DEV Community

Practical AI Workflows
Practical AI Workflows

Posted on

A source-first workflow for studying PDFs with ChatGPT and NotebookLM

AI summaries are useful, but they can hide the source trail.

When I use ChatGPT or NotebookLM with a long PDF, the fastest thing to ask is usually:

summarize this

The problem is that this compresses the document too early. A clean summary can look useful while making it harder to tell which claims came directly from the source and which parts were inferred or smoothed over.

For studying, that matters. I do not just want a nicer explanation. I want notes I can check.

The source-first order

The workflow I have been testing is:

  1. map the source
  2. extract source-backed claims
  3. ask retrieval questions
  4. generate the study guide last

1. Map the source

Before asking for a summary, ask the model to map the document:

  • what sections exist?
  • what does each section cover?
  • what claims appear in each section?
  • where should I check those claims?

A source map is not the final answer. It is a way to keep the document visible before the model starts synthesizing.

2. Extract source-backed claims

Next, separate the claims into buckets:

  • directly supported by the PDF
  • reasonable inference
  • needs verification
  • not supported

This makes weak points easier to catch before they get turned into polished study notes.

3. Ask retrieval questions

Instead of jumping straight to flashcards or a study guide, ask retrieval questions first:

  • what does this section define?
  • which paragraph supports this claim?
  • what evidence is given?
  • what is missing?

If the questions are vague, the source map probably is not good enough yet.

4. Generate the study guide last

Only after the source is mapped and checked should you generate:

  • a summary
  • flashcards
  • practice questions
  • exam-style explanations

The summary is still useful. I just do not want it first.

I wrote up the full workflow here:

https://practical-ai-workflows.pages.dev/

Top comments (0)