Your Answer Has No Page Two: How AI Search Decides Who Gets Cited

DISCUSS YOUR BOOK

Robert: Thanks, guys. Richard, why don’t you? I’ll turn it over to you. And if you want to, you know, intro, share screen. I know you have a limited amount of time today. So let’s get on with it. Thanks. Appreciate you being here.

Richard: You should see my screen.

Robert: Yeah.

Richard: You all see it?

Robert: Yep.

Clicks Falling While Impressions Hold

Richard: Okay. So I wrote a book called The Day Your Website Died. And this isn’t an explanation of how LLMs work because you guys should know all that or you’re learning all that. But what I do know is how it works from the other end of the pipe. So you’ve got the LLM side and then you’ve got the consumer side. And I’ve got a website and a business and I’ve used AI a lot and I learned a lot of things in the last year and a half.

So I’m the guy whose content the retrieval system is deciding whether or not to cite or to pull. And things started looking different about two years ago, 18 months ago. And I started looking for an explanation. Of course, I found all kinds of sales pitches and things. So I discarded that and assembled a lot of data and then wrote a book.

And everything I’m going to show you here today is either measured, and I’ll tell you by whom, either now or in the book, or it’s inferred, and I’ll tell you that too. So let me give you some quick context. I spent 33 years in enterprise IT before I did this for a living. I ran computer operations for Trader Joe’s for 20 years of that. I know the difference between a system changing and a system being replaced, and this is the second one. The web is being replaced with this AI stuff.

And one more thing before I start, I sell ghostwriting, and I wrote a book about this. So I have an obvious interest. I’ve tried to build the deck so you can check every claim without taking my word for it, and the sources are at the end. All right. So that’s the name of my book. And pick it up if you want. It’s on Amazon. It’s very interesting. I think so anyway. Wait. There we go.

So the first thing I found is that impressions are holding or going up a little bit, but clicks are not. So clicks are falling on websites and they’ve been steadily decaying for about 18 months. Well, why is that? That seems kind of weird. At first glance it looks like my site’s been penalized. That’s the first thing I thought was, wow, my site’s been penalized. And that’s when you wake up on a Tuesday and you find out it’s down 30, 40, 50% and it stays there.

On the other hand, seasonality is a wave and it comes back. So this wasn’t either of those. It just went down on a graph like that and it kept going down and it’s still going down. So something is answering the query instead of routing. It was the conclusion I came to. So when a user says, I want to find a ghostwriter who knows technology, search engine would send it to me and I’d get lots of clicks on that. Now I’m getting impressions still, but no clicks or a few clicks.

The reason is they got what they wanted at the top, from the answer machine. So Google or all the other things are telling them what they want to know. So what does the ghostwriter charge would be one of the long tails that I wanted to rank for, and I was ranking for it. Now the AI engine just gives the answer. It doesn’t ever get to my site. And that’s why the clicks are going down. They never get to my site. They stay on the answer machine.

And that’s what you’ll see on websites and various other things. And just so you know, LinkedIn is having the same kind of thing. They’re changing their algorithm and the same thing’s kind of happening there. But that’s outside of the scope of this. And it took me about eight months to finally drill down into this and figure it out. All right. Next one.

Five Access Points, Two You Control

So there’s five access points. You’ve got the query, retrieving the query, the rankings, grounding, and the site. So where I can reach is the green ones, the query and the ranked chunks. I can influence directly whether my content enters the candidate set. And I can affect what a passage looks like when it’s being scored. That’s it. Everything else is out of my control. So that’s the way the industry is moving. And there’s an asymmetry inside of this second box. If a system is doing a live fetch, I have meaningful influence.

If we’re reading somebody else’s index, then I’m influencing that index and I’m hoping that it propagates. If it’s answering from weights, I have no influence at all on this quarter’s answers and probably none other. So your influence on the answers that your site being cited, no pun intended, on the AI engines is limited because you can’t affect these other things.

