually when I get a good piece of video content where I'm on tv, for example, I will write a blog about it. I think one thing about generative engine optimization that's important to understand is it also looks at the content that you write about yourself. And so being able to put into context what's happened with you on video to provide all the assets around it, is another way to make sure that you are helping the AI understand you, even if it's a video placement. I think the short answer though is that when it comes to video, AI isn't sourcing that as strongly as it's sourcing written online content, which is easier for it to digest.

Sean Weisbrot: But wouldn't it make sense for, for people that are building LLMs too. Have these ais become better at looking at video directly because there's so much video being created and it's obviously the future.

Curtis Sparrer: I think so I, first of all, I'm not a large language model engineer. Sure. I'd probably be getting paid a lot more money. Uh, so I can't speak to all their challenges that they're facing, but I know that when I am using like Google Notebook, for example. To do research and I'm using video clips, it does take it a little longer and there are things that it misses, and I see that's not quite the case when I'm having it digest A PDF. And so I would think of it in terms of. I guess how you would think of any meal where you have, you know, crispy potato chips that go down quickly versus a steak and probably this is going to be a problem of your, and in a few months or years, this will be something that seems very quaint and that's the real challenge about this emerging science is it's just moving so quick. So probably if someone watches this in a year from now, they'll say, what are they talking about? Of course video is all index, but right now I don't think it is.

Sean Weisbrot: So what you're saying is for every video that I publish on YouTube, I should have a separate page on my website that also includes the transcript, the full transcript, so that the AI can find it and know more about me and my business and what I do.

Curtis Sparrer: I think that's a good idea. I think that when it comes to content, putting everything in context is important. And one of the things that I recommend is that, uh, companies really consider their press release policy back in the age of your, just a few AI ago. We would think seriously about, well, do we even put this as a press release? Will anyone care? And nowadays the thinking is different. It's like, well. Certainly a journalist may not care about this story, but I bet chat GPT will care about it. I bet Claude will care about it. We should put this in, and we have even seen in some of the, uh, answers that we've received back that. CLA or chat. BT Orrock is citing a published press release as an authoritative source of content. And as PR people, we would think, well, I don't think a ricochet of a press release is that important. But nowadays we're thinking, okay, well that still might move the needle on a couple of different, uh, prompts and queries. So that might be something we might want to consider in our PR playbook.

Sean Weisbrot: So on top of making these YouTube videos and, and putting the, the transcripts on there, I should also be making blogs about the content that are separate from the, the actual interview to provide more information that creates more authority for me. So like basically the, the more pages you have on your website, the more the AI can get to know you, the, the more it can understand when to recommend you to people.

Curtis Sparrer: You know, I would say I'd want someone to check my work on that. Uh, my inclination is to say yes, but I have SEO experts that I rely on for. This emerging field right now, I know that I am trying to make it as easy as possible for AI to be able to scan and find my content and my transcripts. So we have a podcast that we put on called Politely Pushy, and now what I'm trying to make sure happens all the time is yes, that the transcript is easy to find. And yes, that there is a. AI friendly blog. Putting that whole conversation in context and talking about what it means and why people might be interested in it. I think the other thing that people are, are taking a lot more seriously is the art of the FAQ. I think making sure that you have a. AI friendly, uh, frequently. An, uh, ask questions with answers is another important way in which people can invest in their own online properties and online identities to make sure that you are coming up at the top of any answer.

Sean Weisbrot: How do you make an AI friendly FAQ? What's the difference between a stat and a standard FAQ?

Curtis Sparrer: I think there are certain formatting issues that, uh, would probably be best, you know, as a step-by-step demonstration. And I think the difference is every different platform has their own preferences on what they're looking for in terms of summary and word link. Uh, so that would be, you know, my short answer on that one. But the point is, is that when I partner with any SEO expert, we come up with all the sort of questions and then we'll ask Chachi, PT or Claude, hey, what questions did we not think of? What would you be looking for? And that way we can backstop to make sure we're providing the most thoughtful answers possible. I think the best thing about AI is that we can really challenge ourselves to be better by asking it. What did I miss?

Sean Weisbrot: I love doing that. I'm constantly putting in transcripts and saying, what have I done? What haven't I done? Right? Based on best practices, how can I improve?

