The Essential AI Research “Stack” (The Writer’s Process, part 4)

OR How I Went From Library Janitor to AI Research Master

This is the “old guy” in my brain. “Oh, you can’t remember that thing? I’ll have it for you next time you’re in the shower.”

This is the fourth post in the series “The AI Writer’s Process.” You can catch up in the puill-down nav menu on The AI Writers’ Room substack.

We’re getting deep into the process now. If you know someone who might find all of this valuable, please share this with them.

Subscribers -- please share your thoughts about researching with AI.

WHY IT MATTERS:

As you write, you will be going back to the research well repeatedly. You’re always going to be checking your assumptions about facts, characters, settings, your audience and more.

Setting up your research “stack” properly will ensure that you have accurate results that are stored and organized for easy access and that can be reviewed and revised easily.

I’ve developed a research stack that will help you:

• Set up and manage your AI research “team”

• Capture and organize discoveries

• Verify findings while preserving serendipity

I developed these techniques by combining traditional writing methods with modern AI capabilities. They work across genres - fiction, non-fiction, technical writing, and business content.

IN THIS POST (It’s about a 10 minute read)

- Set up 3 AI tools as your research team

- Follow the 40/40/20 research time allocation

- Use the "librarian prompt" technique

- Balance structured research with serendipity

LIBRARIES ARE FOR DISCOVERY

My first job was as the janitor / gardener at my local library. I worked when the library was closed. I’d clock in, do my job, read for an hour or two and clock out. Years later, I ran into the head librarian.

A word about the head librarian: She was a lot more than the town had bargained for. She was educated, well-read, and was as likely to put Ken Kesey or Lawrence Ferlinghetti on the “This Just In” shelf as the latest bestsellers.

I told her I thought I owed the library money for all of the hours I had been reading instead of doing my job.

“Oh, honey,” she smiled, “We knew. You left the books on the table. Libraries are for discovering... that’s what you did.”

THE LIBRARIAN TECHNIQUE

Remember card catalogs? That moment of panic when you're looking for one book and find 150? That's actually the sweet spot for AI research.

Here's how to turn AI into your knowledgeable librarian:

• Ask for "adjacent" viewpoints and opposing perspectives

• Request identification of your biases

• Challenge conventional wisdom

• Share your experiences and ask AI to analyze them

• Use this prompt: "Pretend you are a wise and knowledgeable librarian. What advice would you give about this research?"

YOUR RESEARCH TOOLBOX

Here’s what I’m using to organize my writing and research.

• Claude: Goes wide, exploring alternative viewpoints and connecting seemingly unrelated ideas

• NotebookLM: Goes deep, working with your loaded documents to integrate new ideas with existing research

• Gemini: Brings in current information and easily exports to Google Docs

• Google Docs. The Docs interface doesn’t “get in the way” like other writing tools. AND, they are easily accessed by both Claude and NotebookLM.

Here's how each tool serves a specific purpose:

Claude

• Explores alternative thinking paths

• Finds unexpected connections

• Challenges your assumptions

NotebookLM

• Analyzes your existing research

• Integrates new ideas with your work

• Maintains context across sessions

Gemini

• Provides current information

• Exports easily to Google Docs

• Remembers conversation context

Google Docs

• Stores your findings

• Feeds back into NotebookLM and Claude

• Creates a searchable research library

---

THE PROCESS IN ACTION

Here's my workflow:

1. Heart: First draft with initial ideas

2. Head: Second draft with structured thinking

3. Second Brain: AI review and suggestions

4. Integration: Blend AI insights with your vision

5. Deep Dive: NotebookLM analysis

6. Fact Check: Gemini for current details

Pro Tip: Name your files consistently: "Research - [topic] - [project] - [date]"

A REAL-WORLD EXAMPLE: RESEARCHING AN AI COLLAPSE

Let me show you how this works with my current project, "Autonomous" - a near-future story about global AI failure when AI is woven into everyday life and then... suddenly collapses. I needed to understand how systems collapse and how people react.

There’s a “Job To Be Done” for the reader, the kind of job writers like Michael Crichton do incredibly well: Ground the imaginary scenario in facts and patterns that will be recognizable to the audience. The townspeople in Amity react to the Jaws shark the way we know people react. The people in Jurassic Park -- at first -- react the way we know people tend to react when faced with runaway technology.

I asked all three AIs: "What historical incidents parallel a broad systemic computer failure?”

Each AI approached it differently:

• Claude explored broad patterns and human reactions

• NotebookLM connected incidents to my existing research

• Gemini found current parallels and expert sources

The results were surprising and interesting. Here are the most relevant historical parallels:

• The Y2K bug (didn’t happen, caused billions to be spent)

• The 2003 Northeast Blackout. 50M people without power for at least 2 days.

• The Morris Worm (1988): First major internet worm infected 6,000 computers (10% of the internet), causing $100M in damages and exposing how interconnected systems could be vulnerable

• The Flash Crash (2010): Stock market plunged 9% in minutes due to automated trading systems, wiping out $1 trillion in market value before mysteriously recovering

• The Arpanet Crash (1980): A single corrupted status message spread between nodes brought down the entire predecessor to the internet for 4 hours

• The AT&T Network Collapse (1990): A single software bug caused a 9-hour nationwide outage affecting 60 million phone calls, revealing how centralized systems can fail

• The Knight Capital Failure (2012): A trading firm lost $440 million in 45 minutes when new software deployed incorrectly, showing how quickly automated systems can spiral out of control

As I review these, I’m making notes... Will my characters look at the collapse and dismiss it as a Y2K hype? Will it be caused by a tiny software error, like the Knight Capital Failure? Will the world be plunged into darkness like the 2003 Northeast Blackout. ANY of those things could work in my story, and all would be “recognizable” by the reader.

Bonus Discovery: Gemini pointed me to E.M. Forster's "The Machine Stops" (1909) - a prescient story about humanity's dependence on interconnected machines.

---

REALITY CHECK: VERIFYING YOUR RESEARCH

Always verify what AI tells you:

1. Cross-reference with multiple sources

2. Ask AI to challenge its own findings

3. Contact real experts (they're usually happy to help)

4. Test ideas through character dialogue

5. Feed conclusions back to AI for consistency checks

I had to do tons of research when I was writing a story about the possibility of earthquakes in New York City. The people I called loved hearing from someone who was genuinely interested in their work. I still remember the ominous answering machine message I got one day: "Fred. This is Carl. Yes, 50 years. You can expect a quake within 50 years"

---

RESEARCH IS PART OF PRODUCT DESIGN

Good research helps you understand both what your audience needs to know and what they already know. Research reveals the facts you need to know. AND it can ignite new ideas and approaches that make your writing stronger.

YOUR TURN

I'd love to hear about your AI research experiences:

• What unexpected connections has AI helped you discover?

• Which historical parallels resonate with your work?

• Have you tried the librarian prompt? What happened?

Share your stories in the comments - I'll be there to discuss research strategies and surprising discoveries.