How this Times journalist used AI to write a book in months
A conversation with Katie Prescott on NotebookLM, research at scale, and the future of journalism
Elea.notes is a newsletter about tech, writing and creativity. If you haven’t subscribed, join me.
When ChatGPT was released in 2022, I was an editor at the FT-backed Sifted, and I remember the exhilarating feeling of using the very tools we were reporting about. Three years later, journalists are still just scratching the surface of how these tools can enhance reporting.
This week, I had the pleasure of getting the perspective of The Times’s technology business editor, Katie Prescott. Her book on controversial British tech billionaire Mike Lynch, who died in a yacht accident last year after being acquitted of fraud charges, is out next week. We discussed how she used tech in the time-crunched writing process and what developments in tech mean for the future of her profession.
Key takeaways: Katie’s tool of choice, Google NotebookLM, can be a game-changer for projects — like writing a book! — where you’re dealing with lots of information and accuracy is key. But it still takes an expert to correctly analyze and interpret findings: this is the journalist edge.
EW: Tell me about why AI is so relevant for your upcoming book about entrepreneur Mike Lynch?
KP: In the last phase of his life, after he sold his business Autonomy to HP and was accused of fraud, Lynch started a fund called Invoke and invested in AI businesses. He was very early to legal AI — I think he saw how much money he was spending on lawyers and realized that a lot of that work could be automated — AI in healthcare, with Sophia Genetics, and also in finance. So it felt appropriate that I was using AI to sift through documents for the book.
EW: How exactly did you use AI in the process?
The challenge I had was this extraordinary volume of information. [Lynch] was involved in three major trials…On top of that, the Financial Reporting Council, the British accountancy watchdog, also looked over Autonomy’s accounts and produced an enormous report.
I don’t think you could have shoved it into an AI engine and understood what was going on. I needed to read through [the documents], particularly when it came to Mike Lynch, because he died before I wrote the book, and I had planned to interview him.
What AI allowed me to do — and I use Google NotebookLM — was to essentially create my own mini LLMs for each trial and interrogate them. That was incredibly useful for fact-checking, but also to find little nuggets of information or testimony that I wouldn’t have found if I printed them all out and read through.
It wasn’t like using a chatbot where you’re afraid of hallucinations. I could check that something was accurate and where it was coming from.
I also uploaded all of the transcripts of the earnings calls at Autonomy from when it listed [in London] in 2000 up until it sold to HP in 2011. That was helpful in terms of tracking the progression of the company and seeing, for example, points during the financial crisis when it was facing challenges — what were [Lynch’s] arguments about that?
EW: So you were able to connect disparate threads you might not otherwise have been able to connect and fact-check.
KP: Yes, and also find testimony from people that I might not have read because I thought they were more of a bit part.
When I was fact-checking the book, I could put numbers in. For example, one of the numbers was that Autonomy’s quarterly revenue rose by £200 million. But it didn’t, it was £200 million.
You couldn’t have done it cold. You needed to understand the story across the documents and who the people were, but once you got there, it was a game-changer.
“You couldn’t have done it cold. You needed to understand the story across the documents and who the people were, but once you got there, it was a game-changer.”
EW: If your book were to be used as a textbook in business schools, what do you think the takeaways would be?
KP: The HP-Autonomy case is already a Harvard Business School case — on how not to do due diligence. HP did 18 days of due diligence for an $11 billion purchase. There was noise in the market that they should have listened to, but they were in a rush to buy the company.
Autonomy wasn’t a failing business by any stretch of the imagination. It’s just that Mike Lynch was trying to flatter the numbers and make them look better than they were. His team used various techniques, like selling hardware as well as software and not telling the market.
I think there’s also something about founder CEOs. Lynch ran the company from an inner circle of people who he was very close to, and who could not say no to him, or did not say no to him. I think that when you get to a certain scale as a company, there is an argument for having a professional chief executive at the helm.
They always describe [Autonomy] as being run like a startup. It wore that as a badge of honor because it means you’re quick and reactive and creative, but that can also have repercussions.
