Speed-Reading Dense PDFs
Research triage.
Real researchers triage. They figure out which papers (or which parts of which papers) deserve close reading, and they skim or skip the rest.
The skill of triage is what AI accelerates. Not "Claude reads it for you." The honest version is: you've already verified the source exists (Lesson 5.2); you upload the PDF; Claude gives you the structural map of the paper; you decide which 10% to read closely; you read it. The synthesis comes from your reading of the part that matters, not from Claude's summary of the whole.
The Structural Map prompt: your first move on every PDF.
This is the prompt that turns a 30-page paper into a navigable thing. It doesn't tell you what the paper says. It tells you where the paper says the parts you care about. You still do the reading; Claude does the table-of-contents work.
What this prompt does that "summarize this paper" doesn't
"Summarize this paper" gets you Claude's general gist and removes you from the source. The structural-map prompt does the opposite: it gives you the navigation, not the substance. The bullets and section map tell you where to put your eyes; the actual reading still happens in your head, on the PDF. You finish this prompt knowing which parts of a 30-page paper you need to read closely. Then you read those parts.
The reading order that actually saves time.
Once you have the structural map, here's how to actually use it. The order matters; it puts you in the author's head before you go hunting for quotes.
1 · Abstract + intro + conclusion
Always read these in full from the PDF, not from Claude's summary. Authors compress their actual claims here. The intro tells you what they think they did; the conclusion tells you what they think it means.
2 · The 5 paragraphs Claude flagged
These are the paragraphs the structural map said are most relevant to your topic. Read them carefully, twice if needed. Take notes in your own words; don't copy-paste.
3 · The "what it's not arguing" section
The author's own caveats are gold for your synthesis lesson (5.4) and your critical-thinking work (5.5); they tell you exactly where the argument is weakest.
4 · Verify any quote you're going to use
Ctrl-F the quote in the PDF. Confirm it's there. Confirm the page number.
What you skip: and what you don't
Skip: technical methods sections that aren't relevant to your argument; literature review sections that just summarize prior work; long quantitative tables and appendices unless your paper actually uses the data. Don't skip: the introduction, the conclusion, and the parts of the paper that touch your specific argument. The whole point of the speed-reading workflow is that you still read, just better-targeted.
Two more PDF-handling prompts.
The structural map handles most papers. These two come out for the harder cases: dense jargon, or class-assigned readings where the prof's framing matters more than the paper's framing.
What you wrote in your notes is what counts: not what Claude said.
The notes-from-real-reading rule
- What goes in your notes: things you read in the actual PDF and put in your own words.
- What does NOT go in your notes: Claude's summary text, copy-pasted. That's not your reading; that's Claude's reading, attributed to the source.
- Why it matters: if your paper is built on Claude's summary of a paper you didn't read, your paper is one degree of separation from the source. When you cite it, you're citing what Claude told you the source said. That's how mis-attributions and invented quotes show up in final drafts.
- The fix: use Claude's structural map to navigate the paper. Read the relevant 10%. Take notes from your reading, in your own words. Cite from your notes. The chat with Claude is a tool, not a step in the citation chain.
Try this: run the workflow on your next reading
Pick a PDF you actually have to read this week. Run it like this:
- Verify the source exists (you have it from class, so this one's free).
- Run the Structural Map prompt with the PDF attached. Read Claude's output.
- Open the actual PDF. Read abstract + intro + conclusion.
- Find the paragraphs Claude flagged. Read them closely.
- Read the "what it's not arguing" passages.
- Take notes in your own words. Don't copy-paste from Claude's output.
You'll have a real working understanding of a paper, done honestly, with notes that came from your own reading. That's the speed-read move done right.
The deeper move: AI for navigation, you for the reading
Honest Work Code · Rule 2: your work survives scrutiny. The line between "AI helped me read more efficiently" and "AI read the paper for me" is exactly the line between this workflow done right and done wrong. The structural map is navigation: a table of contents for the parts of the paper that matter to your topic. The reading itself, the comprehension, the notes: those are yours. When your paper claims this source argues X, that claim should be rooted in your own eyes on the page where the author argues X. Run the workflow this way and you'll read more papers, more carefully, and your bibliography won't contain a single sentence you didn't earn.
Up next: putting the papers in conversation with each other.
You've now read three or four papers carefully. The hard part of any research paper isn't the reading. It's putting the papers in conversation with each other. Lesson 5.4 is the synthesis matrix: comparing arguments side-by-side, finding the gap between sources, and using that gap to power your own thesis. (Plus: how this scales up when the assignment is a long-form research paper or thesis.)
Continue to 5.4 → Synthesizing Across Sources