Latest / Found in AI: AI Search Visibility, SEO, & GEO / Writing for Humans vs. LLMs: What Actually Needs to Change?

Transcript

Writing for Humans vs. LLMs: What Actually Needs to Change? · Found in AI: AI Search Visibility, SEO, & GEO

• 0:41
  SPEAKER_00: Like, yes, you will get cited more if you play by those rules, but you know, not all citations are created equal.
• 0:49
  SPEAKER_01: Hey, welcome back to the Found in AI podcast, a show about AEO, GEO, SEO, and everything in between.
• 0:58
  SPEAKER_01: I am Cassie Clark, an AI Search Visibility Consultant.
• 1:01
  SPEAKER_01: And if you have spent any amount of time thinking about GEO lately, you've probably heard the advice to structure your content for AI engines.
• 1:11
  SPEAKER_01: You know, make it easy to extract, answer questions directly, give the models something they can actually use.
• 1:18
  SPEAKER_01: All of that stuff matters.
• 1:20
  SPEAKER_01: But at some point we still have to ask, are we creating better content for AI engines?
• 1:26
  SPEAKER_01: Or are we creating better content for the people who we actually want to reach?
• 1:32
  SPEAKER_01: My guest today is Ben Hale.
• 1:34
  SPEAKER_01: He is a content marketer who has spent more than a decade working across B2B software and content strategy.
• 1:40
  SPEAKER_01: And he's also been doing his own experimentation around what actually works in AI search.
• 1:46
  SPEAKER_01: Today we get into why treating LLMs as one giant search engine is probably a mistake.
• 1:51
  SPEAKER_01: How content should change depending on where it's distributed.
• 1:55
  SPEAKER_01: Why getting cited is only one piece of the measurement puzzle.
• 1:59
  SPEAKER_01: And what actually changes when you're writing for humans and AI engines at the same time.
• 2:04
  SPEAKER_01: And I think that last piece is pretty important here because content is not the only thing that makes an AEO or GEO strategy work.
• 2:12
  SPEAKER_01: There's distribution, authority, technical accessibility, third-party conversations, brand signals, and plenty more happiness around all of that.
• 2:22
  SPEAKER_01: But yeah, content is a pretty big piece of the puzzle.
• 2:25
  SPEAKER_01: So let's just get into it.
• 2:36
  SPEAKER_00: Yeah, I'm Ben Hale, uh, content marketer first, marketing generalist second.
• 2:41
  SPEAKER_00: Uh, I've been doing marketing in different forms and flavors for about 11 years, uh, primarily in software and primarily B2B.
• 2:51
  SPEAKER_00: But I've done B2C, I've done non-software, even done a few stints as an entrepreneur of a small business.
• 2:57
  SPEAKER_00: So kind of a wide range of experience, but content, I would say even within marketing, was my first love.
• 3:04
  SPEAKER_00: And I have done it in every single role I've had.
• 3:07
  SPEAKER_00: And I just love everything about the creative and strategic process when it comes to content.
• 3:13
  SPEAKER_00: I love doing the research, figuring out what works, writing, interviewing, all that good stuff.
• 3:18
  SPEAKER_01: Yeah.
• 3:19
  SPEAKER_01: So that brings us to the topic for today is writing for humans versus writing for LLMs.
• 3:24
  SPEAKER_01: So I know you've got some thoughts on this.
• 3:27
  SPEAKER_01: What are your initial thoughts before we really dive deep into it?
• 3:31
  SPEAKER_00: Yeah.
• 3:32
  SPEAKER_00: Um so I think when it comes to writing for LLMs, there are a few mistakes that we can make.
• 3:41
  SPEAKER_00: And the main mistake would be uh treating LLMs as one entity, a monolith.
• 3:48
  SPEAKER_00: Um, and I found that if you can focus in more with your search strategy, it actually then becomes easier to write for humans, also.
• 3:58
  SPEAKER_00: Um, and there's kind of three areas I would say that I've been learning to focus in more on.
• 4:04
  SPEAKER_00: Number one would be like focus on the kinds of queries that you're optimizing for.
• 4:10
  SPEAKER_00: Um, number two, focus on specific engines that you want to optimize for.
• 4:16
  SPEAKER_00: And then number three, uh being much more focused with your research because that kind of helps with content quality and it helps you write for humans, essentially.
• 4:25
  SPEAKER_01: Right.
