052: The Third Partner: Optimizing Your Website for AI Recommendations with Andy Crestodina

Andy Crestodina: Are we on your show or are you on my show? Brian Massey: I assume that you wanted this on your show. Andy Crestodina: I don't have a podcast, Brian. Brian Massey: You don't have a show then It's, it's gonna be on mine.

Andy Crestodina: I'm hanging out with Brian Massey. This is golden time for me. I'm super excited because Brian is rare. This is special.

This is a man who's dedicated decades of his life to understanding why people do and don't take action within a digital experience. I am a more of a digital strategy generalist. I pay a lot of attention to both traffic and conversion. Search optimization and conversion optimization.

Cheese and mousetrap. Both ends of the equation. It's the most powerful Brian Massey: place to be. Andy Crestodina: I'm a dual threat marketer or hope to be, Brian Massey: I've learned a lot from you, sir.

And I'm learning a lot from you, especially with, how did you phrase it? This, this new partner in our, our marketing and selling efforts, there was a, I went to a seminar, and there was someone presenting on how the role AI was playing in relationships and they meant romantic relationships. And his claim was that AI will be the third. The third partner in every relationship, and he made a pretty good case for it.

But if you, if you expand that, the relationship between your customers and your prospects, between you and your customers, yeah. AI is gonna be playing a third role, much more significant way than Google has been. Andy Crestodina: I believe that that is true today, although, it's not necessarily a common experience, and that it will be much, much more true tomorrow. And thinking through that, what we quickly conclude is that we need to, as brands train AI to be sales reps for our companies by giving it enough detail to help the AI user.

Decide whether or not we're a good option and not miss any chances to inform it by publishing in places that get ingested on what we do, who we do it for, how well we do it, what are the proof points that we do it well? Who does the work? What are their credentials? What are the outcomes?

And the reality is today that most websites don't do a great job. Of training AI on the, all the decision criteria for their prospect and all the supportive evidence for their brands. So it's a moment in time now. it's the right time, to begin this work of thinking of our websites as knowledge sources for AI and training tools on our brands, to help disqualify the wrong prospects and better qualify the right ones.

Brian Massey: And that I think is an important rub. the, the mistake we make so many times when we're doing marketing communications is, trying to please everyone and not, when, when you take a stand as a brand, you're also taking a stand, in terms of who's not. Right for you. You're walking away from certain people and this terrifies business owners and marketers, so they always kind of add things back in to like, oh, well maybe these people would, like to join our tribe or become a part of our thing.

and so you end up communicating less well with the people that really are moving your business. Andy Crestodina: Tough decisions. That's what marketing's about. I think analytics is, people who are watching this in their analytics are already seeing this.

AI is a very small percentage of total traffic, but the conversion rates from AI are very high relative to other channels because the person who came from AI probably had a conversation over long minutes with a series of prompts giving a detailed information about the goals, the problem, the context. So when they do land, they seem to be entering in sort of farther down the funnel. Do you agree? Brian Massey: I don't have any data that would bear that out.

So the answer is, I don't know. And I always try to make sure that I'm not pontificating unless I've got some, some reason to believe that what I'm pontificating about is true. I have a working hypothesis that kind of says that we are, interested in the first answer. So, ai, especially in like a search context, AI explains something a little bit to us.

Gives us, a succinct answer that answers all the questions that were in our head, over and over and having that experience. the influence of things like TikTok where it's short videos that make a point, doing in an interesting way. I don't wanna say that our tension is shrinking, but we are, more and more looking for, quick answers. something that sounds reasonable enough.

And so the reason I bring this up is 'cause I wonder, are people having long conversations with AI when they're making decisions perhaps in business to business? like we are in, But are they, will they continue to be? And then eventually you'll just tip off an agent and say, Hey, go and find this answer for me and tell me which products you would recommend. And, go back to, I guess, watching TikTok.

