How Do You Turn a 40-Page Report Into a Decision?
Asking for a summary gives you a shorter document, which is not what you wanted. Five decision-shaped questions do the job instead — and the best tool for it is free.
By Merxtio Staff
You have forty pages and a meeting in an hour. You paste it in, type "summarize this", and get back three paragraphs that are accurate, bland, and no help at all.
That is not the tool failing. A summary is a shorter document, and you did not want a shorter document — you wanted to know what to do. This page covers the five questions that actually produce that, and the free tool that does it best. About seven minutes.
Why "summarize this" disappoints
A summary optimizes for coverage. It tries to represent the whole document proportionally, which means the two sentences that would change your mind get the same weight as the eight pages of background.
What you actually need is asymmetric. You want the parts that bear on one decision, and you want the parts that contradict what you already believe. Neither survives a process designed to be balanced.
So stop asking for the document back in miniature. Ask it questions that have consequences.
| Feature | What you get | What it is good for |
|---|---|---|
| Summarize this | A proportional, balanced condensation of everything. | Deciding whether to read the document at all. Genuinely useful for triage, useless for a decision. |
| What would have to be true? | The assumptions the recommendation rests on, stated plainly. | Finding the one load-bearing assumption nobody checked. |
| What is missing? | Gaps, unanswered questions and absent evidence. | Working out what to ask for before you commit. |
| Argue against this | The strongest case for the option you are not choosing. | Testing a decision you have already half made. |
| Where do these disagree? | Contradictions across several documents. | The thing manual reading is worst at and this is best at. |
That last one deserves emphasis. Finding where three reports quietly contradict each other is genuinely hard for a person reading sequentially, and it is the single most valuable thing these tools do.
Ground it in the documents
The property that matters for this work is not raw capability — it is whether the tool answers from your sources or from everything it has ever read.
NotebookLM is built around that constraint: you supply the documents, and answers come from them with references back to the passage. The free tier takes 50 sources per notebook, each up to 500,000 words, with a limit of 50 chat queries a day. For a stack of reports, contracts or research papers, that is more than most people will use.
Being source-grounded is what makes the answers checkable. When it tells you the recommendation depends on a growth assumption, it can show you the paragraph, and you can decide whether you believe the paragraph. A general assistant answering from memory offers no such thread to pull.
When the forty pages are a meeting
Often the long input you need a decision from was spoken, not written — a client call, a board meeting, three interviews.
| Item | USD per month |
|---|---|
| NotebookLM, free tier | Free |
| Otter Basic, 300 minutes | Free |
| Otter Pro, annual | $8.33 |
| Otter Pro, monthly | $16.99 |
| Otter Business, per user annual | $19.99 |
The five questions
State the decision before you open the document
You’ll have: One sentence naming the choice you are actually making. · about 2 minutes
"Should we renew this contract." "Which of these two vendors." "Is this market worth entering."
Everything downstream is aimed at that sentence. Without it you will get general-purpose analysis, which is how you end up with a summary again.
Ask what the recommendation depends on
You’ll have: A short list of load-bearing assumptions. · about 5 minutes
"What would have to be true for this recommendation to be correct?" is the highest-value question in this article.
Documents argue for conclusions and bury the assumptions underneath them. Surfacing those turns a forty-page case into three claims you can check, and usually one of the three is the whole decision.
Ask what is absent
You’ll have: The questions to send back before committing. · about 5 minutes
"What is missing that I would need to decide this properly?" Reports are written by people with a preferred outcome, and the most informative thing about them is often what they decline to mention.
If several documents are in play, add the contradiction question here: where do these disagree, and on what.
Have it argue against you
You’ll have: The strongest version of the case you are inclined to reject. · about 5 minutes
Name your leading option and ask for the best argument against it, then ask it to make the strongest possible case for the alternative you like least.
This is uncomfortable and it is the step that catches the expensive mistakes. A confident tool agreeing with you is worth nothing; a tool that makes you defend your position is worth the hour.
Check the passages before you rely on anything
You’ll have: A decision you can defend in the meeting. · about 10 minutes
Follow the references back into the document for anything you plan to say out loud. Source-grounded tools make this quick, which is the entire reason to prefer them for this work.
Anything you cannot trace, drop. Confidently wrong analysis is more dangerous than no analysis, which is the subject of its own guide in this series.
The pattern underneath all five is the one from the prompting guide: specificity and constraints narrow what the model can plausibly say, and "summarize this" is about as unconstrained as an instruction gets. If you are weighing whether any of this justifies a subscription, the four-question test applies — and for most people the honest answer here is that the free tier is the permanent one.
Questions people ask
- What is the best AI for summarizing long documents?
- For genuine analysis, a source-grounded tool such as NotebookLM, because it answers only from the documents you supply and references the passage. But summarizing is usually the wrong request — asking decision-shaped questions of the same document produces something far more useful.
- Is NotebookLM free?
- Yes, with real limits rather than a trial. The free tier allows 50 sources per notebook at up to 500,000 words each, 100 notebooks, and 50 chat queries a day. For most people working through a stack of reports that is more capacity than they will use.
- Can AI read a PDF and answer questions about it?
- Yes, and this is where it is strongest. Ask specific questions rather than for a summary: what the argument depends on, what evidence is missing, where two documents contradict each other. Prefer a tool that cites the passage so you can check the answer.
- How do I get AI to analyze a contract?
- Name the decision first, then ask what obligations it creates, what happens on termination, and which clauses are unusual for this kind of agreement. Verify every answer against the text. It is a reading aid for a professional review, not a replacement for one.
- How much does Otter cost?
- Basic is free with 300 minutes a month. Pro is $8.33 a month billed annually or $16.99 monthly, with 1,200 minutes. Business is $19.99 per user annually or $30 monthly and adds unlimited meeting transcription. Checked 21 August 2026.
- Can AI help me make a decision?
- It can make you better at making it, which is not the same thing. The useful moves are surfacing assumptions you have not examined, naming what is missing, and arguing against the option you already prefer. Asking it to simply choose gives you a confident answer with no reasoning you can audit.