Highlighting vs summarising: which one should you actually use?
These two get discussed as competing tools for the same job. They are not. They produce different artefacts, and the difference is not a matter of degree.
A summariser is lossy and generative: it reads your text and writes new, shorter text about it. A highlighter is lossless and selective: it keeps every word you gave it and adds emphasis on top.
Nearly every practical question about which to use follows from that.
What you actually get out of each
A summary is a new document. It is shorter, it reads smoothly, and it has been separated from the original — the hedges, the conditions, the specific numbers and the author’s exact phrasing are gone, because removing them is how it got shorter.
Highlighted text is your original document with a layer of emphasis. Nothing has been removed or rewritten. You can read the sentences on either side of any mark, check whether a claim was qualified, and see the full context of a statistic.
So the real question is whether the context you are throwing away matters for what you are about to do.
When summarising is the right call
Summarise when you need the gist and will not be held to the detail:
Deciding whether to read something
A three-line summary of a forty-page report is enough to tell you if it is relevant. You are going to read the thing properly if it is.
Briefing someone else
Nobody wants your highlighted PDF. They want four sentences in an email.
Collapsing a long thread or transcript
Meeting notes and email chains are mostly redundancy. Summarising removes it, and there is no exact phrasing worth preserving.
Writing something in your own words
If the next step is your own prose, having to restate the argument is useful work, not lost information.
When highlighting is the right call
Highlight when the original wording carries the weight:
Revising for an exam
You are often assessed on precise definitions. A summary that says “osmosis involves water moving across a membrane” has quietly dropped the water potential gradient, and that is where the marks were.
Anything you will quote or cite
You need the author’s actual sentence. Working from a paraphrase and re-inserting quotation marks later is how accidental plagiarism happens.
Legal, medical, technical and contractual text
Here the hedges are the content. “May” versus “shall”, “within 30 days” versus “promptly” — a summariser treats these as noise to compress.
Text you do not fully trust yet
Highlighting keeps you able to audit. You can always read around a mark to see whether it was stated as a finding or a speculation.
The hidden difference: where the errors go
This is the part that rarely comes up and matters most.
When a summariser is wrong, it is wrong fluently. A generative model can produce a confident sentence that is not supported by your text at all, and because the output is smooth and plausible you have no signal that anything went wrong. Checking requires going back to the source — which is the work you were trying to avoid.
When a highlighter is wrong, it marks a boring sentence. You read it, notice it is unremarkable, and ignore it. The failure is visible, local and cheap. A highlighter cannot assert anything the author did not write, because it has no mechanism for writing.
For anything you will be held accountable for, that difference in failure mode is worth more than any difference in output quality.
There is also a cost and privacy difference
Generating new text requires a large model, which in practice means sending your text to someone else’s server and paying per request. That is why summarisers have word caps, sign-ups and subscriptions, and why pasting an unpublished draft or a client document into one is a decision you should make deliberately.
Extractive highlighting is just statistics over the words in front of it — graph ranking and keyword weighting, no model. It runs in a few milliseconds inside your own browser tab, which means it can be free with no limits, and your text never has to leave your device.
The answer is usually both, in this order
Highlight first, summarise second — and write the summary yourself.
Use highlighting to reduce the document to the ten per cent that carries the argument, with everything still in context and checkable. Then write three sentences from those marks in your own words. You end up with a short summary you can trust, built from verified content, plus the annotated original to go back to when the summary turns out to have dropped something.
That sequence also happens to be the one the learning research supports: the restating step is where the work pays off, and it only pays off if you are the one doing it.