AI Book Companions™ vs Chatbots: What's the Difference?

Why uploading your book to AI is only the beginning

I think one of the reasons authors are becoming interested in AI Book Companions™ is that the technology makes the whole idea look relatively simple. You can upload your manuscript into ChatGPT, give it some instructions and, within a few minutes, start having conversations about the contents of your book.

And I can understand why that would be appealing. If you've spent months or years writing a book, the thought of readers being able to ask questions about your ideas, explore something they didn't quite understand or return to a particular chapter is quite exciting.

The first time you try it, the results can be impressive. You ask the AI a few questions, it gives you sensible answers, and it appears to understand what you've written. At that point, it's quite easy to think you've created something that could be useful to your readers.

But there's a difference between giving AI access to the information in your book and deciding how that information should be used.

That's where I think authors need to look a little more closely at what they're actually creating.

Knowing what's in your book is only part of the job

Let's imagine you've written a book based on a methodology you've developed through years of working with clients.

You upload the manuscript and ask the AI to explain one of your concepts. It gives you a perfectly reasonable explanation, perhaps even using an example from the book.

But what happens when somebody asks a question that isn't answered directly in the manuscript?

Perhaps they've read a chapter and want to understand how it applies to something happening in their business. Or they're struggling with an exercise and don't know how to move forward. They may even start describing a situation that has very little obvious connection to the chapter they were reading.

The AI now has to decide how to respond.

It could give them an explanation, suggest an exercise, ask them to think about something or offer advice based on its wider knowledge. Some of those responses might be useful, but they might not reflect how you would approach the situation.

And if you've developed a particular methodology, that distinction becomes important.

When I work with experts, I often find that there's a considerable amount of knowledge behind the way they do things that isn't immediately obvious from the finished material.

They know which questions to ask, which stages somebody needs to work through and why one approach might be more appropriate than another. Sometimes they've been working this way for so long that they don't even think about those decisions consciously.

A book may explain the methodology, but that doesn't automatically tell an AI how to follow it during a conversation.

So part of my work involves understanding what's behind the author's approach and turning that into something the Companion can use consistently.

The decisions that sit behind an AI Book Companion™

When I began creating AI Book Companions™, I was already bringing several years of experience developing AI agents for experts and business owners.

What interested me about books was the amount of knowledge that was already there. The author had done the work of bringing their thinking together, and in many cases there was a framework or methodology that could be developed into a conversational experience.

But I quickly realised that each book needed a different approach.

I've now created Companions for books covering subjects including self-love, relationships, mindset, wellbeing, ADHD, marketing and personal development. Although I use a consistent development methodology, the Companions themselves don't all work in the same way.

That wouldn't make sense, because the books are doing different jobs.

Some authors want readers to understand their ideas more deeply. Others have exercises or processes that readers can work through. In some cases, the Companion may sit alongside an existing course or programme.

Before I start building, I need to understand what the author wants the reader to be able to do and how the Companion should help them do it.

That involves looking beyond the manuscript itself. We need to establish how the Companion should approach questions, how closely it should follow a particular process, what it should do when it doesn't have enough information and where its boundaries need to sit.

Those decisions form the framework around the author's knowledge.

Without that framework, the AI may still produce some very good answers. What you have less control over is whether those answers are consistently appropriate for the experience you're trying to create.

The problems authors may not realise they need to solve

When I look at a book and start thinking about how its knowledge could become an AI Book Companion™, there are several things I need to work through. Some are fairly obvious, but others only become apparent when you start considering how real readers might use it.

These are the areas that take considerably more thought than simply uploading a manuscript. And they're also the areas where my experience developing AI agents becomes particularly useful, because I've learnt that what looks like a small decision during the build can make quite a difference to the experience somebody has when they start using it.

What real readers taught me about this

One of the most useful experiences I've had during the development of AI Book Companions™ came from the reader testing we carried out with Sarah Grace's From Shadows to Self-Love.

Sarah's book explores how childhood experiences can influence the way we think, feel and behave as adults. Her Companion was developed around the ideas and approach in the book, with particular attention to how readers might explore those ideas in relation to their own lives.

What became interesting was how differently people used it once they had access.

Some readers asked questions about feelings or experiences they were struggling to understand. One woman used the Companion after an upsetting encounter with a stranger while walking her dog, and the conversation led her to reflect on experiences of being bullied at school.

Another reader used it to explore why she disliked Christmas.

And then there was the woman who was having difficulty packing clothes for her holiday. She started talking to the Companion about why she felt uncomfortable in certain outfits and eventually worked through what she wanted to take. She later told Sarah she'd packed her suitcase within ten minutes.

I certainly hadn't anticipated somebody using a Self-Love Companion to help pack a holiday suitcase.