Never Retrieved, or Retrieved and Not Cited

So there’s two different failures. You’re never in the candidate set at all. And this is the one most people look at. They look at their .htaccess file, they look at, they’ve got queries failing, they’ve got JavaScript failing, their site’s got noindex tags, stuff like that. The AI engine is being blocked. This happened to a lot of people at the beginning ’cause they started seeing all these weird crawlers starting to affect their site, all the AI crawlers, and they just blocked them, well, they blocked themselves from the AI engines.

Now, then the other problem is, is you’re retrieved but not cited. So a passage that’s hedged, meaning it’s, I wrote 54 books is not hedged, but I wrote somewhere around 54 books is hedged. So you want to not hedge things. You want to say exactly what you do and exactly what your title is and exactly how long you’ve been in the business. So I put 54 plus.

I don’t have to keep changing it every time I finish a book for my number of ghostwritten books because I just finished four and I don’t have to update everything every time I sell a client. And that seems to work very well. If the claim is not confirmed by any other source, it’s a problem. AI looks at different sources and it pulls them together and it determines whether or not you’re trustworthy. You’re trustworthy if other sources cite it.

For example, if you’re on podcasts and it comes up, if you’re in Wikipedia, if you’re in some other type of website or citation, then this will show up good. The more citations you got, the better. And if it contradicts a higher level, higher confidence level record. So if you say I’ve been in the business 22 years, but there’s a record out there that says you’ve only been in the business 12 years, then and if that’s higher confidence, like say it’s a public record, then you’re not going to get cited.

And a corroborated competitor answered it instead. So if you say you’re the best ghostwriter in the world and another competitor who’s got more weight answers that they’re the best ghostwriter in the world, then they’re going to get cited instead of you, and that’s going to be a problem for you.

Branded Mentions Beat Backlinks

So look at the split, not the individual numbers. And you’ll notice that page is indexed. Well, let’s look branded web engines correlated 0.66. So that’s down here. Branded search volume is close to the same. So and then there’s the gap. Everything else drops. So backlinks don’t matter that much anymore. Referring domains don’t matter that much anymore. Pages index don’t matter that much anymore. What matters is branded search mentions and branded search volume. And this doesn’t say anything about content quality.

It doesn’t matter whether your writing is good or bad or written by AI or anything else. What matters is the branded search volume and the branded web mentions. Are you being mentioned somewhere else on the web by somebody else? Take this podcast. If this was a podcast out there on the web, and we were all listed. This would be a reasonably good web mention or search mention. And what this implies is it isn’t only your own website. Your website has to be there.

It’s your basis for being there, and it has to have all your numbers. But you need to have other things corroborate those numbers, lots of other things. The more the merrier. And that’s not what I wanted to hear because the only thing you really own is your website. You don’t own LinkedIn. You know, you’d be rich if you did. You don’t own Facebook. These are rented systems and that’s a big problem.

So you spend half your life trying to build up your LinkedIn thing and then boom, you lose it because for whatever reason, they don’t usually tell you why you lost it. And then you’re without LinkedIn. I know several people who’ve lost their LinkedIn accounts recently. They were never told why. Could have been a glitch in LinkedIn’s AI, who knows? And they never got them back.

So all of that work, years and years and years of work, hundreds of articles, if not thousands, and all kinds of other things gone in a flash and they can’t get them back. That’s why you base your root on the website. Even though it’s not a powerful AI thing because you wrote it, it’s important because you own it. So everything else can reference it.

Training Crawlers and Retrieval Crawlers

All right, you’ve got training crawlers and retrieval crawlers. And this is a problem that a lot of people have is they don’t differentiate between them. Your training crawlers are listed there. GPTBot, ClaudeBot, CCBot, Google-Extended. This removes you from future training, from future weights. So your articles won’t be used in the training or your images or whatever. It doesn’t have any effect on citations. You don’t have to worry about those. You can block those all you want. In fact, I do because I don’t want to help these things train.

