If you were paying attention to technology trends at the end of 2022, then you’ve probably heard about a ChatGPT – a chat-based interface to an artificial intelligence (or AI) large language model known as GPT3 developed by OpenAI, a San Francisco-based artificial intelligence research laboratory.
In its simplest form, you can type out a question, and ChatGPT will give you back a well-written answer crafted in surprisingly natural language.
But what’s more interesting…and more impressive…is asking ChatGPT to write something…an email, an article, even a Valentine’s day card for your wife.
In fact, here’s what it wrote when I asked it to do just that and gave it a few sentences about my wife, Ali:
"Here is a short Valentine's Day message that you can use for your wife:
To my amazing wife Alisia,
Thank you for being the most wonderful partner and mother to our son Matthew. I am so grateful to have you by my side through all of life's ups and downs. Your love and support mean the world to me. I am so lucky to have you as my partner in life. Happy Valentine's Day!
Cool. But a bit bland, no?
So, then I asked it:
"Make it funnier"
…and here’s what came back
"I'll do my best! Here is a revised Valentine's Day message with a bit more humor:
To my beautiful and patient wife Alisia,
Where do I even begin? You put up with my terrible jokes, my endless appetite, and my questionable dance moves. You've stuck by my side through thick and thin, and for that, I am forever grateful.
I don't know how you do it, but you manage to juggle being a mom and a superhuman all at the same time. You deserve a medal, or at least a free grande Chai Latte, extra hot with almond milk.
On this special day, I just want to let you know how much I appreciate and adore you. You make our little family complete, and I can't imagine going through life without you.
Here's to many more years of love, laughter, and questionable decisions.
Happy Valentine's Day, my love!
That’s better! Lock and load, right?
But apart from saving husbands the world over hours of painstaking wordsmithing come February 14th, I got to wondering…
Could aspiring authors use this new technology to write their manuscripts in record time? And even if they could…SHOULD they?
I’m not going to pretend to be an expert here. This whole space is developing faster than Elon Musk’s sudden interest in suspending free speech. So, despite the fact that this area is so nascent and that much is going to change, I thought it would be helpful to at least put a stake in the ground on this topic and share some initial thoughts on if and how AI might be used for writing books…
So, in this article, what I’ll do is:
- Provide a bit of a primer on AI, and on ChatGPT specifically
- Touch on some risks and limitations of it; and then finally
- Share an approach to how we might use AI effectively and ethically in writing a book manuscript
But first, a disclaimer
I don’t have all the answers here. I’m certainly not a technology expert. Nor a lawyer. But I do feel like I’m starting to put the pieces of the puzzle together as an entrepreneur, as a content creator, as an author. And I think this is going to be a very powerful tool that we need to understand. That we need to figure out how to wield both productively and ethically, and I have no doubt as to the importance of the topic.
So, let’s get started…
A short primer on ChatGPT
The easiest way to understand it is to try it out, which you can do by visiting http://chat.openai.com and signing up for a free account.
The interface itself is pretty straightforward as it’s simply a chat prompt. You enter text into that prompt and then hit ENTER and the chatbot responds, much like a human being would.
Now, there are many ways you can prompt ChatGTP:
- You can enter text, then let the AI continue it, like when I typed: “This technology is developing faster than…” – ANSWER “…a cheetah chasing it’s prey”
- You can ask it a question, like “What celebrities were born on January 8th?” (which it got objectively very wrong…I’ll come back to that)
- You can ask as it to create something, like “Write an article about writing sales copy” (which it did a far better job of)
- Or you can ask it to do something, like translate a text, create a summary, extract keywords, generate product names, and much (MUCH) more
Some of the impressive examples that users have shared include having ChatGPT:
- Write a movie synopsis for a horror film set on an oil rig
- Prepare a detailed weight loss plan, complete with food and workout recommendations
- Find, explain, and troubleshoot a bug in a block of programming code
- Draft legal agreements from scratch
It really is quite remarkable what this technology is capable of. And from what I’ve heard, the next generation using GPT4, is 50 to 100 times more powerful than the current GPT3. We’ll see.
But perhaps the best introduction might come from asking ChatGPT itself. So, here’s what I asked:
“I'm writing an article about you. Could you introduce yourself and share what you are and what you're capable of?”
