Reverse content creation: Using AI conversations to unlock better writing
Remember the hype around Google Notebook LLM? When it was released a while ago, the AI community’s reaction was unanimous: Podcasts, and so, wow! I immediately created a podcast based on an article for my employer at the time. It was fascinating because you could have AI models discuss a text. Two AI “talking heads,” typical podcast hosts, would exchange thoughts about the article’s content. And it was amazing! It was more than just text reproduction.
I was genuinely impressed by how much better the benefits and value for the target audience were elaborated in this podcast format than in the original text. So I proposed internally to revise the original article to incorporate these new insights – but unfortunately, the project didn’t gain traction. Nobody had time to listen. And now they are missing out, big time!
The new generation of AI conversations
This often happens: New technologies are celebrated initially, but the hype quickly fizzles out. The first Notebook LLM podcasts also quickly lost their appeal because the AI hosts’ voices were always the same. It was a cool proof of concept but not really sustainable. Yet it would be so valuable to have a format that automatically illuminates a text on a different level, opening up new perspectives.
But now there’s a new development that has brought me back to this topic: Notebook LLM has received an upgrade! The new model (Gemini Flash 2) is even better and offers one crucial improvement: You can now give prompts to the AI moderators. This means you can control which aspects they emphasise or from what perspective they should examine a topic. As a result, it’s no longer such a gamble as before. The outcome is much more focused and relevant.
Reverse content generation: Using podcasts to generate content
This gave me an idea that I call “Reversed Content Generation.” Instead of starting with an article as usual, you could go the opposite way. You give your notes, drafts, or ideas to the AI moderators and let them talk about them. These conversations are often more dynamic and entertaining than static text. The whole thing is then transcribed, and from that, articles, blog posts, or even presentations emerge.
One crucial point to emphasise here: the quality of your input directly affects the output. While the AI hosts will eagerly discuss any material you provide, they can’t magically transform poor content into valuable insights. If you feed them unfocused or irrelevant material, they’ll still discuss it enthusiastically – but your listeners (and eventually your readers) will quickly recognise the lack of substance.
The magic happens when you provide well-thought-out source documents and clear direction: Essentially, we’re turning the podcast into a format through which we can better understand and enhance our own content. It’s a kind of brainstorming where AI models interact with us and provide new ideas and aspects. Through the more emotional, dialogue-like character of the podcast, we can better reach the target audience, develop a feeling for their needs, and optimise the content accordingly.
My experiment and the results
I tried this myself. I had some notes and transcripts from my previous AI projects that I wanted to organise and structure. Instead of just sitting down and writing a white paper (actually, I did write a white paper, but did not polish it), I input these contents into Notebook LLM, generated a podcast, and then transcribed it. From the transcription, an article emerged, which I rewrote and refined, and also created a presentation using Gamma AI. But that is a topic for another blog post.
If you’re interested, read the article and listen to the 22-minute-long podcast where the AI-powered talking heads discuss my projects (needless to say, those pundits are quite impressed).
Conclusion: Rethinking content creation
The “Reverse Content Generation” approach is fascinating and is another tool in our content creation toolbox. Instead of always starting with a lengthy article, we can use AI-driven podcasts to illuminate the content in a new way, making it more emotional and relatable. This captures the target audience’s interest, and subsequently, we can develop a much better text.
I recommend everyone try this approach themselves. It’s worth revisiting technical developments after a while, as they often evolve and open up entirely new possibilities. Interested to explore with me? Drop me a line.