SearchResearch (1/22/26): Searching for insights with AI (NotebookLM for sensemaking)

 I have two really big interests these days... 

... sensemaking...

(that is, how people search for and discover how to make sense of a complex topic)

AND 

...  unanticipated consequences...

(that is, how things go wrong despite your best intentions).  

So it won’t surprise you to learn that I often try to do both at once.   This week I was doing a bit of sensemaking of my book, unanticipated consequences.  

Of course, I’ve been trying to use Google’s NotebookLM to help me understand (that is, "make sense of") the unanticipated consequences space a bit better.  (Henceforth, I’ll abbreviate it as UC.)  

One generally useful method of sensemaking is to try and look at your collection in different ways.  Here's how I got NotebookLM to help me look at my UC collection in several different ways.  

1. Try creating a mindmap to see the main categories of your collection.  

In my case, I’m trying to see the key ideas in a collection of texts all about “unanticipated consequences.”  I added several UC texts to my NotebookLM and then clicked on “Mind Map.”  It created this lovely chart with the major categories that it saw in my texts… (click on the image to see the full thing):  

2. But I need this as a text outline–how to do that?  

That’s a nice graphic, but I really want this as an outline to copy/paste into the text I’m writing.  How can I do that?  

Easy.  Pull it into Gemini and ask for a textual outline version of the image.  

The full text looks like this: 

Unanticipated Consequences

1. Core Reasons for Failure

• Scope Creep

• Resource Overallocation

• Requirements Volatility

• Cost Overruns

• Schedule Slippage

• Benevolent User Fallacy

2. Categories of Consequences

• Perverse Incentives

• Population Explosions

• Hypersuccess Effects

• End-of-Life Failures

3. Notorious Historical Failures

• Technology & Software

• Sony Betamax

• Apple Lisa/Newton

• Microsoft Bob/Zune

• Google Glass/Google+

• Healthcare.gov

• Stuxnet Virus

• Automotive & Transport

• Ford Edsel/Pinto

• DeLorean DMC-12

• Concorde

• Boeing 737 MAX

• Tesla Cybertruck Rust

• Consumer Goods & Food

• New Coke

• Crystal Pepsi

• McDonald's Arch Deluxe

• Colgate Kitchen Entrees

• Iridium

• Environmental & Infrastructure

• Australian Rabbit Invasion

• Plastic Pollution

• Bangladesh Arsenic Wells

• The Big Dig

• Garden Bridge

4. Lessons and Prevention

• Mental Model Analysis

• Post-Mortem Documentation

• Maintenance/Repair Planning

• End-of-Life Strategy

• Real-time Progress Monitoring

• Rigorous Market Research

Which is a bit more useful.  (More generally, you can often use an AI tool to do conversions like this.)

3. Another way to look at the categories of your collection: Ask for dimensions 

Now that I have this collection of documents about UC, what would it mean to “make sense” of them?  One way to do this (which is something I might as a clever research assistant) is to ask for a list of the “key dimensions” of the collection.  Here’s what I did: 

NLM gave me a nice breakdown, not a perfect breakdown, but a great starting point for my writing.  

4. Yet another way to look at your collection:  Infographic 

Click on the “Infographic” button on the right side of your NotebookLM (in the Studio panel).  It created this for me: 

Interesting that this is rather different than the outline from above.  Nonetheless, it’s pretty compelling AND it gives yet another way to make sense of the collection by giving us the “Core Pillars of Project Failure,” “Lessons from the ashes,” and “Financial Impact.”  

 

5. When NotebookLM gives you an error for a web page, try using the Pure.md hack  

This happens to me all the time–I add a source to my NotebookLM and get this error: 

The best workaround I’ve found is to prepend the string: https://pure.md/ in front of your url.  

So if you’re trying to load up 

https://www.forbes.com/sites/the-prototype/2026/01/22/these-ai-models-might-take-down-superbugs/ 

Just prepend (that is, add to the front of the URL) that magic string so it looks like this: 

https://pure.md/https://www.forbes.com/sites/the-prototype/2026/01/22/these-ai-models-might-take-down-superbugs/

The Pure.md site converts the text of the page to “pure markdown” format, which NotebookLM can easily ingest.  Which leads to success: 

Hope you find these lessons of use in your work. 

Keep searching!