How to Get ChatGPT to Actually Summarize a Long Document Right

Quick answer

A generic “summarize this” prompt gives you a generic summary, one built around what's usually important in that kind of document, not what's important to you specifically. The fix isn't a longer prompt. It's a more specific one: say who the summary is for and what to keep, before you paste the document. That one change fixes most bad summaries without adding much typing.

Why the default summary keeps missing things

Ask ChatGPT to summarize a report with no other instructions, and it makes a reasonable guess at what a typical reader would want, which is usually the headline numbers and the stated conclusion. If what you actually needed was a caveat buried on page four, or a specific number that contradicts the headline, a generic summary has no way to know that mattered to you, because you never said so. It's not a failure of the model. It's a prompt that didn't include the one piece of information it actually needed.

Same document, but only the second version surfaces the one number that actually changes the decision.

The structure that fixes it

Three things, in order: who the summary is for, what to keep even if it seems minor, and then the document itself.

“Summarize this for someone who has to decide whether to approve the budget, not someone learning about the project for the first time. Keep any number that changed from the original estimate, even if it's mentioned only once. Here's the document: [paste it].”

That's a longer prompt than “summarize this,” but not by much, and the difference in what comes back is real. The model now has a target to summarize toward instead of a generic best guess.

What to specify, and what to skip

Always worth specifying: who's reading it, and what counts as important enough to keep even if it's a small detail. Worth specifying sometimes: length (a specific sentence or word count, if you have one), and format (a list vs. a paragraph, depending on where the summary is going).

Not worth specifying most of the time: tone. Summaries are already a plain, functional format, and over-directing tone tends to add words without adding value.

For documents too long to paste at once

Split the document into logical sections, summarize each section using the same keep-this instructions, then run one more pass asking the model to combine the section summaries into a single final one. This keeps your original instructions applied consistently across the whole document instead of the model gradually drifting as a huge single paste gets longer than it can hold onto well.

Turning this into an actual habit

For more prompts covering writing, decisions, and learning beyond just summarizing, see 20 ChatGPT Prompts to Try This Week.

FAQ