Summarize long PDFs into a one-pager

I read a lot of dense PDFs. This gets me the signal in two minutes instead of forty.

The workflow

1. Upload the PDF to Claude
2. Ask for a one-page brief with key numbers
3. Request a short list of risks and open questions
4. Save the brief next to the source

Result: A one-pager you can actually act on.

The prompt

Summarise this document into one page: the thesis, the 5 key data points, the main risks, and 3 questions I should ask. Be concrete.

A note on what follows. The workflow, the prompt and the result above come from the member. The notes below (why the prompt works, what to watch for, how to adapt it) were written with AI help.

Why this prompt works

The three questions at the end are the reason I still use this. A summary tells me what the document says. The questions tell me what it avoided, and in a vendor report or an analyst deck the avoidance is the story. Asking for exactly five data points forces a choice too. Say "the key data" and you get every number in the file, reordered. "Be concrete" I added after a run that gave me "the market is growing rapidly" three times, which is not a data point. What I stopped doing is trusting the thesis line on long documents. On anything past forty pages it tends to describe the executive summary, which is the part the author wrote to be quoted, not the part that matters.

What to watch for

• Charts are mostly invisible to it. If the good number only exists in a graph, it will not be in your one-pager.
• Footnotes and methodology get skipped, which is where the caveats on the headline figure usually live.
• Long documents get summarised from the front. Sections past the halfway mark are underrepresented.
• Risks come back generic unless the document names them. "Market risk" means it found nothing.

How to adapt it

• Force page anchors: Give the page number for every data point. It makes checking possible and cuts invention.
• For anything with an agenda, add Say who paid for this document and what they want the reader to conclude.
• If the back half matters, split the file and add This is part 2 of 2, do not summarise part 1.

What good output looks like

The five data points should be numbers with units and a page reference, not sentences containing a number. The questions should be ones I could actually put to the author, and at least one should be uncomfortable. If the risks section reads like a disclaimer, the document either has none worth naming or it buried them, and either way I go read the methodology myself.