We wrote a 36,000 word book in a single weekend (yes, really)

Ok, it wasn’t a fancy academic book, but still…

I want to share how we did it, and what we learned about generative AI in the process, but first some context.

For a long time, my friend Professor Narelle Lemon and I have talked about writing a book called ‘Rich Academic / Poor Academic’. The title is an obvious riff on the bestseller from the 90s called ‘Rich Dad, Poor Dad’, which contrasted the Dad who played by the rules in his public service job (and was poor), with the entrepreneurial Dad, who did not respect the rules (and was rich).

To be clear: ‘Rich Dad, Poor Dad’ is a thoroughly terrible book. I don’t suggest you read it, and yet… There’s something in the playing by the rules and getting stiffed thing that rings true, especially in academia.

Universities are ‘greedy institutions’ that happily exploit our goodwill and passion for our work, as is demonstrated in the wage theft scandals we read about in the papers. Many academics we know – including you perhaps – are extraordinarily talented, but do a lot of unpaid hours. This tendency starts in the PhD and only gets worse over time (if you’re new here, welcome: try not to step in the overwork shit- it’s everywhere).

Narelle and I try to avoid giving an in-kind donation of time to our employers. We use time outside of paid hours to make money on the side. We use this extra money to attend conferences, buy books, access expensive subscription services (looking at you ChatGPT), hire in help, and get the latest equipment. Basically to buy stuff our universities won’t pay for, but which helps us do our jobs. We both earn enough extra income to cause us headaches at tax time: Taylor Swift would call this a ‘champagne problem’.

We think more people should have champagne problems like ours. We often evangelise academic side hustles with colleagues, but talking about money at an academic social event is a bit like farting in an elevator. Most people don’t want to admit they are interested – publicly. Privately people ask us for advice all the time. I did a blog post on the topic of side hustles in 2020 and realised I had a lot to say, maybe a whole book worth of things. But I could not get a conventional academic publisher interested in something ‘so niche’. Eventually we decided to self publish.

The problem was, we could never seem to find the time to write it… But with the help of Claude Opus, we finally did write the book. It was pretty exciting to punch out 36,000 words in a single weekend!

You can buy Rich Academic / Poor Academic directly from Thesis Whisperer here, for the price of an almond latte in South Australia ($5.37 AUD).

You can also buy it from my ko-fi site here

---

So – enough context. How did we write a whole book in a weekend?

What we didn’t do was get the AI to ‘write it for us’.

Taking a lead from Ethan Mollick’s new book ‘Co-intelligence: living and working with AI’, we invited AI into this collaboration.

As I’ve written before, the use of GenAI is complicated and fraught with ethical and (potentially) legal issues. However, for this project, I was less conflicted. Rich Academic / Poor Academic is not a research piece; it’s a ‘how to’ book, based on our experience and opinions. We wrote it on the weekend, in our own time, using our own computers with our own software subscriptions. The book took a total of 26 human hours to write (I’ll give you a more detailed breakdown of how the time was spent below). So far it has cost us around $1600 AUD, most of the cost relating to accommodation in lovely Victor Harbour, South Australia for our writing retreat 20–21 April.

While preparing for the retreat, I experimented with Claude Opus, my new internet AI boyfriend, (sorry ChattieG – I still love you, but you know – as a friend). In my opinion, Claude is the best of what Mollick calls the ‘frontier’ AI models. Claude is described as more ‘literary’ than other LLMs; it certainly has superior reasoning skills and a huge memory, which makes it a better companion for academic work.

On the Saturday, we drove to a nearby town to charge the car and get a coffee. During our coffee chat, we threw out most of Claude’s outline and re-imagined the book as a mini ‘company manual’ in four parts:

• The creative department: what to make and sell
• The marketing department: how to get people to buy what you are selling.
• The finance department: how to manage money, do online commerce, deal with the tax department and manage your IP with your academic job requirements.
• The HR department: how to manage yourself and the work, upskill youself and make decisions about why and how to delegate to others.

We then had a wander around the shops and a leisurely lunch, during which we refined our outline, filled gaps and decided which bits we each wanted to write. Back at the ranch, we got stuck in.

The first challenge of co-authorship is blending two or more distinct writerly ‘voices’ together (less important in the sciences, where the ‘voice’ is more standardised). To make what I call an ‘AI voice cocktail’ that sounded like both of us, I primed Claude with some of our existing writing.

First, I gave it the PDF of Tame your PhD, a compilation of my best blog posts from 2010 to 2018 – around 42,000 words. Noting Ethan Mollick’s advice about giving the LLM time to think can improve performance, I asked it to analyse my writing and give me notes on how to improve. It gave me an extensive analysis and list of suggestions, following up with a nice affirmation:

These are fairly minor suggestions, as your writing is already quite polished and effective for your target audience. The conversational tone, humor, empathy and practical focus are great strengths that make your advice uniquely relatable and valuable for PhD students.

