Intro

How will generative AI change search and book discoverability in the years ahead? How can you make sure your books and your author website can be found in AI tools like ChatGPT? Thomas Umstattd Jr. joins me to discuss Generative Engine Optimisation (GEO) and how it will replace traditional SEO marketing.

I first covered this topic in Dec 2023, How Generative Search Will Impact Book Discoverability in the Next Decade. As ever, I was early, but those changes are now starting to happen. Thomas recently covered the topic on his Novel Marketing Podcast on Does ChatGPT Recommend Your Book?

This episode is supported by my patrons. Join my Community at Patreon.com/thecreativepenn 

Thomas Umstattd Jr. is the CEO of AuthorMedia.com, as well as an award-winning professional speaker, non-fiction author, and host of the Novel Marketing Podcast and the Christian Publishing Show.

You can listen above or on your favorite podcast app or read the notes and links below. Here are the highlights and the full transcript is below. 

Show Notes

• How authors can benefit from AI optimization
• Principles authors need to keep in mind as search is changing
• Different AI models and their capabilities
• Making your author website LLM-friendly
• How to utilize Goodreads to improve your GEO
• The future of AI agents in book buying
• Staying positive and curious in the ever-changing AI landscape

You can find Thomas at AuthorMedia.com.

Transcript of Interview with Thomas Umstattd Jr.

Joanna: Thomas Umstattd Jr. is the CEO of AuthorMedia.com, as well as an award-winning professional speaker, non-fiction author, and host of the Novel Marketing Podcast and the Christian Publishing Show. So welcome back to the show, Thomas.

Thomas: Thanks, Joanna, for having me.

Joanna: It's great to have you back. Now, for everyone listening, you were on the show a few years back. So we're going to dive straight into the topic today, which is based around a recent episode on your Novel Marketing Podcast on “AI Optimization For Authors: Does ChatGPT Recommend Your Book?”

I was like, yes, I really want to talk about this. Why did you decide to get into this topic now?

What did you see in the author community that made you want to help authors see AI differently?

Thomas: Well, what triggered this topic was actually the Google I/O Conference, where one of the features they were demoing was the ability to take a picture of a stack of books and then get recommendations on additional books that were like that book.

As somebody who spends a lot of time in tech world, books and authors are often the example that the tech people use to demonstrate new capabilities of AI models.

Often, unless people listen to your show, that new tech does not actually get translated to the author community. Most authors are not watching the Google IO Conference or even summaries of it.

Joanna: Except you and me!

Thomas: So I was like, oh, I need to do some tests with this. So I started testing different models to see how they would recommend books. I kind of realized, oh, this is already happening. People are already asking AI all the questions of their life.

Google search traffic is way down.

People are moving those big questions of their life conversations away from traditional search engines and towards AI interactions.

If you can get the AI to recommend your book, you'll be well-positioned for ongoing sales in this new era.

If you're holding on to ranking on Google search, or even Amazon search, as your only way of finding customers, sales are going to keep slipping, and you won't understand why.

Joanna: It's interesting. I have been using ChatGPT primarily since November '22 when it first came out. I use it instead of Google.

So I have started to use Gemini again, but I mainly use ChatGPT. Also on my phone, it's what I use. So what about your personal behavior?

Do you use a lot of AI for normal life that you once would have used Google for?

Thomas: I do.

In fact, AI has boosted my productivity so much that we've been able to launch a new podcast, a whole additional podcast, called Author Update. It is a news podcast once a week, just covering publishing news.

So much of the pieces of that, like taking the transcript and turning it into a blog, creating the timestamps for YouTube, creating the thumbnails for YouTube, creating the titles for YouTube. That's all done by AI.

Different AI tools that I've built for each one of those pieces that two years ago would have been incredibly time consuming. There would have been no way we could have added yet another show to the mix.

Joanna: I didn't know that. Interestingly, I have also brought back my Books and Travel Podcast, which I stopped doing a couple of years ago because it was too much work, and it's not one that's monetized.

I also brought it back in the last few months, because I was like, do you know what? I can now do so much of this with AI that it doesn't matter so much.

Actually, one of the things with that show which is interesting, is a lot of the times I'm interviewing people with different accents. A lot of the speech to text, the transcription previously, has been very good with American men, but it hasn't been so good with British women or anyone else of any nationality speaking English.

Now I find it's all very good. So it's like people who maybe last year might have said, “Oh no, this still isn't not good enough,” it really is now, isn't it, for a lot of use cases?

Thomas: Yes, there's a kind of person who tried ChatGPT when it first came out. They tried GPT 3.5, they played around with it for a couple of hours, they weren't impressed, and then they came to a conclusion.

The conclusion that they came to was not that this particular tool isn't ready, but instead, the category of AI is no good.

What they haven't realized is that so far in 2025 a new model that's the best in the world has come out almost every 10 days.

Almost every episode of Author Update we're like, “And there's a new AI model on the top of the benchmarks.” It's like they all take turns, and now they're starting to snipe each other.

So Gemini was number one for like two days, and then Anthropic is, like, “Here,” and pushed them off.

Joanna: “Here's Claude 4.”

Thomas: “You want to be number one. We're going to take that away from you.” If you were to go back and use GPT now, even the free version, it would be dramatically better than that first experience you had.

Really where the power is once you start paying for the AI models. Once you're using GPT-4.1 or -4.5, or Gemini 2.5 Pro. I really like Grok for research. I found that Grok's Deep Search functionality is unbelievable.