EW: Lynch was definitely ahead of his time in recognizing the applications of AI in business. When you speak to business leaders now, what don’t they get about this technology?
KP: I think there is a lot of fear among chief executives and in boardrooms that they are missing something, but I think the wisest people are cooling their jets and watching how [AI] develops.
I spoke to Sebastian Siemiatkowski, the boss of Klarna. They’re slimming down their workforce by basically half. Not by firing people, but by natural attrition, and he’s expecting [the business] to grow just as much. His inbox is full of execs asking, “How are you doing that?” There’s desperation and a lot of FOMO, but it’s probably unjustified. Apart from the obvious examples of customer service, for example, and Copilot…I’m not seeing any absolutely, oh-my-god transformational thing going on.
EW: What isn’t being talked about with respect to AI?
KP: I don’t think we are thinking enough about the copyright issue. The Getty Images and Stability AI court case hearing was published today. [Note: Getty largely lost the case.] That comes with a cost. I think we all need to be mindful about giving too much power to these incredibly wealthy companies that have market valuations of more than the UK’s GDP.
Recognizing that and thinking about the impact on society — spinning back to that copyright point — making sure they pay their fair share while they’re reaping hay in this AI revolution.
I think we’re almost in danger of deifying these people.
EW: So are we in an era where tech defines the age, versus politics?
KP: Tech is defining politics. Look at how our governments treat tech leaders. Wasn’t it extraordinary that Trump brought the US tech guys on his state visit?
EW: What about journalism going forward? What does this new technology mean for the profession?
KP: I’m so excited about it, I can’t tell you.
The way I used it for my book, I used it similarly in a court case recently. It was another fraud case with [luxury property developer] Nick Candy accusing someone called Robert Bonnier of basically conning him into investing in a social media company.
I put all of the testimony into NotebookLM. I literally asked it: “Find me the most interesting bits of testimony.” It found that Bonnier’s wife used to collect champagne corks, and [a story about how] she hired a private jet to fly her handbag to Dubai from Jakarta.
I think in terms of helping journalists process large volumes of data — journalists who often aren’t very good at doing things like that and who are very time poor — it is invaluable. You still need the skills to understand the information, process it, assess it, and check it. It’s not like that’s going to be taken away from you.
EW: Right, it’s still fundamentally about judgment and expertise.
KP: I was at Bletchley Park at the AI Summit, the year after ChatGPT came out. I sat outside and did some radio hits, and someone said to me, “AI is never going to be able to put a red dress on and crawl the halls of power.” It’s true. What you want journalism for is someone who’s making sense of this all, putting it together and speaking to people about what’s going on.
“AI is never going to be able to put a red dress on and crawl the halls of power.”
Journalism cuts through the PR and gets to the truth of things. Back to the copyright point, these engines are being fed by people and human-written information. So we need to make sure we don’t lose the value of that.
📖 Next steps:
→ Experiment with NotebookLM (it’s free). Start with The Economist’s World Ahead notebook to see it in action.
→ Pre-order Katie’s book, The Curious Case of Mike Lynch, out next week.
→ Share your experience: What AI tools are you using for research or writing? Hit reply and let me know.
✨ Reading this week
“How do you change your destiny?…You change your destiny with knowledge. How do you gain knowledge? You read the news.” Great interview in the Guardian with former journalist and commentator Wang Jian.
What AI means for the business of journalism (h/t
)What is quality work in the era of AI? Thoughtful musings by
will bring you back to those halcyon Ghibli days.Best take on the AI bubble debate I’ve read, quoted this in so many conversations this week. From
.
Thank you all for reading! If you haven’t subscribed yet, join us.
Thank you Eleanor for this fascinating interview with Katie. She is using AI in such a smart and effective way. I really hope she doesn't get shade for this, as IMHO it's not too different from using a team of research assistants. You still need to check the work and have oversight but it also frees you up to better develop the narrative and line of inquiry. Would also be really interested to see how she addresses this in the book itself. A fantastic model to be inspired by though. 🙏