• 4:25
  SPEAKER_01: So I think the research part is something that we've all kind of been doing for a hot minute.
• 4:31
  SPEAKER_01: I know that there's some that are just let me have chat GPT write it for us, and we'll just put it out there, and there's nothing in it.
• 4:37
  SPEAKER_01: Um, but let's talk about how to write for the various different engines.
• 4:43
  SPEAKER_01: So just fair fair warning for everyone listening.
• 4:47
  SPEAKER_01: We just had to re-record this, but we were just talking about how um each different like Google surface is different.
• 4:54
  SPEAKER_01: So, are you noticing that different kinds of content pulls into those different, like the AI overviews versus Gemini?
• 5:01
  SPEAKER_01: Have you noticed that?
• 5:03
  SPEAKER_00: Yeah, yeah.
• 5:04
  SPEAKER_00: Um, and it's interesting because I think there are trends uh that kind of do cross uh the Google platform specifically, right?
• 5:15
  SPEAKER_00: Like one I'll give for example is Google tends to like to cite Reddit a lot.
• 5:20
  SPEAKER_00: Um, there are other search engines, AI search engines that will do that as well, but there are others that will hardly cite Reddit at all.
• 5:26
  SPEAKER_00: Like Claude, I don't think likes Reddit very much.
• 5:29
  SPEAKER_00: Um so that kind of informs your distribution strategy, I would say, right?
• 5:34
  SPEAKER_00: Like if you well, hold on, let me back up a second to something I said earlier that I want to repeat.
• 5:39
  SPEAKER_00: Um I think it's really important to start uh with an understanding of your audience and how they search and how they consume content, right?
• 5:48
  SPEAKER_00: Um, each one of these search engines, and you can apply this within Google as well, right?
• 5:54
  SPEAKER_00: Um, each one of Google's different search engines has different rules, different algorithms.
• 5:59
  SPEAKER_00: Like I've already mentioned with the Reddit thing, different sources that they like to cite more often than others.
• 6:05
  SPEAKER_00: Um, so you really need to understand who you're trying to reach and where they are going when they're searching, right?
• 6:13
  SPEAKER_00: Um, you'll probably find that there are multiple engines that your audience uses.
• 6:19
  SPEAKER_00: I personally use quite a few, right?
• 6:21
  SPEAKER_00: So I think that's probably how most people are gonna be.
• 6:23
  SPEAKER_00: But there will probably be in the data that you can find um trends toward maybe two or three engines that you could focus on, I would say.
• 6:33
  SPEAKER_00: Um so that that I at least from a first principles thing, when it comes to writing for LLMs, I do think you have to start there, right?
• 6:42
  SPEAKER_00: Like, where are we going to focus and how are we going to optimize content for that specific engine?
• 6:48
  SPEAKER_00: Um let's see, you asked about within Google then.
• 6:52
  SPEAKER_00: Um I may have lost my train of thought.
• 6:56
  SPEAKER_01: Well, that's okay.
• 6:56
  SPEAKER_01: Let me ask you another question because I think this might clarify how I'm thinking about it before we go on.
• 7:02
  SPEAKER_01: So when you're thinking about talking about writing for LLMs, are you talking specifically about the blog content that you're putting out, or are you talking about the content that you're putting on these other channels like Reddit, LinkedIn, wherever?
• 7:15
  SPEAKER_00: Both.
• 7:16
  SPEAKER_00: Um, as a general rule, I would say that LLMs tend to weight third-party sources much more heavily than your own.
• 7:26
  SPEAKER_00: That's not always true.
• 7:27
  SPEAKER_00: I've seen queries where it was like 95% brand content, which was really interesting to me.
• 7:33
  SPEAKER_00: Uh, but as a rule, I mean, and this is just, I think this makes sense from a human behavior perspective as well, right?
• 7:40
  SPEAKER_00: Like if I tell you I'm great, you might believe me.
• 7:43
  SPEAKER_00: But if Cassie says Ben's great, well, oh, that's another person saying it.
• 7:48
  SPEAKER_00: Maybe there's more truth to that, right?
• 7:51
  SPEAKER_00: Um, so I do think you need to be thinking about this cross-channel for sure.
• 7:57
  SPEAKER_00: Like it does apply to your blog content, and you do need to keep it in mind for your blog content, but you also need to be thinking about distributing on other channels.