It has real time access to knowledge and real time access to X. So for doing research on basically any topic, Grok has won in every test that I've done.

Joanna: Oh, that's interesting. So I use o3.

My primary model is ChatGPT o3 for pretty much everything, unless it's just something very basic that I would Google.

Then I use Deep Research on ChatGPT with o3, and also Gemini 2.5. So I do use Grok, but only when I'm on X. This is interesting—we're going to come back to search—but interestingly, with all the stuff with the Deep Research, for example.

People listening, you get like, a 20- to 30-plus-page report on what you want to research with loads of sources and links, and most of them never, ever surface social media links. Grok on X obviously does, but that's the only one. So I find that really interesting, too.

Thomas: Yes. In fact, that was one of the things I researched for my episode on AI optimization. I was curious which social networks affect which AI models, because some social networks affect all of the models, and some social networks have impacts on basically none.

TikTok and Bluesky don't touch anything. You can be the biggest deal on TikTok, and none of the AI will know you exist.

YouTube influences Gemini. X is exclusively for Grok. Facebook and Instagram supposedly are tied to Llama.

Joanna: Who uses Llama?!

Thomas: Llama is so bad, it doesn't matter if it's connected to Facebook. Talking to all AI is like talking to a child, but talking to Llamas is like talking to a toddler that hasn't quite figured out how the words work and how the sentences work.

You can learn to understand it, but it's like, why bother when all the other AIs are like talking to middle schoolers who can now do research reports and are actually quite smart?

Joanna: I was going to say, yes, it depends on the context. Well, let's bring it back. You mentioned the Google IO Conference, and I also went to the overviews of that.

Sundar Pichai said a few things. I've just got a quote here. He said,

“AI overviews have scaled to over 1.5 billion users in 200 countries, driving over 10% growth in the types of queries that show them.” Sundar Pichai

So if people have used Google, I guess in the last six months, really, but a lot more in the last month or so, is if you ask something on google.com and then you will get this AI overview.

So you don't necessarily have to click into the article. So given that, I've heard it also called GEO, generative engine optimization, instead of SEO.

What are some of the principles that authors need to keep in mind if search is changing this way?

Thomas: So one of the fascinating things about AI is that it's very much a last shall be first and the first shall be last technology. So it's taken a lot of things that didn't used to matter very much, and it's making them suddenly matter a whole lot.

The two biggest winners of this new era is the author website, which has been kind of declining in popularity, particularly amongst indie authors because most indie authors are all in on Amazon all the time. They're not wide. They dream of maybe someday going wide, but the KU money is just too good.

So if your only existence is on Amazon, it's very easy to ask the question, why does my website matter? Now, the website did matter, right? Being able to sell direct was important. Being able to build your email list was important. Being able to communicate directly to readers was important.

There was a kind of author who's like, “Eh, I'm just on Amazon. I can ignore the website.”

Now your website is your primary way of influencing large language models that train on the open web.

You can't fully control Amazon, you can't fully control anywhere else on the web, but you can control your own website down to the robots.txt file. You have full control over it.

That is really, really important for educating an LLM about your book, and about your book's relationship to your other books, and about your book's relationship to the other books in the world.

So it's like, “This book is like this other book by this other author,” and your blog, on your website, is a really useful tool for that.

Joanna: You mentioned the word control, and that's exactly what I've been thinking about. Now, I've had my own author websites since 2008. You know, old school like you.

I also have Shopify stores. Shopify is actually interesting in that they are going AI first, and there are rumors of some kind of collaboration between OpenAI and Shopify in terms of surfacing direct links, which is interesting in itself.

So, yes, your control, your author website. Also we've seen—well, we're going to come back to Amazon—but they're doing a lot of things with their own AI.

What are some specific things that we can put on our author websites? I mean, if I say, okay, so I've got an about page, which is about me. Then do I have a book page?

On a book page, what are some of the things that I might add that the LLMs would be interested in?

Thomas: So here's the classic mistake.

An author gets started writing, and they have “Home”, “About”, “Book” and “Contact”. It's kind of the classic author website. Then they write a second book, and they're like, oh, well, I need to put this new book at the top of my book page, and I'll rename it to “Books”.

Well, that is a blunder, believe it or not, because now you no longer have a page dedicated to either one of the books. So you've done the new book you just wrote a disservice, and you did your existing book a disservice.

So one really easy change that many of you listening can do right now is you just create a new page for each book, and you copy and paste the content from your Books page into each individual's book page.

Then you make your Books page just a bunch of thumbnails for your covers. Big, beautiful covers, even bigger and more beautiful than the thumbnails on Amazon. They click on that cover, and it takes them to an entire page just about the book. So that's step one. You can do that in 15 minutes.

Step two is now realizing this page isn't just for my readers trying to decide about the book. This page is for large language models trying to understand my book. So you want to actually make that page as rich and as in-depth as possible.

You also want to make it really good for humans, right? So put discussion questions, have sample chapters, have your audio book resources. So I'm a big reader of fantasy, but I listen to fantasy books, and I really want to see the map.

When I go to an author's website, it's some low res garbage map, and I can't see the towns, and it makes me very sad. All I want, fantasy authors, please, for the love of good maps, just upload a five gigabyte version of your map to your book page.

I will love it. AI will love it. Your readers will love it. It will make everyone happy, and it already exists on your hard drive. It's what you put in your book. It's not going to keep anyone from buying your book, the fact that they can get the map of your fantasy world for free.