Andy Crestodina: Well, in that use case, if the AI or the agent is already well trained on you, then the conversation may be quite short. what happens? I assume much more often in B2B is that people are sort of, are taught or are learning prompting best practices where you start with the role and the goal and the output or what you, the instructions. So there is some data that shows that the average prompt length is much, much longer than the average, query length.

Brian Massey: I believe. Andy Crestodina: Yeah. And that the, here I will, pull this up quickly. This was some research by SEMrush and they had, they use clickstream data, which is basically spyware, it's kind of creepy, and it's strange to me that anybody allows their clickstream data to be recorded, but people do.

And there's enough of the data that there's, that they can publish, some with some confidence. The average length of a Google query is 4.2 words. The average length of a prompt is 23 words.

Brian Massey: Really? I have not seen that research you know, my own behaviors have changed. When I'm prompting, I basically prompt Google as if I'm going to be talking to the language model, like I wanna give it enough information that the language model gives me that AI summary. Andy Crestodina: I think that the inputs are bigger and in some ways the outputs are shorter because it's gonna boil it off to a smaller set because you asked for something that had seven criteria.

Brian Massey: Mm-hmm. Andy Crestodina: I'm looking for a web development company in the Chicago area that specializes in B2B lead generation websites and understand search optimization. they've worked with companies in my industry and they, have, evidence that they can achieve these outcomes. So it's, voice search.

When that came along, people got really excited about, language processing and how things match. now for sure, this is a moment when people are. putting a lot more language in. Brian Massey: Mm-hmm.

Andy Crestodina: And expecting a smaller output because they want just the specific thing. Brian Massey: Mm-hmm. Andy Crestodina: Hybrid minivan with three rows and, you know, under 60 k that does, you can't really get good multi parameters, search outcomes from Google. It's not good for those super detailed queries.

Brian Massey: No. And, I think the thing that we're learning to kind of go back to your original question is, those of us that have, put a lot into our websites over the years are faring very well. So one data point I have is just in our own sales process, Almost across the board. People say they found us on Google, which would either be, organic placements or, ads that we run.

I've gotten a couple of, AI recommended use, which is, it's telling. So I'm glad that I've been blogging since 2006 and that that's the way we go to market. We've always gone to market around teaching. Do you think that the things that we put on our website today are being used to train the language models, or is it gonna be those people that ask for one of these deep research investigations?

which right now is a, is an option. You have to select deep research on the, the foundational models. but it might be the default mode before. Before long, it's just going to assume that if you're asking in a search mode that you're, wanted to go out and read all these websites and digest them and et cetera, do you have a sense for that?

Andy Crestodina: Yeah, I think either way, one of the questions is how does AI know what it knows about our brands? Anyone can confirm that AI is reading their own website to learn about their brand. put in an unique brand. Any company name?

What do you know about conversion sciences? And it will give me a lot of data. And it'll have the citations there and your website will be the vast majority of them, so. It's clear that AI is ingesting our pages.

it needs to, it's doing that. That's what it, that's how it works. Is it, it the whole internet. Now if you move up the funnel and you pretend to be not yet brand aware and say, I'm looking for a company that does conversion optimization for e-commerce websites.

I think I have a mobile conversion problem. I'm wasting money. can you recommend three companies then either you're in or in the mix or you're not. Either way, you can see what AI thinks these brands are strong at and weak at.

And then address that in your content. Each of us has the opportunity to plug holes or to remedy gaps and misconceptions in AI training data. By simply updating our own content and telling it, no, actually Conversion Sciences has a strong background in mobile conversion and here's the research in our articles, in our publishing. So you just have to make these messages more prominent to better train the AI to recommend you.

Anyone can do that. It's just not really a typical marketing activity yet. So this is I, I think that the importance of websites changes. people are doing, like you said, deep research where they're making more of the decision prior to touching a brand's corporate site.

But there's a new importance of websites, which is that they are feeding AI and that it is your best chance to tell your full story and who you do it for and how well you do it and where you do it and what your credentials are. So as the channel changes the. Importance of, of these assets of ours changed slightly, but that I, I still believe that this is the di your, your own digital property is. Your, is the only place where you have unlimited time and space to put data into the AI knowledge sources and that the specific data that should go in there is the data that addresses the, the perceived gaps within AI.