But when you look at what was happening during the conversation, it made sense. The reader was exploring her feelings about the clothes she was choosing, which connected with the ideas Sarah works with in her book.

That experience helped me understand why it's so important to build the Companion around the author's methodology rather than trying to anticipate every possible question.

Readers will bring their own lives into the conversation. We can't predict all those situations, but we can make decisions about how the Companion should approach them.

What about the things you don't want your Companion to do?

This is an area I think authors can easily overlook, particularly when they're experimenting with AI for the first time.

Most people naturally concentrate on what they want the AI to be able to do. They want it to answer questions, explain their ideas and help readers work through the material.

But when you make something available to other people, you also need to think about the situations it shouldn't attempt to handle.

With Sarah's Companion, for example, readers sometimes brought quite personal experiences into their conversations.

Sarah is a therapist, and it was important that the Companion supported the ideas in her book without presenting itself as a replacement for therapy or for Sarah herself.

That meant establishing boundaries around the type of support it could offer and when it should encourage somebody to seek appropriate human help.

This is what I mean when I talk about governance.

It includes the instructions and boundaries that help determine how the Companion should behave, particularly when a reader asks something outside its intended purpose.

There are also questions about how the Companion represents the author's work, what information it should rely on and how it should handle situations where it doesn't know the answer.

And because AI responses aren't entirely predictable, those decisions need to be tested rather than simply assumed to work.

Testing something yourself is very different from watching readers use it

I think this is probably one of the easiest things to underestimate when creating an AI Companion.

When you've built something yourself, you know what it's supposed to do. You understand the instructions you've given it, and you tend to ask questions that make sense within the framework you've created.

Readers don't have that background.

They arrive with their own questions, experiences and ways of communicating.

During our testing, one reader described something she was experiencing as a "fuzzy rectangle floaty thing". Another asked for information to be broken into smaller chunks because that was easier for her to work with.

Those are the kinds of interactions that help you see whether the Companion can actually support different readers in the way you intended.

They also reveal things that need changing.

Sometimes the issue is how the Companion explains something. Sometimes it's the way it follows the methodology. And sometimes testing reveals a situation where the boundaries need to be clearer.

I've found that this process can tell you considerably more than asking the Companion a handful of questions and deciding that the answers sound good.

It's also why I don't see an AI Book Companion™ as something you simply upload, configure and hand over without examining how it behaves.

So, what's the difference between a chatbot and an AI Book Companion™?

The underlying AI technology may be very similar, and a chatbot can certainly be developed with sophisticated instructions, knowledge and safeguards.

For me, the distinction is in what has been deliberately designed around the book.

When I create an AI Book Companion™, I start with the author's work and the experience they want their readers to have. From there, I look at the methodology, the way the author communicates, the decisions the Companion needs to make and the boundaries within which it should operate.

I then bring those elements together and test how they work in conversation.

That process is important because a book often represents years of somebody's experience and thinking. The author has made decisions about what to include, how to explain it and how they want readers to understand their work.

If we're going to give readers a new way to interact with that knowledge, I think we need to be just as thoughtful about the experience we're creating.

And this is where my experience developing AI agents becomes particularly relevant.

I've created more than 100 AI agents over the past four years, working with experts to structure their knowledge and approaches into conversational systems. What that work has taught me is how much attention needs to go into the thinking and methodology behind an AI experience, as well as the technology that makes it possible.

An author can certainly experiment with uploading their own book to AI. I think that's a useful way to begin exploring what's possible.

But when you're ready to create something for your readers, there's considerably more to think about.

What does this mean for your book?

If you're considering creating an AI Book Companion™ for your own book, I'd suggest starting by thinking about what your readers actually need from it.

Perhaps you want them to be able to ask questions when they don't understand something. Maybe your book contains a process they could work through, or exercises they might benefit from revisiting. You may have a course or programme that the Companion could support.

Once you've established that, the next question is how your methodology should guide those conversations.

And then there's the practical work of deciding how the Companion should behave, what boundaries it needs and how you'll test whether it's doing what you intended.

These are the areas I work through with authors when developing their AI Book Companions™.

I think the opportunity is much bigger than simply allowing readers to ask questions about a manuscript. But to make the most of that opportunity, we need to understand the knowledge, experience and methodology that make each book different.

If you're exploring what an AI Book Companion™ could look like for your own book, you can find out more about my work at https://www.wendykier.com/.

About Wendy Kier

Wendy Kier is the founder of EmpowerAi™, creator of the AI Book Companion™ and author of The AI Book Companion™ Field Guide.

Over the past four years, Wendy has developed more than 100 AI agents, working with founders, authors and other experts to turn their knowledge, methodologies and intellectual property into practical AI experiences.

Her work now focuses on helping nonfiction authors bring their books to life through an AI Book Companion™ that allows readers to explore ideas, ask questions and apply what they've learnt.