That's just one example of the sort of thing that you can put on your site. Also, frequently asked questions. If you do frequently asked questions, there's a Schema, Schema.org that you can add to a page through Yoast SEO. So if you're using WordPress, it really is the best for this sort of thing.

It's called a Question Schema, where it will actually surface that question, not just on Google search, but also to the to the LLMs, where they'll see the question and then see the answer. This will really reduce the likelihood of the LLMs hallucinating if somebody else asks that question, or a similar question to the LLM.

Joanna: So just on that frequently asked questions.

We just mentioned the AI overview on Google, if you have a frequently asked question on your website that it can easily pull from, it is more likely to do that.

Also useful is NotebookLM, where you can upload your book and it can actually generate those frequently asked questions for you.

So this is another thing I would say. I mean, again, read the terms and conditions but NotebookLM, in particular, says it doesn't train on the data you upload into a Notebook, if people are worried about that. You can actually use the AI tools to help you build this material.

The other thing I was going to say on images, one of the things I was reading about is the alt text. Now, the alt text on images, we've been encouraged to use for accessibility reasons. So if somebody is blind or partially sighted, the alt text gets read when there's an image.

Alt text is used by the LLMs when they're going through a website.

Yes, they can “see” now, but they use the alt text. So is that something that you've considered? Because I guess I didn't think about that before.

Thomas: So this is one of the techniques that I think is helpful right now and won't be helpful in two or three years, because this is purely a way of you adding human labor to your website to save the bot from doing bot labor.

You can upload an image to GPT's Image-1 Engine and ask it to describe it, and it can describe that image with paragraphs and paragraphs of detail, but for GPT to do that, it requires a lot of compute. They don't have the compute to do that for all of the images on all of the websites on all of the internet.

Now, the compute cost is going down. You know, more efficient chips are coming out. The models are getting more efficient.

So several years from now, the AI will be able to just go to a page, look at the image, and generate a much more useful understanding of that image than what it can currently get with alt text.

In the meantime, adding some descriptive alt text could help before it's understanding the image. Also, not all LLMs are multimodal, which I realize is a big term. So multimodal is being able to interact with text and image at the same time.

So GPT is, I would argue, the most multimodal. It's just unbelievably good with images. I'm not a big fan of GPT in general, I find that the other models are better at most of the other specific use cases, but for images, it is just hands down the best.

It's often the second best in every other category. So it's a good one if you're only using one.

Some of the other engines aren't very good with images yet. I haven't been impressed with Claude's handling of images. Grok is only kind of so-so. Gemini has made some big steps forward, but I still think it's behind GPT in image rendering.

So you're also helping these other models more because, you know, Anthropic may not be able to describe the image in a very suitable way right now. It will in a few years, but right now, maybe not so good.

Joanna: It's interesting you say that. I think ChatGPT o3, that is my favorite model. I don't really use the 4-models. I also think where if people are saying, “Oh, well, you know, Thomas doesn't rate it,” well, I think everyone prefers different models as well because of personality things.

A lot of writers like Claude, for example, for the more creative side of things.

As we've also said, If you don't like a model this week, try again in a couple of weeks, and it may well have changed.

I mean, GPT-5 is rumored to be coming out, which I think will be interesting. One of the things is, you and I are quite technical, so we're like, “Oh, this number and this letter,” but GPT-5, apparently it will do that for you. So you'll just put your query in, and it will choose the best model for you, which I think will really help.

Thomas: Yes, one thing to help simplify this, because GPT has probably the worst naming schema in the history of naming. So they have GPT-4o and GPT-o4, which are entirely different models and have almost nothing to do with each other. Then they have 4.1, which is actually better than 4.5. So the numbering doesn't work.

Then o3, which is based off of 4, is actually better at a lot of reasoning tasks. It's very confusing. So let's simplify it in a way that actually will help across all of the companies.

There's kind of three main flavors of LLMs in terms of main features, and that is the kind of default model, default model with reasoning, which is what o3 does really well because it can actually think about your question.

So if you think about if somebody asked you a question, you can answer off the top of your head, or you can sit down with a piece of paper and kind of think about it a little bit. That left brain slow thinking is what we mean by reasoning. When you're interacting with a reasoning model, it's slower to get back to you because it thinks about it.

Then the third kind of model is deep research, where it will actually go and do research. I don't know if you ever do this, Joanna, but on a live call, somebody will ask a question, and you'll do a quick Google search to refresh your memory about that thing that they're talking about. That's kind of how the search functions.

It's called Deep Search on Grok. I think it's called Deep Research on GPT [and on Gemini]. Those three features are rolling out to all of the different models, and they're useful in different ways.

So if you want a quick answer, you just want to talk to the core model, but if you want some deep, in-depth analysis, you want to turn on research, or maybe turn on thinking as well.

That'll simplify it to make it not quite so confusing because if you're not following this every week, the numbers and the letters and the models and the companies will just make your eyes water. It's so complicated.

Joanna: Absolutely.

Or my tip is, whatever your favorite model is, you just say, “I want to do this. How can you help me do it?”

Most people aren't as technical as we are, so they won't necessarily be driving the machines in that way.

Let's come back to the website. So I agree with you that sort of the last shall be first. So the author website has sort of fallen out of favor in many ways. For example, blogging.

I was blogging from 2008, and then about five years ago I stopped because there were some really, really good websites doing the kind of content I was. So I was like, right, I'm just going to do the podcast.

Of course, for our podcast, we have transcripts and all that. I thought, well, that gets indexed, and my site does still rank for lots of good things because of the podcast transcript.