• 8:07
  SPEAKER_00: And that's like part of that's part of the um LLM strategy, right?
• 8:12
  SPEAKER_00: Okay, we know our users are mostly using Chat GPT.
• 8:15
  SPEAKER_00: So what sources does Chat GPT like to use?
• 8:19
  SPEAKER_00: And even you could, and you could and probably should get more granular of like, well, here are some of the queries that we want to optimize for on ChatGPT.
• 8:27
  SPEAKER_00: Let's actually test those and find and pull all of the sources that ChatGPT is citing so we can kind of like build our strategy off of that.
• 8:35
  SPEAKER_01: Yeah.
• 8:36
  SPEAKER_01: Okay.
• 8:36
  SPEAKER_01: So just wanted to make sure that's how you were separating out the two.
• 8:40
  SPEAKER_01: So let's talk about the other channels first and then we'll get into that blog content.
• 8:44
  SPEAKER_01: So say that you have a really great blog content, newsletter, something, and then you want to distribute, you know, you want to put that on Reddit or you want to put it on LinkedIn.
• 8:55
  SPEAKER_01: How are you changing the actual format of the content so that those AI engines pick it up better?
• 9:01
  SPEAKER_00: Oh, significantly, I would say.
• 9:03
  SPEAKER_00: Uh again, I think it comes down to understanding the rules of the platform and like the intent behind why people are using it.
• 9:12
  SPEAKER_00: Um, I'll go back to Reddit since I already brought that example up.
• 9:16
  SPEAKER_00: Um I well, I'll and I'll give you an example of something I worked on with Reddit data to kind of optimize it, right?
• 9:23
  SPEAKER_00: Um so I went out and did a product piece about uh what was it?
• 9:28
  SPEAKER_00: Pipeline management software, super sexy, right?
• 9:31
  SPEAKER_00: Uh part of my research process was going through all of the Reddit threads I could find about that specific topic, right?
• 9:41
  SPEAKER_00: Um and so those were kind of threads that I was also thinking back to back to in the distribution process of like, oh, we can and should probably show up in those conversations as well.
• 9:52
  SPEAKER_00: Um you cannot just take your product-centric piece and like copy paste stuff into a Reddit thread.
• 10:00
  SPEAKER_00: That does not work on that platform because people don't want to hear from your brand on Reddit, they are there to hear from other real people and get real thoughts and opinions and feelings, and they can sniff that BS out so fast.
• 10:15
  SPEAKER_01: Yeah, absolutely.
• 10:17
  SPEAKER_01: Like just a quick example, like I mentioned before we started having these headaches.
• 10:22
  SPEAKER_01: And so I was Googling today, is the magnesium glycinate making my headache worse?
• 10:28
  SPEAKER_01: Sure enough, I know, very specific.
• 10:30
  SPEAKER_01: Sure enough, there were like 10 different Reddit threads, and so I read one of them, and the very first comment was from like 148 days ago, and it was very clearly a product placement inside of the comment when everyone else was all giving, hey, this was my experiment, you know, experience with this.
• 10:48
  SPEAKER_01: That comment was try this specific one, and like all the comment underneath it was, this is such a weird ad placement.
• 10:55
  SPEAKER_01: So definitely don't do that, don't do that at all.
• 10:58
  SPEAKER_00: Yeah, that's that's a bad idea.
• 11:00
  SPEAKER_00: Bad idea.
• 11:01
  SPEAKER_00: So I think um, and even even within Reddit, like different um subreddits have their own sets of rules, right?
• 11:07
  SPEAKER_00: So you kind of have to play by those as well.
• 11:10
  SPEAKER_00: Um, but I there is still a way to show up and build kind of authority around your brand by just kind of offering helpful content, I would say.
• 11:21
  SPEAKER_00: Even if it's going to be product centric, like it can be, but you need to be very fair and balanced and open about like, oh, hey, I am building this thing, right?
• 11:30
  SPEAKER_00: Or I sell this.
• 11:32
  SPEAKER_00: Here are the kind of patterns that I've observed from my own research, right?
• 11:36
  SPEAKER_00: Like, um this is a rule that applies, I think, as well.
• 11:40
  SPEAKER_00: And in if you're doing like comparison product or content on your blog, right?
• 11:45
  SPEAKER_00: You don't want to talk trash on the competition.
• 11:48
  SPEAKER_01: Right.
• 11:49
  unknown: Right.