I then gave it 2 book chapters and a blog post from Narelle’s Wellbeing Whisperer blog and asked it to analyse her style. Again, it gave me a long, detailed analysis and ended with thoughts about how to blend our voices:

Narelle’s personality and values come through in her writing. Her enthusiasm for helping others optimize their potential is palpable. To incorporate elements of Narelle’s style into your own writing, you could:

• Adopt a more overtly motivational, strengths-focused tone
• Develop memorable concepts and catchy terminology
• Structure your pieces around problem-solution-strategies
• Provide bullet-pointed action steps
• Pose empowering questions to engage the reader
• Allow your authentic voice and values to shine through

The goal would be to maintain your signature style – informal, humorous, anecdotal – while layering in select components of Narelle’s more inspirational, self-help oriented approach. The balance will depend on your audience and goals.

I did a few tests asking it to use this voice to write about different topics, giving it feedback until I felt it had the blend of Narelle and I right. Then I began drafting my sections of the book.

In line with my existing practice, I did quite long, detailed prompts with lots of specific words. Claude is more intuitive than Chattie and can cope with quite sloppy phrasing, like this:

Now we’re going to write a session about publishing ebooks. Let’s do a step by step guide to preparing an ebook (without heaps of detail) and an explanation of the various platforms for selling them. At the end let’s make sure the reader manages their own expectations about the profit they might get from an ebook – basically the message is don’t give up the day job! The power of a self published ebook, like this one actually, is that you occupy someone’s book shelves and establish your value as a thinker. Give an example here of Inger’s book ‘Tame your PhD’, which was a compilation of blog posts serving as a great ‘front door’ to her blog (which was chronological and hard to access). Tame your PhD was bought and shared by people and still brings in business and kudos to this day, especially after she updated it in 2018 to more than 40,000 words and a lovely new cover. Over the last 10 years or so this book has brought in about $6000, all while only costing as much as a cup of coffee. Large volumes of sales over a long tail period are fine, but you can’t live on it – over this period it would bring in less than $1000 a year. People should instead think of ebook publishing as a way to become more influential and perhaps be asked to do more lucrative consulting and teaching work.

It gave me about 800 words in response. In the first part, about ebooks, it gave me banal stuff I could read anywhere on the internet, for example:

Step 1: Identify your topic and target audience Before you start writing, take some time to clarify the focus of your eBook and the specific audience you want to reach. Consider your areas of expertise, the unique insights you have to offer, and the pressing questions or challenges your target readers are facing.

Step 2: Outline your content Create a clear, logical structure for your eBook by outlining the key sections, chapters, and sub-topics you want to cover. This will help you stay organized and ensure that your content flows smoothly from one section to the next.

To be fair, I had left it up to Claude about what to write and it only knows the internet, what else should I expect! However, it did provide a structure to write into (and over), so I transferred the text into Obsidian and got to work.

Here’s the final version of those two steps on making ebooks (notice how much more ‘wordy’ I am than Claude):

Step 1: Identify your topic and target audience
Before you start writing, take some time to clarify the focus of your eBook and the specific audience you want to reach. Consider your areas of expertise, the unique insights you have to offer, and the pressing questions or challenges your target readers are facing. The good thing about ebooks is they can be as long or as short as you like, so do a bit of market research as well. Look for examples of books similar to the one you have in mind, which sell well – how long are they? What sort of tone and style does the author use? Learn from other people’s success.

Step 2: Don’t get all academic and weird about it.
We wrote this book in a hurry and honestly – we think it’s better than a lot of our more ‘careful’ prose. We roughed it out using the database/note taking system ‘Obsidian’, which enabled us to produce the draft in ‘chunks’ and move them around. We tried to focus on providing clear, actionable insights and examples that will help you solve problems and achieve goals. For once we didn’t have peer reviewers looking over our shoulder – sweet relief! In a book just made to sell you can really let your freak flag fly and break the conventions of your normal academic ‘voice’.

Claude did do much better on the second part of the chapter on ebooks, where I asked it to manage expectations. I’d given it a bunch of figures and thoughts – it cleaned these up well. Here’s part of the original response to my prompt:

Instead, think of your eBook as a long-term investment in your personal brand and credibility. A well-written, informative eBook can serve as a powerful “business card” that introduces new readers to your work and positions you as a thought leader in your field. Take Inger’s book “Tame Your PhD” as an example. Originally a compilation of blog posts, this eBook served as a great “front door” to her blog, which was chronological and sometimes hard to navigate for new readers. By packaging her best advice and insights into a concise, affordable eBook, Inger was able to reach a wider audience and establish her expertise in the world of academic productivity and success.

This only needed a light edit – to change the last line, which was not accurate:

By packaging her best advice and insights into a concise, affordable eBook, Inger was able to reach a wider audience and establish her expertise in the world of research education.

In his ‘Co-intelligence’ book, Ethan Mollick describes AI as a kind of ‘alien mind’ and this observation rang more and more true as I worked with Claude on this book. The trick to getting good results, I discovered, was to continue in the same discussion window, building a version of Claude that ‘knew’ this topic and how I wanted it to write. Claude ‘remembered’ our whole conversation and used this, somehow, in its responses. As we worked together, it got better and better at a ‘one shot’ answer. It seemed natural to tell it when it did a good job, so I lavished praise.