So if people are now thinking, okay, well, if these AI engines want this rich content, but we don't want to upload our books onto our website, for example, what are some of the other things they could put on the website? Is it just the book page, or—

Could people be thinking about other forms of content on their sites?

Thomas: So blogging is really powerful, and I will share this with your audience. I left it out of the blog version of my episode on AI optimization, but it's in the audio and video version.

One of the big things that the LLMs look for when it comes to ranking a book is something called context, where it's in relationships, specifically. So relational context is really important for LLMs, and you can guide that with a blog post.

So you can say, “The top 10 posts on such and such trope” or, “The top 10 authors who are similar to JF Penn.”

“So if you love JF Penn, you'll love…” and you just got these other nine authors and includes JF Penn. So if I'm writing books that are similar to JF Penn, I would include my name in that list, and then train the LLMs to start associating our names and putting them in a semantic cluster. A blog is really powerful at this.

The other really good thing to do with blogging has been the best thing to do with blogging for the last 15 years, which is just answer questions. Your inbox fills up with questions, and so you just write one really good answer, you email it to that person, and then you copy and paste it to your blog.

Take out all the personal bits, add some bullets, add some headings. Now you've got a really good blog post that already existed in your outbox, that you know for sure a real human being asked. If one human being asked it, probably others asked it. If they're not asking Google, they're asking the LLM.

It's not that much more work to have a blog of some kind, the topics of which are driven by your own readers.

Joanna: You mentioned there, it's easier with nonfiction because people will ask questions about that. For example, on my Books and Travel, I did the Camino de Santiago, and people email me all the time saying, like, “What shoes did you wear for the Camino?” I mean, just a question like that.

It is in my book, and I have actually put it on the website now, but it's interesting because that's easier for nonfiction/memoir. For fiction, like you said, I have done blog posts in the past like, “10 Action Adventure Series with Female Main Characters,” stuff like that.

This is what I was also wondering, because if you use any of the LLMs, and you say, for example, “What do you know about author JF Penn?” and it will kind of look at everything.

I found that Goodreads is actually incredibly highly ranked.

I wonder if that's because a lot of those posts, like you're saying, are often on Goodreads, their blog. That's literally what their blog is. They're always posting lists of relational things, and obviously they're owned by Amazon. What do you think about fiction authors in particular?

Would it be better to be posting lists of that kind of thing on Goodreads and/or their website?

Thomas: I love Listopia. That's Goodreads' list feature. I don't think authors are allowed to add their own books to lists on Listopia, which means you'd have to work with a compatriot to add each other's books into the list, which adds a little bit of friction.

Goodreads has become incredibly important because Goodreads is one of the only places on the internet that has Schema.org information on books. There's actually no good way to add this to your website right now.

This is making me feel like I shouldn't have given away MyBookTable, which is a WordPress plugin for making book pages that I developed years ago and I'm no longer a part. Yoast SEO doesn't support the book schema, but Goodreads does.

So Goodreads has become like the go-to source for metadata and context and information about books. It also has reviews and rankings and relationships. It can look at shelves and which books are connected with which other books and shelves.

It's actually really rich data, and unlike most other social networks, it doesn't have a login wall to access pages. So you can go to any page on Goodreads without being logged into Goodreads, which means there's no good mechanism to keep the bots away.

Having a Goodreads profile at 100% is really important.

You're like, “But I never use Goodreads, and my readers don't use Goodreads,” like, well, some of your readers do. The mega-readers, the readers who take chances on new authors, they're all over Goodreads.

If somebody reads 300 books a year, they need Goodreads to find that 301st book. If somebody reads one book a year, they just go to the bookstore and buy whatever the James Patterson book is that's facing the door.

So if you're new to writing and you're still just getting started, Goodreads was always important to you, but now it's even more important, because now Goodreads is informing all of the networks.

So when I was doing my research, every single large language model—I don't know about Llama, I don't really care about Llama—but all the ones that matter, they all look at Goodreads quite a bit for informing their context about books.

Joanna: That is quite shocking for some people. You know, when I started in 2008, Goodreads was a separate company. It was really big. It still looks the same as it did.

Thomas: It's like a time capsule to the days before social media got toxic.

Joanna: It really is. It's quite horrible. So I guess maybe a decade ago, I was like, okay, I don't really want to use this anymore. Also, a lot of us were focusing on going wide and building Shopify and all this. Then a couple of years ago I saw that, oh my goodness, Goodreads is becoming more important.

So I've really been making much more of an effort and asking people who buy direct to also review on Goodreads. Of course, let's say you read on a Kindle, you read an Amazon device, if you rate a book at the end, that will automatically appear on Goodreads if you've connected your account.

Even if people aren't writing reviews, all these ratings is another data point that does all the linking, like you're saying. So I can't see that another site can be as rich as Goodreads in the English language, I guess we should say. I think Goodreads is only in English, as far as I know.

So would this be more important than the author website updates? If people are like, oh my goodness, you two, you've just given us too much work—

Should people be thinking about updating Goodreads first, or the website?

Thomas: I think for most people, starting with Goodreads might make more sense because chances are your Goodreads page is already half built because Goodreads pulls data from Amazon.

So if you did a good job with your metadata and having a good Amazon page—which if you've been listening to Joanna Penn's podcast for any amount of time, you've heard her harp on.

Joanna: Harp on?!

Thomas: You've heard me. I feel like I'm mentioning my metadata episode every single episode. I'm like, “Please. This is so important.” I don't know if you if you harp on it, but I definitely harp on it.