Curtis Sparrer: I sound snarky. Do I deserve to be slapped for this? I sometimes ask AI that too.

Sean Weisbrot: Why would you be deserve to be slapped by that? I think AI is a really great thing at, at being a mirror for us.

Curtis Sparrer: Yeah, I do too. And I think that the biggest challenge about being a mirror for us is that sometimes it just has a distinct voice that needs to be challenged. I think that I will frequently get content that I know is from ai. I could tell just by the M dash alone. And so I used to love the M dash until AI ruined it, but now that's just not going to be in my content moving forward because I know it's a tell. And so I think that journalists are telling me. More and more they're seeing pitches that are written that have a lot of ai needlessly wordy. It's always trying to prove a line that you could take for granted, and I think that's going to be the real sort of delta here, is that people are gonna have to sound more human. Age of intense ai, and I think that's gonna change the whole field of corporate communications because right now there's nothing that sounds more robotic than corporate comms, and I think it's gonna need a refresh if it's gonna be impactful.

Sean Weisbrot: You, you think AI will be able to make corporate communications more human?

Curtis Sparrer: No, but I think people are gonna be rewarded for making corporate communications more human. I think that if every journalist I'm talking to is saying, yeah, all the content I'm getting is really stale, it all sounds the same. I'm seeing a lot of M dashes. There's all these different tells, and it's all just kind of. Safe. I think that AI is gonna make, uh, quirkiness and being very human, really rewarded in terms of how that message gets picked up and how likely that message is to be green lit. Hmm.

Sean Weisbrot: I constantly tell the ais that I work with, don't be wordy. Make, make this thing as short as you can. Right. People want to read something that's concise or they want to see something that's concise. So I, I constantly do challenge them. You know, Hey, you said this, but like, I don't think that works. Or I provide more context, and it's like, oh yeah, actually that's right. Now that I think about it, like, that's not a good idea. You know, maybe this is a better idea. So I, I don't accept the first thing that I'm given. Because it's, I see it as a, like a, a planning. You're, you have an idea and you're trying to plan for how that idea can make sense and be something that's acceptable, uh, for whatever your purpose is. And so I, I think accepting the first thing you're given is never a good idea. There's, um, this guy on YouTube who talks to BT and, and he asks it specific questions. He typically focuses on. Certain topics that I can't name here unless, uh, YouTube wants me to not be allowed to run ads. So I'm gonna avoid saying what they are, and they're not bad. But YouTube is very sensitive. Um, and what I find is the, he uses certain phrases in his prompting in order to get a really fantastic result. And what he says is, no bs no fluff, no headers, give it to me straight. When you include those things, it's usually shorter and, and more, uh, realistic and and honest.

Curtis Sparrer: I usually give a prompt that says 150 to 200 words because I think that also keeps the AI honest. And I will also say no M dashes. No dashes, uh, no flowery stuff. No fluff, no adjectives. I have a whole list of things that I don't want in email con conversation anyway, and so I'm just trying to design that. I think the next step of AI is really going to be AI remembering our settings, because I think the challenge is, is that each time we go into a chat Cpue or Claude or rock session, it's almost tabular rasa and it forgets your preferences. It forgets your voice, and I think that's going to be almost as important as a agentic.

Sean Weisbrot: Yeah, I think one of the issues with it is the context window. There's just a certain number of tokens that the AI can use before its information starts to break down and it starts to hallucinate. And so I think they, they have this multi chatt system designed so that the context window can be refreshed for every new conversation or thread that you wanna have. With the ai, they, they all do have a memory based system for the most part. Whether or not it's good is a different conversation. I use Gemini Pro and I find that Gemini is pretty good, but I've also found that recently when I have a new chat, it doesn't really remember things, but if I ask it to pull information from the other, other, his, the other chat history, it's sometimes able to do it and it's sometimes not. So I have a, a general. Prompts that I start with, Hey, this is some information about me and or what we've talked about in previous chats. Use this as a foundation and, and let's go from there. And that's also pretty helpful. Um, chat VT is better at storing memory of you. Um, I can't stand Brock, so I don't touch it. Um, but I, I used Chat two Beauty for a while until I started paying for Gemini. Actually, I got it. A friend of mine gave me a three month free trial to Gemini Pro, and so I've been using that. It's been amazing, so I'll probably start paying for it.