It just uses my bandwidth. But then you’ve got retrieval bots. This is like OAI-SearchBot, PerplexityBot, Claude-User. If you block these, you’re blocking yourself from the AI answer engines. Don’t do that. You want to get the answer. You want to be the answer for the AI engines. And to give you an example of that, I’ve gotten several clients this year and I always ask, where’d you find me? And several of them said, over half of them said, well, I found you in Perplexity. It recommended you.

Or I found you in ChatGPT and it recommended you as number two. So I went and asked you, I didn’t like the number one guy. I liked you better. Or I found you in Claude. It’s very interesting, and that’s what you want. You want AI saying, when they ask the question, who’s the best of or who knows, who can I contact, who knows how to do agentic AI, then it’s going to, you want it to reference you. Very, very important.

And this is the way the web is going, so this is not something to ignore.

Four Kinds of Answer Engine

There’s four systems. You’ve got grounded live fetch. This fetches at query time. So if Robert, if you’re sitting there and you say, I want to find somebody who knows, who has worked at Trapezoid before and who knows this and this and this and this and this, or I know somebody who’s worked on machine code, then it will bring, this is a live fetch. It will go out to your site and it’ll come back. Right then and there and give you the answer. So this is important. This is great.

This is exactly what you want. Index mediated. This means that it reads somebody else’s index. So you inherit the index crawl policy and its staleness. This is second best. It works okay. Parametric first. You get answers from weights. It retrieves only on uncertainty and the training cutoff is your ceiling. So you get answers. Nothing you publish this week matters. This is over time. And then hybrid is not documented, and it switches strategy by query class with no published rules for each.

So the same query against all four returns four different citation sets. And there’s no shared ranking function to optimize against and no publishing contract that says any of it stays put. In fact, there’s no contracts at all. It just they all work differently, and they all use these these same kind of things before anything will cite you.

One Entity, Not Three

It has to decide whether you are the one resolvable entity and not three partial ones. This is really important. So high confidence. Several independent records agree on the same attributes. So like for me, I’ve ghostwritten 54+ books. I’ve got YouTube videos that say that I’ve got speeches that say that I’ve got all kinds of things all over the web that say that. It has high confidence that Richard Lowe has ghostwritten 54+ books.

So it knows that if somebody says, I want to find a ghost writer who’s written a lot of books, it knows that I’m one who’s written a lot of books. But if I have sites out there that say, oh. And this happens often because you do more over years. So when I started, I might have ghostwritten 12 books my first couple of years and then 15 books and then 30 books and then 50 books.

That causes disagreement and contradiction and the AI engine will get confused and you get kicked out of its answers. So an outdated job title on a conference page from 2019 actually winds up costing you more than not having a job title at all. So it’s an important thing when you’re out there trying to figure out why am I not getting traffic? Check all those things. Somebody says, I’m looking for AI engineer. If one of your job titles is not AI engineer, then you want to fix that.

That’s the most recent job titles. Of course, you can have a history like on LinkedIn. So corroboration compounds and repetition does not. So one source that’s repeated doesn’t mean much. You can see that the graph is flat. Independent sources that agree, so I have YouTube videos from other people, that’s important, YouTube videos of my own, my own website, other people’s websites, maybe Wikipedia if I’m lucky enough to get one of those, articles on various news sources, press releases, if all these things agree, then my graph goes up.

So the importance is getting agreement, not volume. And that’s a little bit different than it used to be. It used to be, oh, just get a whole bunch of sites that are backlinks. Get a thousand sites with backlinks and I’ll do it fairly well. No, three sites of having them agree is going to be a lot better than the thousand sites that say the same thing.