So if you've been doing a good job with that, a lot of the Goodreads stuff is already done. So it's just logging in, making sure account is attached to your author account, making sure all of the information is correct, tweaking the things that need to be corrected. You could be done in an hour.

Building out these web pages could be done in an hour or two if you're savvy, but if you've never edited your website before, there's actually a bit of a learning curve to do it the first time. If you had somebody build you your website, now it's more complicated because you've got to go find that person and pay them.

So the website could be a higher amount of work, but it's still really important. So don't hear me say, “Oh, Goodreads, start there,” as an excuse to then stop there. Do your website as well.

Joanna: I guess we should also say that it's early days. In 2008, do you remember back then it was, “It's the year of mobile,” or, you know, 2010, 2012 was still the year of mobile. Like it was the beginning of mobile commerce and all that.

Nobody believed it for years, until one time everybody woke up and were like, “Oh, yes, you buy things on your phone. I suppose that's what they were talking about.” This is the same thing. I mean, this is going to grow.

So right now, we're still early, I think, on this. So yes, have a look at your Goodreads. Have a look at your website.

Let's carry on. So you did mention norobots.txt earlier, which everyone's like, “Oh no, no, that sounds complicated.” Or they're saying, “Well, I don't want things to search my site. I'm against AI, or I don't want them to see my website or to search things that I've spent time doing.”

What will definitely stop the AI search engines? Why should we not do that?

Thomas: I think that some of the “all AI is evil all the time,” is actually being advocated by people who themselves use AI and don't want other authors to have the competitive advantage that they have.

I don't think it's all of that, but I think that some of the most vocal people secretly have pen names where they're making a lot of money with AI everything. It's kind of like you're the first farmer in town to get a tractor, and you're way more productive on your farm than all the other farmers who are still doing it with their hoes and their backs.

If you can convince the other farmers that tractors are evil, or if you can get somebody else in the town to do that for you, then you can buy the fields from everyone else who's doing the work with their backs.

So I'm not convinced that this fear is all in good faith. There is some of it. You know, some people all they know about AI is they watched The Matrix and they watched The Terminator, and it's really scary.

Those people tend to not be very vocal because they don't know the difference between a large language model and machine learning. Like in AI, it's all just a bunch of jumble for them, and it's all scary and evil and strange. That kind of person isn't going to know how to put a norobots.txt file on their website.

I don't think that LLMs.txt or norobots.txt are going to go anywhere. I think what's going to happen with AI is the same thing that happened with mobile. So back in 2008 when it's like the year of the mobile, if you remember, we were building mobile versions of our websites.

So you had the website, and then you had a completely separate website for mobile. Then in the early 20-teens, this new approach to web design called responsive web design was developed, where everything had percentages instead of fixed number of pixel widths. Pages could get big and they could get small.

Now you only had to build one version of a web page for both mobile and for web. The robots.txt file has plenty of space for instructions to LLMs and what you can and cannot do. We don't need other txt files on a website to accomplish those purposes.

I think it's simpler for everyone to just use the page we already have for talking to robots, rather than having other pages to also talk to robots, but say different things. So I may be wrong on this, but I don't think the alternate files, norobots or LLMs, are going to take off.

I do think people are going to add instructions to robots.txt, and some of them will be encouraged by other authors, well-meaning or not, to start blocking the LLMs. I don't think that's going to work. For one, I don't know how you block Google without blocking Google. It's like, “Are you the Google search bot or the Google Gemini bot?”

“I'm the Google bot,” right?

Do you really want to block Google? You can. It's your right, and Google won't surface your website, but you have to be found if you want to be read.

You can't hide in the wilderness and not let anyone read your book and then complain that no one's reading your book.

You either want to be found or you don't.

I think for some authors, they're afraid of being found, and they may be using AI as an excuse. I think pretty soon, people are going to be paying money to get AI to learn about their books, this fear of AI doing it for free, I think, is going to go away pretty quickly.

Joanna: Well, yes. We should just come back to Amazon because people are like, “Oh, well, I don't need to know any of this because I am just on Amazon,” but of course, Amazon is also moving to generative search.

I'm kind of annoyed because this mobile app that you have in the US with this Rufus shopping bot is now using it, and you can ask really detailed follow up questions, and it's more granular, but I can't see that because I'm in the UK.

I mean, I'm really pleased about this because I am sick of loading seven keyword terms into my metadata. If I give Amazon the whole book, I mean, surely they can do useful things and do all that themselves. It's very rich data.

Is there anything that people need to do specifically with Amazon, other than listen to your metadata episode?

Thomas: So one advantage you have being in the UK is that you can still be blissfully unaware of how not great Rufus is. So I would rank Rufus above Llama and below all of the other LLMs.

I know that Amazon has a bunch of different AI models that they're developing. I was doing some experiments with one of them—I forget what it's called, started with an A—and it had the most delightful hallucination about me I have ever seen an AI do in my life.

So I got this tool where I had access to all the AIs, so I was asking them a lot of the same questions to see what answers they would give. It was like, “Thomas Umstattd is a professor of book marketing at Texas State University,” and then it started listing all of these book marketing books that I had written.

They were all vaguely associated with podcast episodes that I had done. It was all close enough to be believable, but none of it was true. It hallucinated an entirely different Thomas Umstattd. Like, oh, Amazon is behind on this AI thing.

So right now, Rufus, in my tests, is somewhat better than just a pure review search for answering product questions, but just barely. It's not where the other AIs are at yet.