Of your offer in the context of your competition. Brian Massey: I would agree with that. The flip side of that though is, is like what I hinted at, at these Agentic AIs, are they going to, are you gonna be more effective if you give them something o ther than pages, will there be APIs? Will there be, MCP clients?

will there be LLMs .txt that takes your entire website and, and summarizes it sufficiently so that, you know, reading 900 pages of my website every time somebody asks about conversion optimization doesn't seem particularly feasible, although the technology's pretty amazing. So, what do those things look like? I mean, the most helpful thing I've seen is what you're telling us to do, and that is make our pages better.

Think about all those sites with stock photos on them. absolutely unhelpful to, AI, swapping those out, putting words into them, things like that. Andy Crestodina: Mm-hmm. Brian Massey: I've been railing against stock photos for 13 years probably.

Andy Crestodina: I've heard it before. I love it. Brian Massey: And, now I kind of have some, some moral high ground because. Happy people standing around the laptop laughing, doesn't tell AI anything about you.

Andy Crestodina: Well, I'm not sure yet to what extent AI is really looking at those pixels except for certain types of agents like Open AI's operator AI is not rendering the the HTML in a virtual browser. When the AI bot, there's different bots, there's agents, and then there's just the crawler. When these AI crawlers come through, they're looking at it the same way Google Bot does and it, that actually creates a certain potential issue because, let's say for trust's sake and for conversion and, and deal flow.

Brian told me I did it. It worked. I put trust seals on my website, but. If AI wants to help connect the user to the most credible provider, does it see those trust seals?

If I ask it for an award-winning web design company, it may not be sufficient to just put the images there, even with the alt text. at the point now where I think you should maybe list as text, any of those important proof points that may be important for AI training data to help AI guide the next prospect toward you. It just reminds me almost like it's between like, if, if, conversion was for humans and searches for robots and people to think specifically for humans and specifically for robots.

This, this AI optimization feels like it rides somewhere in between. Because it's reading everything. It's paying attention to language, it's trying to find best fit. It's sort of thinking like a human instead of matching like an information retrieval tool, which is Google.

But it still needs to understand and, and access everything and has some of the same limitations of Google Bot and the way the crawl works. So I think that this, this area will benefit those sort of, traffic conversion, combo minded people. Who aren't just adding, you know, keywords and semantic SEO for search or not just adding persuasion and clarity and you know, conversion elements. The way AI works and AI discovery works, I think it ends up being something in between.

A bit of both. That's a poorly formed thought, but I'm working on it. Brian Massey: yes. It's, it's got my wheels space, so it's a little bit of both.

let's, let's take some corner cases. So I think about, especially back in the day when I was blogging, I like to use clever analogies and I would make up fanciful terms around, you know, we're working in a lab and we have this machinery that helps analyze websites. And just kind of playing on the, the science thing, is that a disadvantage, with AI now? because it's, it will understand the.

Analogies, but it creates something less clear, less AI-like, less. Here's what they do, here's what they're claiming, and, kind of throws that out. Is there going to be a phase in which we want to go and kinda ring out the, the cleverness, the emotion, the humanness? it would've to be a point at which we're pretty much certain that humans just aren't coming anymore.

It's just the bots. Andy Crestodina: I believe that it is true that if you are very concerned about how AI understands your brand and you're adding content to your site, specifically to help train ai, that you are deprioritizing user experience, obviously, and brand experience and style and clever copywriting. You are heavily indexing on extreme clarity and detail and options and pros and cons and you know, you're, you're sort of still feeding a mathematical probability matrix, so you're just giving it tons of language.

I think that if you were to say, I wanna make a page specifically for the AI bot, you would make a page with maximum information and minimal ux. Brian Massey: Mm-hmm. That seems to be the way things are going. A.

