Skip to main content
All Notes

My AI research stack

Most people use AI to search faster. I use it to understand deeper. Here's the actual chain.

Most people use AI to search faster. I use it to understand deeper.

There's a real difference. Searching gives you more information. Understanding gives you fewer, better ideas.

Here's the actual chain:

  1. Start with a question I genuinely don't know the answer to
  2. Ask ChatGPT to write a detailed research prompt for Gemini — specifically to crawl the web for the best sources and expert perspectives on the topic
  3. Use Elicit to find academic papers — it follows citation chains, not just keywords
  4. Dump everything — PDFs, links, notes — into NotebookLM and start asking questions
  5. Switch NotebookLM to analyst mode, shorter outputs. Map the major opinions.
  6. Form my own view. Then ask for evidence that supports or challenges it.
  7. Bring that to Claude for writing.

The step most people skip is step 6 — forming an actual opinion before writing. That's why most AI-assisted content reads like a summary. Because it is.

The chain above doesn't skip it. The goal isn't to collect more. It's to reach the point where I have something to say — something that's true but that most people haven't yet thought all the way through.

That's the only thing worth writing.


Tools referenced

  • ChatGPTchat.openai.com. Used here specifically to generate better research prompts for other tools, not as the primary research surface.
  • Geminigemini.google.com. Better at live web search than ChatGPT. Used for sourcing current expert perspectives and articles on a topic.
  • Elicitelicit.com. Academic research tool that traces citation chains rather than just matching keywords. Significantly better than Google Scholar for building a real understanding of a topic.
  • NotebookLMnotebooklm.google.com. Google's document Q&A tool. Best used in "analyst mode" with your own sourced materials loaded in. Excellent for synthesizing across multiple long documents.
  • Claudeclaude.ai. Used at the writing stage. The reasoning quality for long-form synthesis and structured thinking is noticeably different from other models.

On the methodology:

  • The "form your own view before writing" principle connects to Cal Newport's Deep Work — the argument that real understanding requires uninterrupted synthesis time, not just information collection.
  • The distinction between "searching faster" and "understanding deeper" is the core of How to Read a Book by Mortimer Adler — a surprisingly useful text for thinking about what engagement with ideas actually looks like.

You might like these too!