• 11:50
  SPEAKER_00: Because it's just yeah, sorry, go ahead.
• 11:52
  SPEAKER_01: No, I've I've seen different different people talking about this, where some of them would be, oh, we just paint them negative the whole time, and then others are no, we actually do a fair analysis.
• 12:03
  SPEAKER_01: And it's the fair analysis that gets pulled into those AI answers instead of just the negativity the whole time.
• 12:09
  SPEAKER_01: Because it seems like these engines are, hey, we we see this one as really just a promo piece instead of something that a reader would actually use to evaluate a product with, and so they're pulling in the one that is actually fair.
• 12:23
  SPEAKER_01: Um, so that's just something to keep in mind for the listeners out there.
• 12:27
  SPEAKER_00: Yeah, yeah, I think I think you have to keep in mind as well.
• 12:31
  SPEAKER_00: Um, again, getting back to basics.
• 12:34
  SPEAKER_00: What why are you people even using the LLM in the first place?
• 12:37
  SPEAKER_00: Right.
• 12:38
  SPEAKER_00: They have a specific problem that they are trying to solve.
• 12:41
  SPEAKER_00: So the LLM is incentivized to give them the content that best helps them solve that problem.
• 12:47
  SPEAKER_00: And it's doing that by just essentially measuring what people engage with and trust, right?
• 12:53
  SPEAKER_00: Like on Reddit, an upvote might be a trust signal, right?
• 12:57
  SPEAKER_00: So you are you are also incentivized to provide the most helpful content possible on that channel, right?
• 13:05
  SPEAKER_00: It's just it it's just a matter of understanding what works and what doesn't.
• 13:09
  SPEAKER_00: And you do have to test things, right?
• 13:11
  SPEAKER_00: Like there are there are rules, there are things that work and don't work, but it's also constantly in flux, and you kind of do have to test things and measure them as well, right?
• 13:21
  SPEAKER_01: Right.
• 13:22
  SPEAKER_01: So let's talk about this blog post for a moment because I noticed when I was doing a couple tests of how AI share voice changes when you do GEO specific updates on your blog content.
• 13:34
  SPEAKER_01: I noticed that some of the pieces that were being cited before were these long narrative prose pieces that might have some headings in there, but they were just really, really long.
• 13:46
  SPEAKER_01: When I updated based on like adding TLDR at the top, adding in um definition callouts, adding in those kind of things that made it look look more like a flashcard.
• 13:58
  SPEAKER_01: That's when the LLMs were like, oh, here's a better piece.
• 14:02
  SPEAKER_01: Have you noticed the same thing?
• 14:03
  SPEAKER_01: And how are you taking that into your content writing process if you've noticed the thing?
• 14:09
  SPEAKER_00: Yeah, that's a really good callout.
• 14:11
  SPEAKER_00: Like the structure of the content does matter.
• 14:15
  SPEAKER_00: This has been true with SEO, but there's just the rules are slightly different with AI.
• 14:20
  SPEAKER_00: Um, I definitely notice a difference in my like structured disciplined experiments with that, right?
• 14:29
  SPEAKER_00: Um you do get cited more if you're if you're playing by those rules.
• 14:35
  SPEAKER_00: Um I would also say though, like on this topic of writing for both humans and LLMs at the same time, those are the table stakes, right?
• 14:45
  SPEAKER_00: Like, yes, you will get cited more if you play by those rules, but you know, not all citations are created equal.
• 14:53
  SPEAKER_00: What happens after you get cited really matters.
• 14:57
  SPEAKER_00: Um, so it's it's a really good leading indicator of whether what you're doing is working, right?
• 15:03
  SPEAKER_00: But like you kind of also need to follow it down the trail to those lagging in indicators as well, right?
• 15:09
  SPEAKER_00: Like, I know traffic is not the end all be all anymore.
• 15:12
  SPEAKER_00: We're in the zero click era, et cetera, et cetera.
• 15:15
  SPEAKER_00: But at the same time, especially if you're focused on the bottom of the funnel and product-centric stuff, like if you're trying to show up in those product-centric serve uh searches, you do want that to be generating traffic.
• 15:27
  SPEAKER_00: It might not be as much as it used to be, but like you're trying to show up when somebody's researching a specific kind of product.
• 15:34
  SPEAKER_00: And so you do want them to go to your website and learn more and start the sales conversation, right?
