Research

Summarisation

Compress documents without losing the point.

Knowledge cards
Best practice
Ask for Quotes and Locations, Not Just Claims
A grounding habit: ask the summary to attach the exact supporting quote and its location to each key claim, so you can verify the point without re-reading the whole source.
L2 · Context engineering
Knowledge card
Tell It the Audience and the Length First
A framing step: state who the summary is for and how long it should be before asking for it, so the model keeps what that reader needs and cuts the rest.
L2 · Context engineering
Workflow
Summarise Long Docs in Chunks, Then Combine
A workflow for documents too long to summarise reliably in one pass: summarise each section on its own, then ask the model to synthesise the section summaries into a single coherent overview.
L2 · Context engineering
Knowledge card
Summarise to a Decision, Not Just a Shorter Version
Summarising a document toward the decision you actually have to make — 'what should I do about this' — rather than asking for a generically shorter version.
L2 · Context engineering
ChatGPT summarisation

Summarise to a decision, not a shorter version

A generic 'summarise this' gives you a compressed paraphrase — shorter, but not necessarily more useful. The better move is to tell ChatGPT what the summary is for and who will read it. 'Summarise this contract for a founder deciding whether to sign, flag anything unusual' produces a different, sharper output than an even shrink. The purpose tells ChatGPT what to keep and what to drop, which is the whole point of a summary in the first place.

Ask for the shape you need

Different decisions need different shapes: three bullet takeaways, a one-line verdict with reasons, a table of pros and cons, the key risks only. Specify it. And for anything you will act on, ask ChatGPT to note what it was unsure about or could not find in the source, so you know the limits of the summary. A summary that hides its gaps is more dangerous than a longer read; one that flags them is genuinely useful.

Frequently asked
How do I summarise a document with ChatGPT?

Tell it what the summary is for and who will read it, and specify the shape you want — bullets, a verdict, a risk list. Purpose is what lets it keep the right things and drop the rest.

Why are ChatGPT's summaries too generic?

Because 'summarise this' gives no purpose, so it compresses evenly. A summary aimed at a specific decision and audience is far more useful than a neutral paraphrase.

Can I trust a ChatGPT summary?

Mostly, but ask it to flag what it was unsure about or could not find in the source. A summary that surfaces its gaps is safe to act on; one that hides them is not.