The Passage Is the Unit of Retrieval

The unit of retrieval is not your webpage, it’s a passage. A bad one would be, there are many factors to consider and the right approach depends on your circumstances, which we will explore below. That’s a typical AI thing to say, and it’s not going to get cited. If you want to get cited, you say a first PCI audit usually returns findings for controls nobody was assigned and not for controls that fails. Those are different problems. So that’s a little more specific.

If you want to even add more specifically, you could add more details about the PCI audit or about the controls that failed. The more specific you get, then the more the question is answered and the more you’re going to get the citation. But remember, chunking is done to you, not by you. So you want to write your article so that any 200 words lifted out of your context still assert something checkable. So that’s important.

AI is going to go through your article and it’s going to look for chunks that it can. You know how AI works. You get little snippets and it’s going to look for those little snippets. How hard is it to do a PCI audit? That’s actually a question that I answer in one of my articles. And I answer that question specifically. I have the question, how hard is it to do?

And then I give details on how hard it was for me and how hard, why it’s hard in just a few sentences, boom, that’s going to get cited most likely. It’s really up to it.

Freshness and the Hedging Problem

Cited content is fresher than organic results. So anything that’s dated is a weaker content than the same page undated. So don’t have dates on your pages is what that implies. Just say how hard a PCI audit is and don’t put in in 2019 or 2024 or whatever, because that makes it weaker. Revising a strong page beats publishing a new weak one. Just do revisions on your pages and keep them revising. But the parametric first engines invert this, their cutoff is fixed. Freshness is the per engine lever, not a universal one.

So there’s all these different engines and they all have different ways of working, but fresher and more precise is what you want. Dated and less precise you don’t want. And the typical AI tends to hedge. So when you write articles with AI, look for that hedging and tell it to stop doing that. Because AI does a lot of things that are hedges and you don’t want to do the hedges in your articles that you want cited by AI. Which is kind of interesting, because AI makes the content weak.

What You Can Measure When There Is No Rank

There is no rank. There’s no longer any ranking. So what do you measure? Well, you can sample your citation rate. You can run a fixed query set on different engines per schedule. You can record whether you appear or where you are in the answer. This is very problematical because this changes minute by minute, second by second. I can run a query in incognito Richard Lowe, kill the incognito window, bring it back so it’s fresh, run the same one, it’ll be different. So it’s not 100%, not even close.

And then you can look at competitors across the query set and decide which fractions of the slots belong to you. You’ll find it very noisy and hard to sample. And then referral fingerprints. This is the one that I find most important is traffic from assistant domains and queries inquiries, citing things that I never told them, things like that. I look at my statistics and find out, what am I getting cited on now? Who’s reaching my site and what queries are they using? That works pretty well.

And I look in my logs on their website for that. And single readings are just noise. You need to look at the pattern over many weeks. You can measure crawler access. So whether you’re named for a query at a point in time, and note that it’ll change over time, and correlations across brand samples. You can infer why one source is preferred, what triggers retrieval versus a parametric answer, unknown, citation policy, the ranking function over passages, and whether any of it’s stable month to month.

If anybody is telling you that they have a guaranteed claim about getting you cited and things. They’re trying to sell you something, and I would ignore them. Nobody knows yet, and they’re all different.

Where to Put Your Effort

So put your effort into auditing your crawler access. Look at. Look at your your logs and things. Deduplicate your facts. Remove the contradiction penalty across every claim. Rewrite your passages, because that directly changes what is retrievable. Remove ambiguity. Get corroboration from third parties over volume. Write for readers. Don’t write for the machines. If you write for readers, you’re going to write better, and you’re going to get more citations. If you write for machines, it’s going to kick you out because it knows.

So every figure on these slides and this deck and in the book, I’ve got an appendix that’s a third of the book that’s all of the stuff I looked at for this and all the articles I read. It was a pain in the butt. Like Ahrefs has a study of 75,000 brands, looked at my own crawler statistics, things like that. Nothing in here is proprietary or under NDA and you can use it all you want with attribution.

There Is No Page Two

Robert: Yeah.

