Amanda Smith is a freelance journalist and writer. She reports on culture, society, human interest and technology. Her stories hold a mirror to society, reflecting both its malaise and its beauty. Amanda's work has been published in National Geographic, The Guardian, Business Insider, Vice, News Corp, Singapore Airlines, Travel + Leisure, and Food & Wine. Amanda is an Australian living in the cultural center of gravity that is New York City.
I have a goal to read one book every month. On the weekends, I cozy up on the couch with my coffee, and I leave my phone in another room. The point of this ritual is to create space that's unbound by time and to-do lists -- but as my reading list grows, I find myself trying to get through more pages and more books.
I wanted to see if I could use artificial intelligence to summarize the main concepts, lessons and wisdom of a book I might not get to for months or even years. Keep the beautiful prose for the physical page, but use AI to summarize nonfiction business books, for example.
I picked Deep Work by Cal Newport to test run on ChatGPT. After using various AI tools, I thought ChatGPT, one of the best-known text prompt chat tools, would do the job best. I also have the $20 a month membership, so I wanted to make the most of it.
But the mission wasn't very successful -- partly because of plagiarism protections built in to the tools (and rightly so), and partly because it took a whole lot of prompt engineering and independent research to get anything worth using.
The very first thing I learned was that ChatGPT doesn't have access to full manuscripts -- to avoid plagiarism and respect intellectual property rights -- and just summarizes from existing information online.
If I've learned anything about AI, it's that the preprompt thinking is just as important as the first prompt. I didn't just want a huge summary of the book. I wanted to learn Newport's big ideas, arguments, strategies and frameworks regarding deep work, so I could apply it in my work.
So, I started the chat with some expectation-setting.
Prompt 1: "Do you have access to Cal Newport's book, Deep Work?"
Not ideal, but I had an idea. I found a 6-hour audiobook YouTube clip, so I asked if it could use that to summarize the book for me.
But no such luck. It told me to watch the video -- another plagiarism protection. Watch a video for six hours? That's hardly time saving.
There are over 32,000 reviews of the book on Amazon, so I thought maybe there's enough commentary on the book to come up with an in-depth summary. So I started over from this angle.
Next prompt: "I haven't read Cal Newport's book, Deep Work. Highlight the key ideas, concepts, strategies and frameworks, so I can apply to my business as if I've read it. I don't just want a summary of the book."
ChatGPT had a hard time interpreting "comprehensive," spitting out a ton of suggestions. It also started to give advice without asking me about my line of work, like "educate your team on the importance of deep work and provide training on time management and focus techniques" and "allocate dedicated blocks of time for deep work on your calendar." It also gave generic suggestions like "focus on tasks that provide the most value and minimize time spent on low-value activities."
All pretty vanilla advice, if you ask me. Time to apply the pressure, one question at a time.
I still wasn't getting any breakthrough insights, so I kept pushing.
I asked for an example of deep work in 2024, and we started to get somewhere. I liked one suggestion, about batching shallow work.
While I do this instinctively, it was a helpful reminder to batch tasks and watch context switching.
At this point, I was using ChatGPT to ask random questions, like if there's a limit on deep work hours.
It said 4 hours max of deep work per day.
I remember a concept I like called the manager-maker schedule, which details the two main schedule types. As a writer, I'm on the maker schedule, meaning blocks of uninterrupted time are critical.
I had to do a quick Google search of the key lessons in the book so that I knew what insights to prod ChatGPT for. Apparently "productive meditation" was an important takeaway, so I asked for more details about that.
Finally, a new concept. Deep work isn't just head down at desk time.
This strategy was the unlock -- looking on Google for nuggets in summaries, then going back to ChatGPT to expand. When directed, it was great.
I scanned a second summary and found another concept that sounded interesting: keeping a compelling scoreboard. ChatGPT helped unpack the concept.
If you want to use ChatGPT to learn about a book, you'll have to spend at least 30 minutes browsing reader summaries yourself and use the most interesting ideas to frame your prompts. This inevitably takes a chunk of your potential deep work time. You can't just ask the AI tool to do it for you.
Once you get on the right track that way, you might discover one concept that changes your productivity or perspective -- but ChatGPT still isn't as good as just reading the book yourself.
You also have to consider that since ChatGPT doesn't have access to the actual book, and only summaries and reviews that are available online, you may not even be getting accurate key points.
And whether using an AI summary defeats the purpose of a book about deep work is still up for debate.