• 15:40
  SPEAKER_00: Um, yeah, does that answer your question?
• 15:43
  SPEAKER_01: Yeah, which gets me kind of thinking though, about when you go in and you ask a question about okay, compare these two products or which one's the best.
• 15:53
  SPEAKER_01: A lot of times those answers do not have citations with them.
• 15:57
  SPEAKER_01: I mean, sometimes they do, but most of the time they don't.
• 16:00
  SPEAKER_01: It's it feels like the model just relies on the memory that it already has to compare those, which is a different strategy, I think, than trying to be the citation.
• 16:14
  SPEAKER_01: What do you think about that?
• 16:16
  SPEAKER_00: Okay, so you're kind of talking about like the mentioned citation dichotomy, essentially.
• 16:21
  SPEAKER_01: Right, yeah.
• 16:22
  SPEAKER_00: Yeah.
• 16:23
  SPEAKER_00: Um, I like citations because they are more traceable, right?
• 16:29
  SPEAKER_00: Um, one thing I was recently doing was just using Bing, again, this is kind of more before I was learning to treat AI search as a a monolith or as not a monolith.
• 16:41
  SPEAKER_00: Uh, I was using Bing Webmaster tools to track citations for careers, right?
• 16:46
  SPEAKER_00: Like if you don't have money to spend on like one of these fancy uh search visibility tools, that's probably the place I would recommend most people start just because it's available and it's good data for the platform, right?
• 17:01
  SPEAKER_00: Um, so but the assumption that I made there was like this is a proxy for all of the AI search engines, and that's simply not the case, right?
• 17:10
  SPEAKER_00: Like I it I that's what I've been learning is you just need more data.
• 17:15
  SPEAKER_00: Um mentions are harder to track, but also more valuable because of that, right?
• 17:22
  SPEAKER_00: I've been I haven't figured this out yet, but I've been experimenting with you you have to do it manually first before you automate it, I believe.
• 17:30
  SPEAKER_00: But just kind of setting up a process for tracking those specific queries to see if you're mentioned because that is that I've observed that pattern as well.
• 17:40
  SPEAKER_00: It really depends on the engine and on the query, even.
• 17:43
  SPEAKER_00: But um, there are times when there will be no links, right?
• 17:47
  SPEAKER_00: And how do you know if you're showing up in that search?
• 17:50
  SPEAKER_00: Well, you kind of just have to test it yourself multiple times, which is it takes a lot of work from what I'm seeing so far, but at the same time, like that's the way to learn it.
• 17:59
  SPEAKER_00: So you don't really have a good alternative.
• 18:03
  SPEAKER_01: Have you been testing out the kind of content that helps make a better mention versus the content that's a citation?
• 18:11
  SPEAKER_00: Ooh, I wouldn't say I've been testing that specifically.
• 18:16
  SPEAKER_00: I would be really curious to explore that.
• 18:20
  SPEAKER_00: I uh because I mean the the thing that sticks out to me when you say that is a better mention, like you I think you have to define that to begin with, right?
• 18:29
  SPEAKER_00: Like what is true, what is a quality mention versus just a citation versus an okay mention or a not that like a not a mention that we want, right?
• 19:29
  SPEAKER_00: Um yeah, that's interesting.
• 19:33
  SPEAKER_00: Yeah, I I wish I had a thought uh something concrete to share on that, but I that is something I would love to test.
• 19:39
  SPEAKER_01: Yeah, okay.
• 19:41
  SPEAKER_01: So back to like writing for the LLMs versus the actual human reader, like if we're thinking about just actual blog content, so not really something that you're putting on Reddit or LinkedIn or whatever other channel you're talking about, just the website.
• 19:56
  SPEAKER_01: What's the trade-off?
• 19:58
  SPEAKER_01: Or if you notice that there's a trade-off between I'm gonna put it in air quotes, normal SEO writing versus what we're doing now.
• 20:06
  SPEAKER_00: Trade-off versus SEO writing versus what we're doing now.
• 20:10
  SPEAKER_00: Um It's hard to say.
• 20:14
  SPEAKER_00: I think some of the it's a different world in the same world at the same time, right?
• 20:21
  SPEAKER_00: Like Google has pretty much always told you just make good stuff, right?
• 20:26
  SPEAKER_00: Like make content that your audience values and that is helpful to them.
• 20:31
  SPEAKER_00: I that that has not changed.