Richard: But your answer has no page two. This is different than search. This is the most important thing of all. There is no page two. You’re either cited or you’re not. It’s not like Google where you’re on, you know, you want to get on page one. And this one, that’s the only place to get. That’s all there is. That’s what I got to say.

Reading the Logs With Claude and Cloudflare

Robert: So, Richard, on your, I guess, on your website, you know, a couple things, you know, it’s like, so I think one of the points you make is you should have a website if you don’t, if you’re interested in being in some of these findings. But you kind of mentioned looking at your logs and all of that stuff. I mean, where are you doing that? Are you hosting your site and all that kind of stuff? And how are you looking at your logs and things?

Richard: I buy hosting from SiteGround. On that SiteGround, they have a way to get to the raw log files. I just take that and I shove it into Claude. I say, analyze this for me. I’ve got big prompts that look through it and it says, it looks like all of a sudden ChatGPT is looking at your site and starting to give answers for it. Here’s why and here’s what I found. Then this is garbage. You should block this.

I use Cloudflare’s blocking capability to block stuff so they don’t get all this garbage in there. And that’s Cloudflare is awesome, by the way. I really like Cloudflare. I could do a commercial. But I host my own website on SiteGround. It’s only, I think I’m paying $500 a year. And it’ll be enough for anything, especially if you put SiteGround or Cloudflare in front. Cloudflare is a cache, has a cache. So you don’t need the big old package on SiteGround.

Because Cloudflare is taking all the traffic for you because it’s cached your whole site.

Robert: And so you’ve seen, I mean, it sounds like, you know, you’ve kind of been watching this for a while. And like, when did you first start notice like, hey, I’m not getting traffic on my site like I thought I was previously?

Richard: Yeah, I was getting traffic about two years ago. I was getting not a lot of traffic, but enough where I didn’t have to do a lot of prospecting to get clients. It just came to me because it came through the website. And then that started to taper off. And then I found that I was still getting clients, but they weren’t coming from the website. Website traffic clicks were going down. Impressions were the same. Clicks were going down.

So I started asking my clients and they said, well, I came to you through AI. And then it became obvious they’re not coming to me from Google anymore. Very few are coming through Google now. They’re coming, interestingly enough, through Bing, because some of the AI engines work on Bing as a base, and they’re coming through the AI engines. LinkedIn is the other source.

Contradictions Make You Unreliable

Robert: And I guess I’ll hand it out for other questions in a second. How, you know, you mentioned not contradicting yourself. Like if you say 54 plus books here, but you know, there’s some page on your website that didn’t get updated that says I wrote 12 books. Is that a problem?

Richard: That’s a problem. And the more of that there is, the more of a problem there is because I could see it as I was doing this. Perplexity showed me as not reliable because exactly of that problem. I had different numbers all over my website for all these things.

Robert: So you saw it said you’re not reliable as you were doing this analysis.

Richard: Said my numbers were not reliable.

Robert: Oh, nice. That’s nice to hear.

Richard: Which is, okay, I better fix that. Don’t want clients saying that and doesn’t say that anymore.

Robert: Right. OK.

Richard: And of course, there’s some stuff I couldn’t fix, old stuff that people no longer manage or people I couldn’t get hold of. That doesn’t seem to matter as much because my website is my home. And if I’m wrong on my website, what is that saying?

Robert: Cool. All right. Guys, any other questions for Richard? So the further away from your source website that the information contradicts, the less weight that has. So for example, if I put a blog post up on my website that I was at this meeting and I said, you wrote 12 books and that somehow got picked up, that’s gonna be, is that gonna mess up your contradiction factor?

Richard: Yes.

Robert: Interesting.

Richard: It might not be given a huge weight, but it’s one way, I think it’s a flaw in the whole system that they’ll probably have to address. But the competition can undermine other sites by doing that. And it’s not that hard. And I haven’t seen a lot of that yet. But I’m sure it’s coming.