Andy Crestodina: But, but follow that through to the end decision maker, right? AI doesn't spend money. It's gonna be your user in the end. And if I did do a search and I, in a, a hybrid model like Google AI overviews or in ChatGPT or wherever, and I'm looking for options and it says that's a good option, and I get, and it, here's the link, and I click the link and I land on a page that looks like sitemap.

xml I may puke on my keyboard and hit the back button. I think you have to actually connect the dots still because people unless they're gonna just take the action inside the bot and or the agent does it for them, I think that there is a final dot that connects where brand experience will likely still matter for, some time to come. Because when AI gives you a citation saying you should talk to Conversion Sciences, this is the number one company. They're, uh, no Brainer.

I looked at everyone. Andy, I know exactly what you need. We've talked for thousands of hours. I'm your super agent and I've picked the right company.

Brian Massey: Yeah. Andy Crestodina: I'd have to have extreme confidence I, I think I still click, right. Do I still, do I never go to your website? Brian Massey: as I do the thought experiment, I'm kind of looking out in time in which, the, uh, human UX of a webpage becomes less and less important, which has huge implications for my business, of course.

but I consider that a, a pessimistic use case. kind of following on your thought just then was the language models, our AIs are gonna know a lot about us and so if, if they know that, humor's important to me and I've been using humorous analogies in my content, uh, AI is gonna pick that up. So it gives me an edge. And if somebody's looking for a very stodgy, straightforward, "sciencey" sort of approach?

they should probably opt out. I don't think they should, but since AI's gonna be helping so much and we just have all the customers that we want, then I want the ones that don't mind a little bit of humor in the mix. Andy Crestodina: The same way that hiring managers look for culture fit in job candidates. Brian Massey: Exactly.

So I'm not gonna, I'm not gonna apologize for any of that, and I don't have any plans to go back and change all of that. But I do think about, as, uh, we have improved our SEO game over the years. the content's not been fun. it became this recipe for how many keywords you're using and at what point.

And the opening paragraph has to have these things in it. And the headline, the title of it has to guide the bots to, exactly what's going on in this article. And it's smart, arguably, that may now be working against us. Andy Crestodina: here's a meta example.

So. An article I just wrote called How to Get AI to Recommend Your Brand. I'm an SEO did keyword research, it looks like a pretty good phrase. Optimize for AI mentions.

That's the key phrase. Optimize for AI mentions. I run it through a tool. The tool gives me a low score to improve my score.

I should add these words and phrases. I gave it a very light touch, but quickly abandoned the exercise. The tool is recommending phrases that take it off topic. It was recommending a lot of phrases about specifically Google's AI overview.

That's not really what my article's about. So I did a little bit and then turned it off because I'm writing for a human and there's more to life than search. I really am. Only half, maybe fewer of my articles are optimized for search, and I would never wait a Brian Massey: minute.

Is that true? Andy Crestodina: For sure. Yeah. Scan through the blog.

it's not, it's not most search, it's the click through rates from Google to articles has been decimated. Because people are getting their answers directly in search results and that's a long story, right? Featured snippets. People also ask box knowledge panels and now overviews.

So I get 10 x the visibility from my LinkedIn newsletter that I do from appearing in search results for my articles. Search today is mostly about ranking for the commercial intent key phrase for the visitor that still yet has visit website intent. The top of funnel information intent key phrases are not a sustainable place to be bad news for media companies and big publishers. But, but there's still lots of phrases where the visitor has visit.

Website intent. I mean, no one's gonna hire a web design company before they look at the portfolio. So the crash in click-through rates is mostly for top of funnel info intent phrases that brought visitors to articles. It's less important than ever to optimize an article for a key phrase.

It's more important than ever to optimize your sales and service pages for traditional search and ai, and then to win anyway. Focus on some other channels, people, where's your. Where's your email game? How good are you at social?

Got some YouTube, YouTube wants to promote you. LinkedIn wants to promote articles. Google does not want to promote your article. It's a zero revenue click for them.