• 20:33
  SPEAKER_00: That's also the hardest thing to know if you're actually doing it or not.
• 20:37
  SPEAKER_00: Right.
• 20:37
  SPEAKER_00: Um yeah, I so I don't know necessarily that there is a trade-off.
• 20:45
  SPEAKER_00: I really think that the core difference or the the core function of search has not changed.
• 20:53
  SPEAKER_00: It's just the way that it works is what has changed.
• 20:57
  SPEAKER_00: And um you're really more the the the the core or the the main differences are more technical than anything, right?
• 21:07
  SPEAKER_00: Like the job that content is supposed to do hasn't changed.
• 21:11
  SPEAKER_00: Um, it's more just the format, the delivery, uh, the structure, essentially.
• 21:16
  SPEAKER_01: And where you where you put it, I think.
• 21:18
  SPEAKER_00: I mean yeah, distribution for sure.
• 21:21
  SPEAKER_01: Yeah, great content can live on your website, but the engines don't always look at your website, they're looking everywhere else.
• 21:28
  SPEAKER_01: Which is kind of rude when you think about it.
• 21:30
  SPEAKER_01: But I know, right?
• 21:32
  SPEAKER_01: That's what they're doing.
• 21:33
  SPEAKER_01: Okay, so so last question like when you're when you're testing all of this, how do you know what's a tactic that you should stick with versus one that's like, uh, this might just be a trend in you know, either distribution or actual content writing?
• 21:50
  SPEAKER_01: How do you know what's working and what's not is the question.
• 21:53
  SPEAKER_00: Oh, that's a really good question.
• 21:55
  SPEAKER_00: Um knowing what to stick with, I think you really just kind of have to, when you're planning out your experiments and your tests, I think you need to have time bounds, really.
• 22:08
  SPEAKER_00: Like this is something that's helped me at least, right?
• 22:12
  SPEAKER_00: Uh you need to make reasonable estimates here because you're not gonna know for sure.
• 22:18
  SPEAKER_00: But uh like if you're starting an experiment, you need to say, well, I think by month three, we will know whether this has worked or not.
• 22:27
  SPEAKER_00: And this is how we will know if this has worked.
• 22:30
  SPEAKER_00: We'll see these signals, right?
• 22:32
  SPEAKER_00: Um some in most cases I find that works, actually.
• 22:36
  SPEAKER_00: Um, another thing that I found to be helpful is kind of setting a significance threshold, I guess you could say, right?
• 22:45
  SPEAKER_00: Like we need X amount of data to be confident about whether this worked or not.
• 22:50
  SPEAKER_00: That's something I've used in my experiment tracking as well, just in terms of okay, I want to see 10,000 hits between these two groups, right?
• 22:59
  SPEAKER_00: Uh, and then I will have a relatively good amount of confidence about whether it worked or not.
• 23:05
  SPEAKER_00: Um what else can I say about that?
• 23:13
  SPEAKER_00: I don't know.
• 23:14
  SPEAKER_00: I I I think you just need to be as data focused as you possibly can.
• 23:18
  SPEAKER_00: Is it's easy to be precious about some things that you love, right?
• 23:23
  SPEAKER_00: Um things that you really believe in and that you think will work.
• 23:27
  SPEAKER_00: But if it actually isn't working, you need to be kind of detached from.
• 23:32
  SPEAKER_00: And be willing to kill that darling, right?
• 23:34
  SPEAKER_00: Um I'm trying to think of a specific example to illustrate that.
• 23:37
  SPEAKER_00: Uh I don't know.
• 23:40
  SPEAKER_00: I I I'm more just thinking about uh pieces that like writers really enjoy writing, right?
• 23:47
  SPEAKER_01: Yeah, yeah.
• 23:49
  SPEAKER_01: I have my favorites for sure.
• 23:51
  SPEAKER_01: Yeah.
• 23:52
  SPEAKER_00: But like you need to be aware of your audience too.
• 23:54
  SPEAKER_00: Like, do they enjoy reading that?
• 23:56
  SPEAKER_00: There's not those aren't necessarily mutually inclusive, right?
• 23:59
  SPEAKER_00: Um and the data is what tells you the answer to that question.
• 24:03
  SPEAKER_01: Okay, that was Ben, and I think the biggest takeaway from this conversation is that writing for humans and writing for AI engines probably shouldn't be two completely separate strategies.