Robert: Interesting. Yeah, Eric, I know. I know you’ve been doing, you know, you were doing market analytics for a while and have a marketing background. Any thoughts? I thought it was kind of interesting because, yeah, like I say, on my stuff, if I think about my LinkedIn, X and Facebook or whatever, I think they all say completely different things.

Richard: Of course they do.

Robert: And there’s no consistency, so I’m sure I’m not reliable. But yeah, Eric, any thoughts from you?

Writing Content That Gets Cited

Eric: It’s very fascinating. I’ve never been an SEO guy. It’s a different animal, but I’ve had to learn pieces of it with the PR work that I did, doing press releases and blogs and other content that I would create. I have not even looked under the first page of AI SEO.

Richard: It’s AEO.

Eric: AEO, thank you. I didn’t even know what to call it. So this is all very fascinating to me. And I’m definitely going to get a copy of this recording from Robert because you highlighted a lot of stuff that was brand new to me. I had a question for you about, and thank you for that, by the way.

I had a question for you about something you said, and I apologize if you defined this before and I missed it, but you said something, I’m going to mess it up now, but you said something about using a hitch or something like that in, I guess, blog content. Am I reading you correctly?

Richard: A hitch.

Eric: Or hatch or, I don’t know what you said.

Richard: What you want to do in blog content is you want to make it so that it’s citable in a short sets of paragraphs. So you want to break it down into questions. How hard is a PCI audit? What’s involved in a PCI audit? Why would somebody do a PCI audit? Your blog would be split up like that. So all of those things become questions that the answer engines are looking for. And you give them the answer right there.

And you want that first paragraph, first 3 or 4 sentences to be the answer. Then you can have a little explanation about it.

Eric: So it’s so it sounds like you’re it’s almost become more of a knowledge base than like a media approach.

Richard: Yes, yes, yes. So that doesn’t mean your whole blog needs to be that. What I tend to do is have half of those kind of things and half statements instead of questions, because otherwise it reads weird for visitors.

Eric: Sure. I know you’re a ghostwriter and you probably have some very strong feelings about this. As far as AEO goes, is it bad to use AI to create content?

Richard: Interesting enough, my most recent customers, I think it’s four or five of them, came to me saying they tried to write their book in AI, they hated it, and they wanted me to write it as a human. I’m getting more and more business and it’s more quality business because now they understand how hard it is than I was getting. It’s not turning out to be a problem for me.

Eric: It’s turning out to be business. For getting cited, is it better to use original content or AI content?

Richard: Well, raw AI content tends to be awful. It has that hedging problem. I think you said hedge.

Eric: That’s what it was. I said hedge. What does that mean?

Richard: Hedge is where you take the answer and you bury it in terms that make it less real. Like, a PCI audit is hard, except under these circumstances. If this happens, it’s not hard. However, under these conditions… All kinds of clauses that soften the quote, if that makes sense. So instead of saying PCI audits are difficult, and here’s why. You say, well, PCI audits are sometimes hard, and sometimes there’s a hedge. That’s a hedge word right there.

Robert: Yeah, it’s like, just freaking tell me. Don’t tell me, oh, but this, but that. Just give me the answer.

Richard: And that’s one of the main problems is using AI to write. And I’m giving you very, very tame examples here because I can’t think of any off the top of my head. But AI tends to hedge, hedge, hedge, hedge. Except for, if, if, if. These are AI constructs.

Building AEO Plugins With Claude

Robert: Richard, one of the things you talked about is when we spoke, you talked about doing stuff on your website in JSON, et cetera, plugins to improve AEO. So can you talk a little, I mean, I know you used Claude to kind of help build. So you were building WordPress plugins that help to optimize your site. Is this, it’s not on the pages in the site, it sounds like it’s in the source code or what’s?