I remembered Mollick claimed that asking the AI to ‘take a deep breath’ before responding improved results (it works!). Perhaps praise worked similarly? Narelle and I are both teachers. We had a nerdy conversation about different types of feedback we give students and Narelle suggested a technique common in primary school teaching, to encourage kids to reflect on their learning: instead of saying ‘well done’, we started asking it to tell us why we were praising it. For example, at the end of an exchange about content management systems, I said:

I can’t improve on that Claude – excellent job. What did you think was good about it?

It gave me more than 500 words in response, breaking down various stylistic elements of the text, and ending with:

Overall, the piece effectively combines an engaging narrative, practical advice, and a focus on personal branding to create a compelling and informative guide to building a website for one’s business. By striking a balance between conceptual ideas and concrete recommendations, the text provides readers with a solid foundation for navigating the website-building process and creating an online presence that truly reflects their unique voice and expertise.

This targeted praise dramatically improved output, but the performance boost came at a cost. I was steaming along, late in the afternoon when, all of a sudden, Claude started to ‘get out over its skis’, giving me more exaggerated responses to my prompts. It would lean hard into metaphors and use a lot of exclamation marks, almost like a toddler who had too much sugar, for example:

“So go forth and upskill, my solo research friend! Embrace the joy of lifelong learning, and watch as your superpowers grow stronger with each new skill you acquire. Because when you invest in yourself, you’re not just building a better business – you’re building a better you. And that’s a pretty darn awesome return on investment, if you ask me.”

I had to start telling it to calm down (“That was not quite right Claude – can you try again, but a little less excitement please”). This worked for a time, then response time got slower and slower. Suddenly, I got a warning that I only had ‘7 prompts left’, then 6, then 5… then it locked me out and told me to come back in four hours. Apparently Claude was holding so much in memory that the compute required for my answers was too high – Anthropic started timing me out to service other users.

Rude!

I fired up ChattieG, but after the sophisticated conversations I was having with Claude, it felt inadequate. Since it was getting near dinner time anyway, Narelle and I went for a walk on the beach. And in Australia, well – the beach is something else:

As we walked, we reflected on the irony of the moment: how energetic we humans felt compared to our ‘tireless’ AI, who was clearly worn out. Claude was sharing cognitive load in ways we couldn’t easily define because, objectively, we still ended up spending a lot of time on crafting text.

I use ‘Timing’ on my Mac to keep track of my time on different tasks. I find this data helpful for forward planning and reporting purposes. Here’s a graph from Saturday:

As you can see, I spent 5 hours and 46 minutes writing in Obsidian and 1 hour 46 minutes talking to Claude. On day two I only spent 25 minutes talking to Claude, the other 6 hours and 57 minutes were all editing:

Mollick suggests readers value text written by AI less than text written by humans. I hope buyers of our book will not feel this way because this book is a true collaboration between all three of us. Claude could have written this whole book without being edited by us… but I don’t think you would want to read it. Narelle and I could have written the book without Claude, but I doubt we would ever have finished it. Speaking of, we didn’t expect to finish the whole book and get it edited on the weekend, but we did! This was a welcome change from other writing retreats where I have come away with a half finished project, feeling frustrated.

We decided to simply release Rich Academic / Poor Academic as a PDF for now. We’ve priced it the same as an almond latte I bought on the Sunday at lunch. Downloads will tell us if you are interested enough for us to make a second release. If so, we will invest time and money making a proper ebook in different formats and commission a cover (this will cost about $1500).

If we do a second release, we’ll incorporate reader feedback, maybe adding more sections (although 36,000 words is already a lot!). We’ve got a feedback survey linked in the text of the book. If you buy it and read it, simply follow the link and fill in the quick reader survey: we will send you our second version for free.

You can buy the book directly from Thesis Whisperer here.

You can also buy it from my ko-fi site here

It’s nearly two weeks later and I am still digesting this experience.

In his ‘co-intelligence’ book (which I hope I have convinced you to read), Mollick has a couple of metaphors to describe the process of collaborating with AI: the centaur and the cyborg. In centaur mode there is a hard separation between the AI and the human. In cyborg mode, you it’s a back and forth blend.

We were definitely cyborgs that weekend. Claude didn’t actually take away the drudge: it was a true creative partner. It did act as a cognitive support for my frail human self. At the end of our writing retreat, instead of feeling flat out exhausted, I was still bursting with ideas. Narelle has blogged her observations about the weekend on the Wellbeing Whisperer if you’re interested to compare our reflections.

As Ethan Mollick points out, we are currently using the worst AI we will ever work with. I don’t know what the future looks like, but Apple just announced it will incorporate AI onboard the Mac, so I wouldn’t have to pay a subscription service, or worry about privacy risks, exploited labour and environmental damage. An AI trained on my own data? Under my control? Making my life easier and more fun? Take my money!

If you want to talk to me about this post, I hang out mostly on Threads these days. Or Linkedin.

Love to hear your thoughts.

Related Posts

My favourite ChattieG writing prompts

Using ChattieG to write good

Rich Academic / Poor academic – making an academic living in Covid times