• 24:15
  SPEAKER_01: The job of the content is still the same.
• 24:17
  SPEAKER_01: It's still to help someone understand something, whether that's evaluating their options or maybe even solving a problem.
• 24:23
  SPEAKER_01: What has changed is how that information needs to be structured, where it needs to show up, and how we measure whether it's actually working.
• 24:31
  SPEAKER_01: Now, before I leave you, I want to add one update because Ben and I recorded this conversation before what I have started referring to it as the Reddit drama.
• 24:41
  SPEAKER_01: Since we recorded, Reddit citations and chat GPT have dropped pretty dramatically.
• 24:47
  SPEAKER_01: We are still talking about this.
• 24:48
  SPEAKER_01: This happened maybe a month or so ago, but it's still all over LinkedIn lately and Reddit and X, wherever you're looking at this conversation, it's still happening.
• 24:58
  SPEAKER_01: Um, but that doesn't necessarily mean that Reddit has stopped mattering altogether.
• 25:03
  SPEAKER_01: I'm gonna I'm hoping I say his name correctly.
• 25:06
  SPEAKER_01: Suganthan Mohanadasin has been digging into ChatGPT's actual retrieval behavior, and in one of his recent tests, which was 84 of the 21 results, ChatGPT retrieved they were Reddit threads, but not one of those threads actually received a visible citation.
• 25:25
  SPEAKER_01: What's especially interesting here is that ChatGPT appeared to already have a short list of brands, then went to Reddit specifically looking for opinions about those brands before ultimately citing other sources, including those vendor pages, in the final answer.
• 25:41
  SPEAKER_01: Now, as of now, we cannot definitively say that those Reddit results change the answer, but it gives us a really important reminder here.
• 25:49
  SPEAKER_01: What an AI engine cites and what it retrieves are not necessarily the same things.
• 25:55
  SPEAKER_01: And if you are measuring your entire AEO strategy based on citations, you may be missing the bigger picture here.
• 26:03
  SPEAKER_01: So I would like to leave you with one thing that you can do this week.
• 26:06
  SPEAKER_01: Here's your one thing.
• 26:07
  SPEAKER_01: Go pick one high value question that matters to your business, use that as a starting point for your own experiment.
• 26:14
  SPEAKER_01: Look at the answer that you're getting today.
• 26:17
  SPEAKER_01: Which brands are mentioned, which sources are cited, what does the model seem to know about your brand and your competitors, and then choose one thing that you can actually change.
• 26:26
  SPEAKER_01: Maybe that's improving the structure of an existing piece of content.
• 26:30
  SPEAKER_01: Maybe that's publishing something that better answers the question or updating an older piece of content with better answers.
• 26:37
  SPEAKER_01: Maybe that's strengthening the comparison page or showing up on a third-party platform where you know your buyers are already talking about the problem.
• 26:45
  SPEAKER_01: Definitely document what you changed, give that experiment enough time to actually produce a signal, and then test the question again.
• 26:54
  SPEAKER_01: The goal here isn't really to reverse engineer Chat GPT from the answer sitting in front of you.
• 27:00
  SPEAKER_01: As that research shows, there can be an entire retrieval process happening behind the answer that we simply can't see without digging much, much deeper into the mechanics and the data.
• 27:12
  SPEAKER_01: So don't make the assumption that what the citation tells you why you appeared or the absence of a citation means that the source didn't actually influence that answer.
• 27:22
  SPEAKER_01: Just go make one intentional change, measure what happened, and start building your strategy from evidence instead of those assumptions.
• 27:29
  SPEAKER_01: And if you are listening to this and thinking, okay, but I don't actually know what's shaping the answers about my brand right now, that's what I look into in my AI Search Visibility Audit.
• 27:38
  SPEAKER_01: I dig into where your brand is showing up across AI Search, where your competitors are beating you, and those sources and signals that influence the answer, along with helping you identify the biggest opportunities across your content, authority, structure, and distribution.
• 27:53
  SPEAKER_01: Then you walk away with a roadmap for what to actually do about it.
• 27:56
  SPEAKER_01: You can learn more at Cassie Clark Marking.com.
• 27:59
  SPEAKER_01: There's a link in the show notes.
• 28:01
  SPEAKER_01: Okay, that's it for today's episode.
• 28:03
  SPEAKER_01: I will see you on Thursday.
• 28:04
  SPEAKER_01: Until then, stay visible.