Richard: Well, a plugin is a piece of code that you put into something that modifies it. So say in my website, I have one that does a table of contents and it improves WordPress to have this new feature. Well, what happened to me is my site got hacked and it got stopped right away. Boom. There’s some scanning software that found it and stopped it. And I was like, how did it get hacked? I’m a security guy. How can my site get hacked? Typical arrogance.

And I looked and it turned out that this whole package of software had been bought by another company that they had put in malicious software unintentionally and 600,000 sites got hacked within a matter of a month. So I thought, I don’t ever want to have this happen again. Claude, can you help? Sure, I can write some plugins for you. You can write plugins? This was news to me. Now I’ve got dozens and dozens of plugins that it’s written, like for table of contents and things.

What I do for AEO is I wrote one I call Discovery, which handles what’s called a schema. And a schema is what is this page about, but in language that the AI engines at Google understand. It’s a language, it’s called a schema. My book has a lot about that. And they have to match. So one part of your scheme is who wrote this page. Another part would be, what are all the FAQs on this page, frequently asked questions. Another one might be, what sites do I own?

So this would be my LinkedIn, my PR sites and so on. All these things need to be consistent page to page. Well, that’s a pain in the butt if you change something. So I made the Discovery plugin that injects that schema in everything else. So everything now has a consistent schema because it’s injected at runtime. Does that all make sense? And that improves AEO because it looks at my site and it says, oh, Richard wrote all these things. Cool. Now I know they’re all related. And oh, these are the FAQs.

Good. I got questions now and I got answers. Cool. That’s what I need. And this is all the other sites that Richard’s working with. Good. I like that. That means I can go look at those sites and find out more about Richard. And it has all that data in machine format. But it’s a nightmare trying to code that. So I wrote plugins to say, okay, I want a schema for the FAQs. Boom, schema in three seconds. Good. I want you to make a plugin so that it can do that.

So I don’t have to ask you that every single time I do a page. Good. And then I wrote another plugin that scraped every one of the thousand pages that I have and took all the schemas off and put them into a plugin so that they’re centralized because they’re all different because I wrote them over 10 years. So I used plugins to automate a lot of this stuff. And something that would have taken me months or years to do took hours.

Robert: Yeah. And so this is using Claude Code to build WordPress plugins to kind of optimize your site for AEO, et cetera.

Richard: It wasn’t even Claude Code. I just used straight Claude. You know, I just used the Claude chat. Claude chat. I guess it calls Claude Code. And then I always do one thing with it. This is important. So you finish your plugin and you say, okay, do an audit of this plugin. It goes, oh, this is bad, this is bad, this is bad, and it fixes them.

Go to another session, do another security audit, exactly the same thing, and it’ll find a few, and go to another session, and keep doing that until you’ve basically getting no more security problems. It’s because it’s an LLM. So everything’s different in every prompt. So you need to move to a whole new context and have it check again. But it’ll give you nice hardened code, very low cost. It’ll fix it right there. It’ll say, oh, I found this place where you’re doing this and you’re doing a SQL injection.

I’ll fix that for you. Well, thank you very much. Why did you put it in the first place? Well, I don’t know. I did. Okay, fix it. And then you go to another session. It says, oh, I found another SQL injection. Well, why didn’t you tell me the first time? I don’t know. And it fixes it for you. And then when you stop getting those, then you probably got it pretty hardened. In fact, it’s a lot more hardened than normal.

Robert: Okay. Awesome. Thank you. Any other questions, guys?

Can AEO Be Used Against Someone?

Chris: Yeah, I’m really curious. It’s fascinating on AEO and how you’ve been able to use that and how people can use that in the future. I’m curious what you’ve learned about AEO that can be used in an adversarial sense. You touched on it. Understanding how it works, how can you use it against a customer or a company, an organization?