Curtis Sparrer: I wish all these services would pay us just to talk about them, right? I, I think that's a part of it. I love Claude for writing, and I also love Claude for writing in French. I have in-laws that are French, and it really does help there. What I don't like is that it doesn't understand my voice. Even though there's a lot of published stuff that I have written, when I say, write this in my voice, it has no idea what that is. And so a lot of times I find Claude to be really useful when I'm assembling content where there's a lot of quotes and I use. You know, obviously order, and I say assemble this in a reasonable order based on the conversation, and it's great for that. I think ultimately AI should never be your final draft. It should be your first, and I think it can move things along much faster. I think that people who will use it as their final draft or think, ah, it's good enough, are always shortchanging themselves 'cause it's not good enough. And you know, there are so many ways you could tell, ugh, this is ai. And from a media point of view, when journalists see that. And they kind of get the sense of, oh, this is ai. They hit delete. 'cause journalists are reading to delete, they get about a hundred to 200 pitches each day and they are just trying to get to the good stuff. And the good stuff is frequently the stuff that's written by humans and not ai. Not yet.

Sean Weisbrot: I get about 10 to 15 pitches a day. A hundred percent of them were written by ai. How could you tell. Because they will often say, I love how you had this conversation with this other person where you talked about this. And I'm like, you didn't watch my interview. There's no way you sat there and watched a specific, you went through my different interviews looking for the one that's gonna help you connect me to this person. This, you know, this interview was so great and it reminds me of my clients and how they could do this as and like. You obviously threw in a bio of your client and you threw in my YouTube channel and, and AI told you what to say. I'm like, fair enough. You're trying to do a bunch of pitches. You're trying to do your job. That's fine. That's annoying. All I want is, Hey, here's the person's name. Here's their LinkedIn, here's their website, URL. Here's a paragraph about them. That's all I need. I don't need you to suck up to me and I don't need you to tell me how you like what I did and, and how it relates to other people. I just don't care. I don't have time to read that. I instantly think it's useless. I will offer intro calls mostly anyways to those people because the people they're pitching are actually really interesting. But the point remains that I don't need any of the nonsense that they're throwing at me.

Curtis Sparrer: You know, if anyone writes to you and says that they really loved your interview with Curtis Spur, probably nine times outta 10. It's my mom. So I, I apologize in advance that she emails you, or it might be my dad. Um, but that's called the me Too pitch. Hey, you did something about X and me too. I got X. Let's write about X. Let's talk about. And as a former journalist, I find the Me Too pitch, so lame and boring. I kind of like the pitch that starts off super short. It's about 10 words, and it gives me, this is new and this is important and it's compelling enough for me to read the second line. And if it's compelling enough for me to read the second line, I think, okay, this person has something they're thinking journalistically. If someone writes, Hey, I saw what you did, blah, blah, blah, blah, blah. I just think, I don't think I'm being pitched by ai. I just think I'm being buttered up in that someone doesn't have a sense of my actual media priorities.

Sean Weisbrot: You sound like a venture capitalist.

Curtis Sparrer: But not rich as a venture capitalist. That's the thing. I mean, I think the thing is, is that I have seen a lot of different companies come through and I've heard a lot of marketers talk about them selves and. I am thinking, okay, let's play kingmaker. This is what's going to make this company really work and really ping. And if the company focuses on this other thing, it's gonna be a lot harder for them to cut through because there's five or other, maybe even 10 companies who pitched me something similar. And they might have a slightly better pitch, but it's just a race to the bottom.

Sean Weisbrot: Mm-hmm. Where is the bottom?

Curtis Sparrer: I think the bottom is where everyone thinks everything is commoditized and everything is interchangeable, and we're seeing that you and I just now, when people are sending us these thoughtless AI pitches that, you know, butter us up by saying how great our last interview was, if even if it wasn't. And I think that the more people think about content that way, that the weaker it's going to ping with other people, which is ironic because the real gold rush right now is getting content that pings with journalists that gets published so that AI will recommend that company to someone when they search for the pain points that they need solving.