Now, Amazon, just this week as we're recording this, signed a licensing deal with the New York Times. So they're now the only company that has an agreement with the Times for licensing data for AI models. It used to be the New York Times didn't matter for AI training because they had walled it off.

Joanna: They were suing OpenAI and all kinds of things. Microsoft.

Thomas: They were suing for big money, and Jeff Bezos is like, “Big money? I have big money.” So now they've got this really good source of data—or mediocre source of data, depending on your view of the New York Times—but they're the only ones that have access to it.

That's not going to fix Rufus's ability to find the needle in the haystack in a bunch of book reviews. So Amazon will get there. If it worked better, you would have it in the UK.

People don't want to give you the kind of mediocre stuff. They want royal quality for the UK. It's like, “This isn't good enough for the king, and the Brits aren't going to take this trash. We'll just keep it for the Colonials right now.”

Joanna: Oh, fair enough. Well, we should also say Amazon is a major investor in Anthropic, which does Claude. So it would be nice if they could use some of the Claude juice or something.

I mean, again, all these things are going to get better. So if people are like, “Oh, well, I don't like the Amazon automatics review overviews,” or whatever, it's like, look, all these things are going to get better and better.

Again, I want to stay really positive about this. As I said, I think we've been in the long tail for like 20 years already, right?

Now we're in the very, very, very long tail when we're thinking of generative search and the conversational search.

So I do a lot of book recommendations through ChatGPT, and often I'll have gone backwards and forwards several times before I'm happy with the level of granularity I'm at.

No longer do we have the sort of basic keywords, but we're having a whole conversation about what we like.

Or like you said, I sometimes just upload a screenshot or a picture of my whole bookshelf. I did this the other day, like with my bookshelf with probably hundreds of books on.

I said, “Here's my bookshelf. What else might I like?” It can read all the books on the bookshelf and all of that. So I think, again, all of this is going to get better.

Thomas: I think it's also going to reward good writing, especially that kind of relational stuff. So you have a really deep relationship with your o3 mini model, right? Like it knows you really well. You've spent months, maybe years, building its context window. So it has a good sense of your preferences.

This is one area where GPT really shines, as like a per user context window that persists across conversations. So when you ask your o3 for a recommendation, it has really good knowledge of books.

The GPT book recommender is already really good, and it has really good knowledge of you. So it's going to make very likely very good recommendations, and those recommendations are likely going to be the kind of books that are well written.

So as we talk about Schema.org and all these technical things, don't forget the fact that none of this will fix a bad book. None of this will fix a bad book.

If your book is not fun to read, if it's not engaging, if it doesn't pull readers in, if it doesn't leave them happy at the end, or they're leaving a good review, if it doesn't deliver on its promises, then it's not going to matter.

Like these things are great tiebreakers, and they can be really helpful if you're obscure, to move you from obscurity to notoriety. You can't make that move if your book doesn't thrill readers.

Your book has to thrill readers first.

I probably should have started with this. This is not like a “get out of learning the craft of writing” free card. It's just the opposite, actually, because the AI is much more discriminating on quality than the search engines were.

You type a search into Amazon, and it's going to surface whoever paid the most first. Then it will pick, based off of the very limited information you gave it, some books to rank.

None of that really had much to do with the writing. Some of it was connected. You know, it would look at review data and sales data and things like that, but a lot of it just became self-reinforcing. Popular books got more attention, which made them more popular.

My hope is that these new AI recommendation engines will have more nuance and make better recommendations.

It will also be better for people on the fringes of society.

So I'm not sure how it is in the UK, but like conservatives got pushed out of publishing. There's hardly any conservatives in the traditional publishing world. If you're writing a book for a conservative audience, it's hard to find your readers right now. Whereas I think AI is going to help bridge that gap.

Conservative authors and conservative readers are going to be able to find each other and they're going to get really excited. That's true with every kind of niche group of readers and authors that were really limited by the search engines that were really reductive.

Now all this nuance is going to be like, “Oh, you're interested in this kind of unique sub-genre that doesn't have a category in Amazon. Well, guess what? I've read all of the books in that, and here's the best one that I think you'll like really well,” and suddenly you're reading more books.

I think this is going to be good for the industry overall.

Joanna: Yes, I think so too. It should reduce, or hopefully reduce, that sort of paid ad effect.

Although, inevitably, these companies are going to have to monetize more than they are now. So it'll be very interesting how it changes.

So I'm also really interested in what else is coming. Now, people have been kind of saying agents and agentic AI for a while, but mostly these have been assistants.

So something like Deep Research reports, they're kind of an agent. You give them a task, “go research this,” and then it goes away and it will come back and bring you a report. So it's early days.

What is really interesting to me is zero-click. So zero-click with agents. So, for example, I just booked some trips in the US to Antelope Canyon, and I got my Deep Research report, but I still had to go buy the tour. I did go with its recommendation.

“Zero-click,” I say to my agent, my travel agent, let's say, “go and find me the best trip to Antelope Canyon. Here's my budget, and just arrange it. You know what I like in terms of hotels and the brands I like and all of that. Go do it.”

People are like, “Oh, no, that's years off,” but Visa, the Visa card, now has intelligent commerce. So they actually have a card you can use with an agent, so you essentially task it with buying for you.

I was thinking about this with books. It's like, “Here's my bookcase. This is what I like. Here's my budget per month. Just send me a book a week, or buy me some cool books and deliver them to my house.” I'm like, actually, that's quite fun.