Here’s the response:
"Sure! My name is Assistant and I am a large language model trained by OpenAI. I am a software program designed to generate human-like text, and I am capable of understanding and responding to a wide range of questions and statements. I can provide information and assistance on a variety of topics, including but not limited to general knowledge, science, history, and current events. I do not have personal experiences or feelings, as I am just a program, but I am here to help you to the best of my ability. Is there anything specific you would like to know or discuss?"
Okay. That’s all pretty cool stuff.
But what about the limitations?
Limitations of ChatGPT
To get an initial idea of the limitations of ChatGPT, I asked it again to give me its POV on the issue:
"As a language model, I am capable of generating human-like text, but I do not have access to the internet and am not able to browse or search for information. I can only provide information and responses based on the data that was used to train me, which was collected up until 2021. I do not have personal experiences or feelings, and I am not able to perform tasks or functions outside of generating text. I also do not have the ability to understand or respond to certain types of questions or statements that are beyond the scope of my training data or abilities. Is there anything else you would like to know?"
Enter your text here...
So, OpenAI trained ChatGPT with data and information collected up until 2021, which means it has no knowledge of anything that occurred after that time. As evidenced by the fact that it still speaks of recently deceased Queen Elizabeth II as the “longest reigning current monarch in the world”.
For the same reason, it does not have real-time information, so it can’t tell you the current weather, foreign exchange rates, sports scores, or the like.
Speaking of which, it cannot (or rather, will not) make predictions or subjective judgements, though it will try to offer pros and cons of alternative options.
Now, another important limitation is accuracy. The magic of ChatGPT is the natural language generation capability and the simplicity and facility of how you can interface with it. It’s actual output, while usually impressive looking (and sounding) is by no means guaranteed to be correct.
Going back to my example above when I asked it for celebrities born on January 8th... While it did give me Elvis Presley, whose birthday was January 8th, it also gave me Paul McCartney (June 18th), Margot Robbie (July 2nd), Goldie Hawn (November 21st), and Halsey (September 29th).
And this points to another limitation – sourcing. Or lack thereof. Birthdates are fairly easy to check, but that’s not often the case, and ChatGPT doesn’t provide any source provenance you can use to verify accuracy yourself (or to credit a source if and when needed).
The text that ChatGPT generates is original, in that it’s not copied or replicated from other sources, so plagiarism shouldn’t be an issue. And in fact, after generating a sample article, I ran it through Grammarly’s plagiarism checker, and it did, in fact, come through very clean, which was great to see.
It’s also worth pointing out that because ChatGPT’s output is driven by the training data used as its input, its output may show bias, so it’s important to be watchful of that.
So, then…given all it’s capable of and tempered with its current limitations…is ChatGPT something that could be useful to help you write your book manuscript?
I think so.
Using AI to write your nonfiction book manuscript
Even if you can use AI to help you write your manuscript, it’s important to be thoughtful about the how if you want to reap the benefits without it resulting in some sort of useless Frankenstein of a book that doesn’t deliver on the goals you have for it.
If you’re keen to try ChatGPT for the purposes of writing your book manuscript, here are some tips for how to go about it both strategically AND ethically:
1) Do the upfront work
This may be the most important takeaway on this topic. Take the time to plan out your book as you normally would. Clarify who your ideal reader is, articulate where they are today and where they want to be. Determine the central question your book needs to answer for them, and then plot out the chapter-by-chapter journey you’ll take them on to get them there. With your chapters planned out, take your outline to the next level, planning out your subtopics for each chapter and answering the important questions:
- What is this topic about?
- Why is this topic important?
- And how should my reader take action on this topic?
In short, spend the time to build out a detailed blueprint of your book. If you need some guidance on this, listen back to episodes 3 through 7 of the Entrepreneur to Author podcast or read chapters 2 through 4 of my book by the same name or check out Jacquie Pretty’s book Book Blueprint.
Spending this time will make sure that no matter who or what is doing the writing – whether that’s you, a ghostwriter, a research assistant or an AI chat bot – that it’s all working together toward a common purpose.
2) Use AI tactically
Could ChatGPT write your entire book for you? Technically, it could. Should it? Absolutely not. Not only would that be a disingenuous way to become an author, but your book would be far too generic to be of use in accomplishing your business goals. You need a book that reflects your unique approach to getting outstanding results for your clients, not a book-for-books-sake.