Sean Weisbrot: Then it makes sense why that guy the other day told me if I can figure out how to connect the interview to something related to GEO, that it could be a far more valuable service for people. Yes, because anyone can do an interview and anyone can make shorts, clips, but if the interview ties into something that the AI is really excited about. Then it helps them get more, a better ranking or, again, I don't really understand how that works, but, but it, it makes it more valuable so that, that makes sense.

Curtis Sparrer: Yeah. And I think the thing is, is that there is the electronic part that makes sense. And I should say it makes sense for now because it's always changing and hey, you know, when Chatt PT six comes out. Everything's gonna be changed. Again, it's just like SEO when Google would change its search parameters. So we have to go with what we have our understanding currently, and that understanding is informed by a lot of SEO right now. And that's the sort of north star that we have as we're mapping out this GEO universe. And you know what it's saying is earned media is the most important. What comes second to that is how you described yourself on your own media sites and owned content sites, and then it becomes a thought of, okay, if earned media is important, how can I really make sure that I'm knocking out the part there and how can I go after either high domain places or places with tons of industry credibility? If my own content's important, what can I do to make that owned content easy? Scannable, findable and useful for other people? And I think that companies and brands need to really invest in both sides of that side of that part of their house to make sure that they're coming off well.

Sean Weisbrot: I think I'm going to ask Gemini this based on what you just said. How can I do this based on my situation?

Curtis Sparrer: Ask Gemini.

Sean Weisbrot: Yeah, I'm curious to see what it'll say. I, I'm not gonna do it right now, but I'm curious to see what it'll say. I, that it's, it's kind of been my go-to. It's like as soon as there's an idea that comes to me so quickly, what does Gemini think?

Curtis Sparrer: We've done research that shows just how people are using the different AI platforms. And so people are using Claude, for example, for perhaps prose. Uh, that's what I do, and letters to the French in laws, but. A lot of people are using chat GPT, first and foremost, that's the favorite platform. We are also seeing that developers are increasingly turning to deep seek and they find that to be useful. And I think in the future it's gonna become a situation almost like Coke or Pepsi. What kind of guy are you? Are you a Coke guy or a Pepsi guy? Granted, we might have a few more flavors, but I feel that that's going to be kind of the knee jerk reaction about what people really care about and what you know matters to them, and how companies will then lean into that. So if companies discover that yes, deep seek is the place that developers go to, then there's gonna be a lot more thinking about how. Brands could really traffic themselves better and make themselves more useful for deep seeks, uh, various unique properties.

Sean Weisbrot: I have not tried deep seek. I lived in China for long enough to know that I don't trust it. Uh, personally I use Claude for any sort of coding things that I need to do. I use chat bt basically not at all anymore, and, and, and focus on Gemini. So I just think Gemini is really good all around, but I don't use Gemini for coding or anything like that, so I'm, I'm pretty strict on the split for those two.

Curtis Sparrer: I have tried Claude for coding and I am no Python expert by any stretch of the imagination, but I have found it to be helpful for someone like me. I do find it really weird when I'll try to run a program and it'll crash and I'll say, Hey, this crashed what happened? And I'll get an answer like, oh yeah, uh, I forgot to do that. And I think, how could you forget? You are ai. You have a transcript of this conversation since it started at 4:00 AM in the morning, but I definitely find it a lot easier to use than chat GPT when it comes to programming.

Sean Weisbrot: Yeah. I, I don't know why people use chat GPT for programming. I, I don't see it at all. I, I just think Claude is far superior. For that, and I'm also not a coder, but I understand business processes from previous experience with my tech company that I understand kind of what good code looks like versus bad code, how to build an art architecture. Not advanced at all, but probably better than 99.9% of the people that I talk to about ai.

Curtis Sparrer: I guarantee you, you're better than me. That is a certainty

Sean Weisbrot: fair. So what's the most important thing that you've learned in life so far?

Curtis Sparrer: To never believe that you've figured it out. Otherwise, don't buy your own pr, even if it's the PR for yourself. I think that. AI has been really interesting, uh, because it's made us all reevaluate our preconceptions about what we're doing, and it's made us have to learn and dig in. And I think that's a healthy exercise, but I think that's a healthy exercise for any profession, and I think everyone needs to really challenge themselves and think, is this, do I really have it all? Have I really figured it out? Because chances are the answer is no.