With ‘zero click' + agents, the human never does the browsing, the human never clicks, and things turn up.

So what do you think about this?

Thomas: I think AI agents are going to start creating zones of the internet that are devoid of humans, and I don't think that's a bad thing. So the first place we saw this was actually about two decades ago, and it was the stock market.

So if you've ever seen an old 1980s video of the stock market, there's actual men in suits with pieces of paper, and they're all shouting at each other. Then somebody was like, “Oh, we'll connect it to computers.” So then somebody had to type on the computer to place an order.

Then there was one company that literally created a robot that just typed on the keys because the NASDAQ wouldn't allow automated orders, so something had to push the keys on the keyboard. Then they're like, “Okay, this is stupid. You can just place the orders.”

Then they're like, okay, well, let's create AIs that will evaluate a stock's price and make purchases. Now almost all of our stocks are managed by agentic AIs. So we moved away from mutual funds into something called an exchange traded fund, which is entirely run by an AI. It's algorithmic. This has been around for 10 or 15 years.

Now the New York Stock Exchange is a film set, and if you go there today, it's just a bunch of reporters reporting on what the computers are doing. Stock trading is no longer the getting the call from the pushy salesman about this really good scoop on such and such stock that you've got to buy and this real scammy thing.

We were taking a lot of our really smartest people and putting them in a room and having them shout at each other to buy and sell. I don't think that was the best use of those really smart people.

Now those smart people are doing other things that are more beneficial for the economy than shouting at each other to buy and sell shares of IBM or whatever. So that's what we're going to see with these other agents and these other sectors.

So you buying a ticket for a train, that's not a very emotionally rewarding experience for you to buy that ticket. For the person selling you the ticket, it's not a very emotionally rewarding experience for them either.

Nobody wanted to grow up, like when they were a kid thought, “Like all I want to do when I grow up is sell tickets for trains and answer questions from tourists who are all asking the same stupid questions over and over again, and I've answered this question 500 times this week.” Having an AI handle that is going to be better.

I think it's going to be slow. I think right now we're in the phase where early adopters like us are playing with it, but right now we're building our own agents.

I think a good model to look at is the spreadsheet. So back in the 80s and 90s, we'd have Microsoft Excel, and you could build your own spreadsheet. The 2000s have all been about taking features that you could do yourself in Excel and building a whole product around it.

Now there's a website that does that thing that Excel could do, but the website is for just one purpose, for one kind of user, and it does that same sort of thing. I think we're going to see that same thing with AI because most people don't want to create their own agents.

They don't want to create their own personal AI butler. They want to buy an AI butler off the shelf that does just one thing. So don't feel like you have to learn how to build your own agents in order to use them, you'll just have to wait. Who knows, maybe Joanna Penn will build some AI agents that she can rent out in the future.

Joanna: I mean, it is interesting to think where it's going to go. I tend to put this in the general category. For example, we've just talked about updating your Goodreads pages or updating your website. Hell, I don't want to do that, so I will just get my admin agent to do that.

Right now, there isn't one particular thing that can do that. I can do it with AI, but I'll still have to drive it. Whereas, I don't know, I mean, let's talk about ads as well. Mark Zuckerberg has repeatedly said that all you'll need to do on Meta—because, let's face it, Meta ads are just awful now. They're so complicated.

What he said is, you'll be able to say, “This is my book. This is the page I want to drive traffic to. Off you go. Meta will do all the creative, do everything, and here's a budget.” I don't have to do anything at all except say, “This is what I want to sell.” That will be great. I think authors will be all over that.

Most people want to do the writing, and they do not want to do the marketing, as you know.

So I think there are some great use cases. I don't know how long that's going to take. I mean, Zuckerberg has said end of next year I think, 2026.

Thomas: Well, it's already here for websites, actually. So if listening to Joanna and I talk about websites stressed you out, if you get Divi from Elegant Themes, it has AI built in, and you can just tell it to build you a homepage and what you want on it, and it will just do it.

I played around with this a couple of weeks ago because I was like, can AI really build a web page? And it did. I was kind of flabbergasted. I went in and tweaked and added, and I would copy what it did and add more things to make the page longer, but it's like it's already here.

It's already here a little bit for advertising too. The Amazon auto-targeting has gotten a lot better.

Joanna: I do use that.

Thomas: Yes, and it's self-reinforcing. So if you want to understand what machine learning is, there's some really good like cartoons on YouTube you can look up, but machine learning is the computer kind of getting better on its own, like improving itself.

These ad engines are using machine learning to get better every month. So if you tried auto-targeting on Facebook or auto-targeting on Amazon a few months ago, just realize that they're now better because the machine learning is training the algorithm.

There's not some developer at meta going, click, click, click, to tweak the algorithm. The algorithm is tweaking itself. Facebook's algorithm has been tweaking itself for over a decade. For people who are against AI and they complain about it on Facebook, I hate to tell you this, but you've been using AI on Facebook.

Joanna: Or buying on Amazon or using Google.

Thomas: It's like, Facebook particularly, the entire experience is AI, start to finish. It's like, “Oh, but I don't want you to have AI. I only want this powerful Californian to have AI, not the regular people.” Like, okay, now we're getting to a class conversation. This isn't really about AI anymore.

Joanna: Well, I mean, we could talk about this forever because you and I geek out on this. There are some people who are like, seriously, how is everything changing so fast, and how are you two so relaxed about the fact that everything is changing?

So how are you staying positive and curious? Obviously, there are bad things about AI, which, you know, we try and stay on the positive side of things.