So, if you use AI to help with your manuscript, use it tactically, deliberately, and thoughtfully. Use it to make your job as author easier…not to replace you as the thought leader and expert. Otherwise, you risk actually BEING replaced as the thought leader and expert, and no one wants that.
3) Use micro-prompts
The more specific your requests of ChatGPT, the more useful the content it creates will be. That’s why it’s important to build out your book blueprint FIRST. Because only then will you have thought through the structure of your book to the level where you can write prompts that will result in content that is useful…and usable.
But balance that with…
4) Provide lots of context in your prompts
One of the wonderful things about ChatGPT is that you can provide all sorts of context to it in your prompt and it will use that context when writing content. So, rather than simply asking it to ‘write a short story about someone who improved their health’, you’d get much better results with a prompt like this:
"David is a married, father of 3 kids. He’s 47 years old and works as a Director at a large commercial bank in Cleveland, Ohio. He used to play football in college. But now he’s out of shape. I worked with David to get clear on his health goals. Then I build a nutrition plan, a weight training plan, and an exercise plan that would help him accomplish them. Then I trained him at the gym every week for 2 months followed by weekly Zoom calls to keep him accountable. As a result, he’s lost 45 pounds and is feeling and looking great. Please write a short case study about David."
Now, admittedly, that was much more work than our one-sentence prompt earlier, but it makes ALL the difference.
Here’s what we got back using this more detailed prompt:
"David is a 47-year-old husband and father of three who works as a Director at a commercial bank in Cleveland, Ohio. Despite his busy schedule and family commitments, David was motivated to improve his health and fitness after realizing that he had let himself go since playing football in college. He came to me for help in achieving his health goals and was provided with a customized nutrition plan, weight training plan, and exercise plan.
Over the course of two months, I trained David at the gym on a weekly basis and for the next 4 months, I provided ongoing support and accountability through weekly Zoom calls. The results of David's hard work are significant. Six months in, he’s lost 45 pounds, greatly improved his physical appearance, and he feels much better overall.
David's weight loss journey was not easy, but with the help of the nutrition plan, weight training plan, and exercise plan, he was able to make healthy changes to his diet and exercise habits. The weight training plan helped him build muscle and improve his overall strength, while the exercise plan provided him with a balanced mix of cardio and endurance training. Additionally, the weekly Zoom calls helped David stay on track and motivated, as he was able to discuss any challenges or successes with me.
By the end of the two months, David had transformed his health and fitness, going from out of shape to feeling and looking great. He is now able to keep up with his kids and no longer gets tired at work. He is happier, more confident and has more energy to do things he loves. He has also become more productive at work and is able to handle more stress.
David's success serves as a testament to the effectiveness of a well-designed nutrition plan, weight training plan, and exercise plan, coupled with ongoing support and accountability. With the right approach, anyone can achieve their health and fitness goals, regardless of their age or busy schedule."
Can you see how adding in specifics resulted in a case study that actually reflects the work that this expert author did with David? Otherwise, the AI would have returned a piece of fiction, which isn’t what you want in your nonfiction book.
5) Treat it as a starter
I said that #1 was probably the most important takeaway, but this one is just as important. Any content generated by ChatGPT or similar AI should be treated as a starting point. It can save you tonnes of time, but I strongly encourage you to go through and edit it yourself to:
- Make sure that it incorporates your unique ideas
- Give it your own unique voice so it’s written the way you would phrase it
- Make sure there aren’t any factual errors (remember the limitations we discussed earlier!)
And, of course, once you’ve got your manuscript completed, it should still go through the regular rounds of professional editing.
6) Consider citing
This last point I’ll make is, I’ll admit, a moving target. When it comes to the ethics of using AI in creative endeavours, transparency is key. While it thankfully seems unlikely that ChatGPT will produce plagiarized content in your book, I believe it’s important to let your readers know if and how AI has been used to write it. This becomes even more true the more extensively it’s been used.
So, there you have it. Like I said at the outset…it’s very early days for AI content generation. The technology will only get better and both the ways of using it and the ethics of doing so will continue to evolve.
But if you decide to dabble with ChatGPT…
The most important thing to remember is that you are the author. Don’t delegate the work of authorship to AI. That’s not going to result in a book that has the right level of strategic connection to your business to make it useful and effective for accomplishing your goals.
Instead, think of AI as a hard-working research assistant. Provide clear and targeted marching orders for specific areas, and then take responsibility for integrating it all together into a cohesive whole.