Richard: Well, if you ask Claude this, it’s going to tell you, it’s going to say, I’m not going to answer that. It runs into its guardrails. But basically, theoretically, I suppose you could make a bunch of websites that have wrong data on it that you don’t own, theoretically, you know, they’re not yours. Or you could put in a bunch of articles that say the wrong thing. Think of how you would tarnish a person’s reputation. How would you do that? Well, you’d probably be interviewed on some major magazine and tarnish the reputation.

You might get sued, but still you’ve tarnished it. And that’s exactly what you do with this.

Robert: Interesting.

Richard: Yeah, and it’s hard to defeat. Other than doing lawsuits and taking down those articles and things.

Robert: I mean, it sounds like even if you were writing articles that had slightly incorrect information, it could impact somebody’s reputation.

Richard: Oh, yeah.

Robert: If you’re planting contradictory articles around different places.

Richard: Well, the key is you’re not writing articles. You’re getting a reporter to interview you for their articles. That’s even stronger. And that’s what you want to do to strengthen your site. Theoretically, you could use it to weaken somebody else’s reputation. And that’s what it’s all about now is reputations. The entire web has changed to reputation now. And I’m not sure I like it. But who’s asking my opinion, right?

Other People Citing You

Robert: How important are the sites that aren’t your owned content? X, LinkedIn, Substack, Beehiiv, whatever, Blogspot. How much do those have an impact? I’m assuming all of the AI crawlers, etc., etc. Are all hitting those sites as well, I would think.

Richard: They are and they have varying impact. Substack is relatively high, but it’s high when other people talk about you. That’s always the key. Other people talking about you, saying good things, hopefully. So if you get a testimonial from somebody, you want it on their site. You also maybe want it on yours, but get it on their site so that they say it. That’s stronger than you saying it on your site. And that’s the key to the whole thing. Other people cite you, not you citing you.

So on your substacks, get other people to like them, copy them, all that kind of stuff, and you’re going to see some growth. Be mentioned in their articles. Get into groups like this one. If this was public, it would be some good credentials because all of us would be in here and we’d have names and AEO would look at it and go, okay, these people know about this stuff.

Eric: Interesting, what you’re describing is a lot like what I’ve done for over 20 years with public relations.

Richard: It is.

Eric: Media relations, getting other people to talk about your company and driving that reputation.

Richard: That’s precisely what it is. In fact, using media in public relations is one of the keys, something that I’m having trouble with right now because I’m an introvert. Getting me to this show was hard. Having me talk on the show is not hard. I’m not shy. I’m an introvert. So I have trouble getting on these, these types of public relations things because that’s not my forte. I’m going to have to start doing that a lot more. Get some reporters to interview me, get on some media.

Quotable moments

Frequently Asked Questions

Why are website clicks falling while impressions stay the same?

Because the answer engine answers the question on its own page. Your site still shows up, so impressions hold, but the searcher gets what they needed at the top and never clicks through. A penalty drops you overnight and seasonality comes back; this is a steady decline that keeps going.

Should I block AI crawlers from my website?

Block the training crawlers if you like: GPTBot, ClaudeBot, CCBot and Google-Extended only affect future training. Never block the retrieval crawlers such as OAI-SearchBot, PerplexityBot and Claude-User, because those fetch your pages to answer live questions, and blocking them takes you out of the answers.

Do backlinks still matter for AI search?

Much less than they did. In the data Richard presents, branded web mentions and branded search volume track AI visibility most closely, while backlinks, referring domains and indexed pages fall behind. Agreement among independent sources counts for more than volume.

How do you write content that AI answer engines will cite?

Break it into the questions people ask and answer each one in the first three or four sentences under it. Be specific, drop the hedges, and leave dates off evergreen pages. Any 200 words lifted out of the page should still say something a reader could check.

Can a competitor damage your reputation in AI search?

In theory, yes. Answer engines distrust entities whose facts contradict each other, so planting wrong numbers or claims on third-party sites could undermine someone. Richard hasn’t seen much of it yet and expects it to come, and the defense is consistent facts plus third parties who corroborate them.