Any tips for people who need encouragement to keep going in this time of change?

You know, they thought they knew the rules, and now it seems like the rules are changing yet again.

Thomas: You're not going to believe this answer, but I'm actually going to encourage you to study history, because we're not actually living in a time of rapid change compared to what our great great grandparents went through.

My great great grandfather was born in 1880, and the steam engine was new, that we were just starting to have the Industrial Revolution. When he was a child, the first car rolled into his town, a horseless carriage. Then suddenly there was electric light bulbs.

The telegraphs that had already existed when he was a kid, now there was lines that would go to people's houses, and they could actually hear a voice of somebody on the other side of town, and even the other side of the country.

It didn't stop there. Then a few years later in his life, something flew over the town that was heavier than air, and yet floating in the sky. Then before this man died, there was an American putting an American flag on the moon. That's not to mention radio, and the nuclear bomb, and like so much innovation.

Then we invented the semiconductor in the 1960s, 1970s, and then the innovation basically ended. After that, it was all of this really slow iteration where the transistors got smaller and smaller, the computers got faster and faster.

There wasn't this big, life changing technology, like what we were getting every two years in the late 1800s and early 1900s, until you have kind of critical mass of the transistors where they get cheap enough to enter people's homes. Then we have computers, and we have the internet.

We've had the internet for a long time. Like the internet's not new. It goes back to the 70s in the States. The World Wide Web was developed by a Brit in the 1980s. Then we didn't have much innovation, right? Web pages got a little bit more complicated, animation got a little bit better. Really slow evolutionary change.

Then the phones came around, and the phone was a big shift, but from a technological perspective, the phone wasn't that different than a computer. It's just smaller. So that same trend of smaller and lighter.

So in this way, AI is the first time for me to experience the kind of transformation that my great, great grandfather went through, but on a much smaller scale.

Like the tractor was unbelievably disruptive. One man in the town with the tractor could do the work of 10 men, which meant that those other nine men had to find something else to do to provide for their families. What it was doing things for the farmer, because the farmer was now making almost the same money.

So this is the flip side, the people who use AI—there's actually a report that just came out yesterday about industries where —

People are using AI, they're three times more productive and they're making 50% more money on that individual worker perspective.

So just like what happened with the tractor, there's nothing new under the sun. The guys who left the farm and started doing jobs for the farmer, those jobs are actually super rewarding. It's things like being a podcaster.

Like, I could not have my job of being a podcaster if there wasn't some blessed farmer somewhere in the sun and toil on his tractor making food for me, because I can't make food. I can hardly keep my grass alive. I realize in the UK, grass just grows on its own, but in Texas, it's a fight. It wants to wither and die in the sun.

So I'm really thankful that I'm not working in the fields like my ancestors did. So, yes, there's going to be some disruption, but the history of technology shows us that technology creates more jobs than it destroys.

95% of us were working in the fields back in the day, and that was awful, awful work for little pay. It was back breaking. It killed us, literally killed us. I don't think anybody wants to go back to that.

Technology isn't good or evil. People are good or evil.

So I'm not afraid of AI. I am very afraid of humans and what humans will do with AI, but I'm too much of a Texan to let those humans have AI and me not to have AI, too. The only thing that can stop a bad man with AI is a good man with AI.

Joanna: Or a good woman!

Thomas: Or a good woman.

Joanna: No, I mean, I think so too. We need to be on the side of the angels, and the more we're involved, that's the other thing. Obviously, people listening, Thomas and I are interested in the technical side, but you don't have to be super technical anymore to get involved.

The more creatives and other types of people who are getting to grips with these tools, the more they will represent the whole of humanity.

So that's why I try and encourage people. But also, you're right, it does make you more productive.

Also it's a lot of fun, so I have fun with my AIs like every day. So, yes, lots for people to think about. We're out of time, so—

Where can authors find you and everything you do online?

Thomas: So my website is AuthorMedia. That's where you can find all three of my podcasts. I have a suite of over 30 AI tools that are really easy to use. They're very specific things like an About Page builder, where you answer a few questions about yourself, and it will write a very interesting about page for you.

Or you upload your book cover, and it will analyze the cover and give you tips on how to make it better. Or create a chapter summary.

I even have a tool here called “Not a Literary Agent” that can review contracts and even write a rights reversal letter based off of the contract that you signed with that publishing company 10 years ago. And creating book blurbs. There's a bunch of different tools that are there.

My hope with these tools is that they're kind of training wheels for using AI, because they're really easy. You just answer a few questions, and you push a button, and that's it. So you don't actually have to be good with AI to try using these tools.

I've gotten just incredible feedback from folks who've tried these out. One, you just upload your book, and it creates a strategy for advertising on Amazon. Like a five page strategy based off the content of your book, including who to target and how to target them.

So my hope is these will help make you more productive. They're almost all focused on marketing. So they're not going to help you write the book, they're going to help you sell the book because that's my focus. If you want to learn how to learn how to write the book, listen to Joanna Penn.

Novel marketing is more focused on getting more sales for the book you already wrote.

Joanna: So those tools, is that on AuthorMedia.com?

Thomas: Yes.

Joanna: Okay, so that's fantastic. Definitely have a listen to Thomas's podcast as well. Well, thanks so much for joining me today, Thomas. That was great.

Thomas: Thank you for having me.

The post Book Discoverability In An Age Of AI. GEO For Authors With Thomas Umstattd Jr. first appeared on The Creative Penn.