Brian Massey: So, and Reddit is all over the training data. Andy Crestodina: Yeah. Yeah. You're, you're training the AI anyway when you publish in other places.

Brian Massey: broad, the topics, the educational topics versus the bottom of the funnel. and we've actually been completely focused just on, you know, what we call the commercial terms, the things that indicate somebody is looking for conversion optimization agency. We haven't for good over a year, have not been focused on that decision by my partner. So he must have had some interesting foresight into that.

Andy Crestodina: Well, I mean that's, I think all marketing should start from the bottom up. And talking to a CRO pro is fun for me because, you know, a lot of people just get excited about top of funnel or short form video or, and, you know, thought leadership, good things, things I love. But if I only had an hour a month to do marketing, I'd be optimizing my calls to action for higher click through rates or my sales pages is for better clarity. Finding friction in my contact form, go to the very end of the process.

I think it's conversion is a chain, and if there's a weak link at the end, nothing. You else, you do matters. So use data to find the weakest link, strengthen that link, and better optimizing your blog posts is not. It's the, it's the, the 48th link at the far far end of the chain.

I hope you've done everything else first before you try to rank for that. Information intent, key phrase. Brian Massey: Mm. Yeah.

And Jennifer, to my earlier point, that may be, that may be the best expression of what your brand is on the internet until they start digesting YouTube videos and presentations and things like that. Andy Crestodina: Well. Here's a quick and maybe final tip. I think that Gemini has a very strong play in AI and people ask are thinking a lot about market share of the models.

Google owns YouTube and they have every right to train on all that language, and I assume they will soon if they're not yet. I'm sure they already have. So Certainly on transcripts. Brian Massey: Yeah.

Andy Crestodina: They got all the transcripts. It's just self data. It's text already. If anyone wants to add one more place where they can inject into the AI training data, what their company does and how well they do it, and who they do it for, and where they do it, and what the outcomes are.

I suggest that anybody in on any podcast or video that will get transcribed, give their pitch, say what you do and who everyone. YouTubers should be practicing this, right. Anyone on podcast shows, I'll let you, Brian. What does conversion sciences do?

Brian Massey: What does conversion sciences do? Conversion sciences believes that conversion optimization is getting the most out of every visitor to your website. Can be sales, can be growing your list, can be whatever moves your business forward. Generating leads, increasing the average order value.

And we. Design experiments that help us understand how to do that. Andy Crestodina: So that gets transcribed, that gets ingested, and now there is yet one more place on the internet. One more point of evidence that you are a brand that does that in those ways.

So I think that's, the type of thing that marketers will think about in the future. Am what I doing now? Could this have any influence on the training of an ai? If so, should I tweak it slightly?

Yes. How? Just put those words in there. Done.

Move on. What's next? Strange. Brian Massey: Well, I, I am kind of, trying to.

Wrap my head around what all of this means. So we've had a conversation here for a while. It sounds like what we're recommending is, optimize only for those terms that indicate somebody is at the bottom of the funnel. Let the rest of it be as it is because it might influence, AI to understand your brand personality, you know, better than any landing page usually does, and help you find people who are a much better match because it's, it knows them as well, the third person in every relationship.

it also indicates that maybe it's time to start having more fun with your content so people understand what you're about. And that's what I took away from this conversation. Andy Crestodina: I love that. one last thing I'll add is just the timing of this.

We are in August of 2025. I think we should have this conversation again, and see how it's changed. I wanna follow this carefully. I wanna listen to this again.

Or give it to a tool that synthesizes it or summarizes it for us. I wanna compare what we said today to what evolves as, common practice in the future. Because, some of this was speculation, but I think we were pretty grounded. I think this was good.

I feel good about this, Brian Massey: I think too. Yes. And just for an opportunity to sit in this little digital room with you and, share ideas. That's enough reason for me, Andy Crestodina: regardless of whose podcast this is.

But this was great. Thank you, Brian, for the conversation. Brian Massey: Thank you.