Using NotebookLM to help with foreshadowing

Being able to switch on and off what it knows can be a powerful critical tool

A while ago, I wrote about how I’ve been using NotebookLM as a writing assistant. I don’t let it do any of the actual writing or plotting, but I’ve found it very useful to help me keep track of details and to critique my work. Today, I accidentally discovered a neat trick for getting insights that I’d find hard to get from a human reader.

I’ve loaded it up with a wealth of background information, notes, and ideas, as well as a complete outline and the latest draft of The Yellow Flowers. So far, that’s just a Prologue and Chapter 1. Not much, I know, but as we all know, if the opening of the book sucks, nobody’s going to read the rest of it.

My initial intention was to get it to cross-check the draft against the notes and make sure that I’d got all the names right, and so on. Instead, I tried something else. I gave it just the draft, with none of the supporting documents and told it to critique it, focusing specifically on anything that seemed unclear or inconsistent. This meant it was approaching the story just like a reader: all they know is what’s in front of them. It picked up several things: this character’s motivation doesn’t make sense, there was too much focus on this aspect of the set-up, the narrator’s perspective seems odd, and so on. All valid criticisms, to be fair.

I then gave it the outline as well, and told it to look again, focusing specifically on the things it had just commented on. Now it gave me completely different feedback. It talked about how all these apparent inconsistencies and extraneous details made a lot more sense once you have the full picture, and how I was foreshadowing and setting up things I’d be returning to later. It also noted how this was going to subvert a lot of the reader’s expectations. (And yes, it did also point out some places where I could streamline the story by shifting the perspective or incorporating descriptive passages into the action. Some good ideas I can work with, some not-so-good ones I’ll be ignoring, just like I’d get from any other beta reader.)

The wonderful thing about this is that I’ve effectively got independent opinions from two completely different readers. One only knows what they’ve read so far, just like the people in my writing group or the people who will (I hope) eventually read the book. There are things they find confusing or odd, but for the most part, that’s intentional. I need to check that what they’re seeing at this point in the story is what I want them to see. I was able to ask it things like “what is the most likely explanation for this?” and it could tell me what a typical reader is likely to be assuming at that point. And, to my satisfaction, it told me exactly what I was hoping for.

The other reader already knows exactly where the story is going and has an entirely different point of view. They’re advising me from an almost technical perspective on whether I’ve laid the groundwork successfully for what’s going to come later so that I’m not cheating and my big reveals are going to work. They can tell me whether I need to add emphasis to some of the foreshadowing, or whether I’ve been too heavy-handed and unsubtle. My writers’ group can’t do that for me because I haven’t told them how the story ends. And again, I was pleased to see that it looks like it’s working. It even pointed out some subtleties about the narrative structure that I hadn’t noticed, and made some interesting suggestions about how I could add some callbacks later in the story to make the reveals even more effective.

It’s like practicing a magic trick. I haven’t done the trick yet, but I need to know whether I’ve successfully misdirected the audience while keeping them interested. I also need advice from a fellow magician who knows how the trick works on whether I’ve done all the things I need to do in order to actually do the trick when I get to it.

I can imagine that this would be a really useful tool for anyone writing mysteries or detective stories, where it’s essential to keep track of what the reader knows and cross-check it with what’s happening out of sight and the things the reader doesn’t know yet. It would also be great for stories with an unreliable narrator.

So, as I said at the beginning and in my previous post, I’m not a believer in AI-generated writing. On the other hand, an AI beta reader is an amazing tool to have at your beck and call.