How to summarise a long document or report with AI, properly
AI can turn forty pages into the bit you actually need, if you ask the right way. Here is how to get a summary you can trust.
There is a forty-page report sitting in your inbox, and you have read the first page three times.
You are not lazy. We are all drowning in documents nobody has time for. Reports, contracts, proposals, research, threads that went on far too long.
AI is brilliant at cutting these down to size, which is why "summarise this" is one of the most common things people ask it. It is also one of the easiest to do badly.
A lazy summary gives you a bland paragraph that misses the one line that mattered. A good summary, asked for properly, gives you exactly the bit you needed and tells you where to look in the original. Here is the difference.
Summarise for a purpose, not in general
"Summarise this report" forces the AI to guess what you care about.
Tell it instead. Are you looking for the risks? The costs? The recommendations? The bits that touch your team? A summary aimed at a question is far more useful than a summary of everything.
Instead of "summarise this," try "pull out the three things in here that affect my budget, and anything I would need to act on this month."
Ask for structure
Tell it the shape you want back. The key points, the decisions needed, the open questions, the numbers that matter.
A structured summary is scannable, and far harder to get vaguely wrong than a wall of prose.
Make it show its working
Ask it to point you to where each claim comes from, a section, a page, a heading.
This does two things. It lets you check the important bits in seconds, and it discourages the AI from inventing a tidy summary of something the document does not actually say.
Every model can only hold so much text at once. That budget is the context window, measured in tokens, the small chunks of text it reads in. A genuinely long report can overflow it, so tools quietly split the document into chunks and summarise each before stitching the pieces together, a process called chunking. That is why a buried detail sometimes vanishes: it sat in a chunk that got compressed away. It also helps to know the two kinds of summary. Extraction lifts sentences straight out of the text, faithful but blunt. Abstraction rewrites the gist in fresh words, smoother to read but the point where things get subtly bent, a "subject to approval" softened into a flat "approved." Asking it to cite the section it drew from nudges it toward extraction for the facts that matter, which is exactly why showing its working makes a summary you can trust. Knowing this is also why, for a very long document, you summarise section by section rather than firing the whole thing in and hoping.
Always spot-check the load-bearing facts
Here is the honest caveat. AI can misread, skim past a crucial qualifier, or smooth a "maybe" into a "yes."
For anything you will act on, a figure, a deadline, a term in a contract, read that specific line in the original. The summary gets you to the right page fast. It does not replace your eyes on the part that matters.
Use it to ask follow-up questions
The real value is that the document becomes something you can interrogate.
Once it is summarised, ask follow-ups. "What does it say about cancellation." "Is there anything here I should worry about." "What is missing." You go from passively reading to actively questioning, which is faster and usually sharper.
Done this way, a forty-page document stops being a chore you keep putting off, and becomes five minutes and a clear head.
If your team is buried in documents that need reading and acting on, setting up a reliable way to cut through them is exactly the kind of thing we build.
Book a quick chat →Related: How to use AI to make better decisions from your own data.
Common questions
How do I get a good summary from AI?
Summarise for a purpose, not in general. Tell it what you care about, ask for a clear structure, and have it point to where each point comes from in the document so you can check the important bits.
Can I trust an AI summary of a contract or report?
Use it to find the right sections fast, but read any load-bearing line, a figure, a deadline, a contract term, in the original yourself. AI can smooth over a crucial caveat, so verify anything you will act on.
What is the best thing about summarising with AI?
It turns a document into something you can question. Once it is summarised you can ask follow-ups, what does it say about X, is there anything to worry about, which is faster and sharper than reading front to back.