Reflections on Co-writing with a Bot

It’s been over a year now since I published The Remote Worker’s Guide to Time Management, which I co-wrote with ChatGPT. It was an experiment worth doing, which resulted in me learning a couple of things I didn’t expect - so I thought I’d jot them down.

(I dictated this during my walk today, then transcribed it and edited it myself (as tempted as I was to give it to Claude for an edit! It’s not a faster process than typing it into a keyboard, but it meant I was able to capture the ideas for the post as they came during my walk.)

Origin Story

I had the idea for the book mid-2023, at the time when lots of people were using the bot to write their blog posts, and people's opinions around it were divided into those who were completely for it and those who were completely against it.

(And a lot of people with opinions along that spectrum, as well as people who had decided not to give it a try because who has time to try one more technology?)

I'd been hearing how the writing community was using it and most of the advice agreed on: don't let ChatGPT write a book for you.

Well, you know what I did?

I went ahead to see whether ChatGPT could write a book for me.

The very quick answer is: kind of, but with a lot of caveats, which I will of course address in this blog post.

Since then I've tried to write more books in a similar manner, but haven’t succeeded, and I'll come to why I think this is later.

For now, here’s a quick overview of the process.

The Bot’s Input

I wanted to write a book that was relevant to my knowledge of expertise, remote work. But I also wanted to write a book that I wouldn’t be able to write by myself.

So I picked the subject of time management and this seemed to work. I new enough about the topic to know when the content would be worth keeping.

The main reason why this topic worked is that there is a lot of good information around time management on the internet and the other sources that ChatGPT might have used as its training data. So the chances of getting some good generated writing from the bot were high.

I co-created an outline with the bot, and adjusted it as I went along.

Bit by bit, I’d say, okay, start writing this chapter. (This was particularly satisfying when I did it at 10pm on the sofa, on the iPad, while I was listening to the news.) I would look at what it had written and prompt me to give me more detail around this or give me an example of that, find some research to support a claim, or write a short story to illustrate a point.

I noticed that, while it had broad material (some of which I’d never come across before), and nuanced material around time management, this wasn’t the case for remote work. For example, all the advice around remote work was about working from home. So I would prompt, “give me some points on how this would be relevant to someone using a co-working space or a café to work in”. It needed a lot of guidance to go behind the very basic vision of what remote work is.

I also challenged any “research” and double checked references with good ol’ Google.

The Pilar Input

Once the first draft was done, I followed the same process as with my other books - but instead of taking months or years as I changed the focus, the structure and other things, I was done in about a week.

I went through the text and edited it. It had lots of filler sentences: It always started with an introductory sentence you could easily scrap and it would end with a sentence which you could easily scrap.

I deleted many unnecessary words. In short, I became an editor. And that was an unexpected benefit of going through the whole process: I have become a much better editor of my own work. (If you’re thinking, ‘well, this piece could do with some editing”, that’s because you didn’t see the original transcript! Wink.)

So having the bot write the first draft saved me a lot of time AND ended up helping with my own writing.

I'm going to be honest: I haven't sold many of these books. One, because I don't feel much ownership of it. Although it was great fun and there is actually a lot of me in it, after all, I was the editor. But there is nothing that drives me to push it out into the world. Nothing that says, “Everyone in the world must read this!”

Having said that, I think it's a very good summary of all the general advice that's out there on time management, and taking charge of your time as a remote worker. However, as my friend Mark said, he missed reading about my own experiences in it. (That was one of the best processes of working with the bot was to then pass the work on to someone else and see what they thought.)

More About the Person, Please

I’d deliberately not added any of my personal experiences or thoughts (well, some opinions, yes) into the book. As I said earlier, time management is not something I think about or reflect on often. And I thought it wouldn’t matter if there was not much of me in it because after all, it was a book generated by AI.

But of course, as a reader, it matters.

Even non fiction books where the topic is really interesting to me come alive when I hear the author's voice.

So that was another benefit of working on this project: I've realised that if I'm going to write any nonfiction, it's got to have a lot of me in it. And another thing I’ve found out is that doing that is a lot more fun.

A One-Book Experiment

As I mentioned, since writing the book on time management, I have tried to replicate the process, unsuccessfully.

I started co-drafting an outline for a book on motivation for the remote worker, because motivation is something I’ve looked at in-depth in the past, and there's a lot of material out there that can be used. But we weren't getting deep enough and I felt like there was not enough solid material to fill a whole book.

I also tried to write one around Connection in remote teams, but this didn't work either because most of the advice out there is not nuanced. It’s mainly about how if you want to feel less lonely when you're a remote worker, you should go on virtual coffees or meet up in person.

So I abandoned those ideas and realised I'd been very lucky with my first attempt at co-writing with a bot.

I think the main reason neither of these ideas worked was that the topics are not as well covered online - I mean, everyone seems to be obsessed about time management in one way or another…

Or maybe with time management for the remote worker, I found a subject with the right balance between knowing enough, being curious enough, and not having a strong point of view on it.

The other subjects I picked would need a lot of time and guidance to get to a text that was worth sharing. And if I was going to spend so much time guiding someone else to write, I might as well write the book myself.

Coda

(I love a good ol’ coda, even though I know this isn’t a book…)

I'm sure things have changed with the bots, and maybe if I tried now, a year later, to create those two books, things might be different. But seeing as I've discovered a style of writing that I really enjoy (part memoir/part self help), what would be the point of that?

As well as editing, I have also learned to prompt the bots to edit and create material, which I’m using to write my other books. A summary of a study here, a transition paragraph there, rephrasing a bunch of sentences which are a mess (technical term!)... It’s done wonders for my motivation, as it means I have an assistant and an editor on tap.

It's much easier when you arrive at a blank page to put down your thoughts and then give them to a bot and say, can you clean this up? Or can you elaborate on this? Or is there anything missing that just gets you going?

It's about finding ways that it can help you to either kick off the process or to keep you going.

By the way, one outcome of trying to write a book on Connection with the bot was that I remembered there was already a lot of material on the topic, in the form of podcast episodes (a seven-part season collaboration we did on the 21st Century Work Life podcast back in 2020). So I turned to Bree, who had created those episodes, and asked if she would be up for turning them into a book, with the help of genAI. We’ve used Claude to help us find material we could use from the episode transcripts, unify our writing voice, and to get some first drafts going.

Now we’ve turned to humans to have a read of our first draft - they’re the experts in the end in whether a book really works. (You can sign up to that mailing list here.)

And there we go.

Another long-form blog post.

Each section could have been a blog post.

But this blog post could also be a book.
Kind of.
I still haven’t discarded writing a book on how to write a book.
I’m still struggling with whether it’s right to encourage others to do so - the companies behind the bots have way too much power right now, if you ask me. (Or others.)
So look out for the post where I talk about the books I’ve been reading about the effect of the recent developments on AI.
(And also look out for the posts on Pilates - as I’ve completed my course, and passed!)

Here’s another random photo from one of my walks in Ravenscourt Park.