Let us begin with simple definitions, because the two words are often mixed up.

Summarizing means shrinking a longer piece of material into a shorter one while keeping the main ideas. A summary answers the question, “What is this about, in brief?” It is like the blurb on the back of a book, or the short recap at the start of a meeting.

Extracting means pulling out specific pieces of information from material and presenting them separately. An extraction answers the question, “What are the particular facts I am looking for?” Examples include a list of names, dates, amounts, action items or complaints. It is like running a highlighter over a page and then copying only the highlighted parts onto a clean sheet.

Reading, understanding and restating are exactly what language models are built for.

Claude has learned from a very large amount of text, which gives it a strong feel for how ideas are organised and how language works. When you hand it a document, it can follow the structure, spot the main points and restate them in different words and at different lengths. It can also scan for particular kinds of information, such as names, numbers or requests, even when they are buried in the middle of long paragraphs.

Another advantage is flexibility. A human summariser usually writes one version. With Claude, you can ask for a one-line summary, a one-paragraph summary and a detailed outline of the same document, each aimed at a different reader. You can also ask follow-up questions, such as “What did the report say about costs?”, without re-reading anything yourself.

Claude can only summarise or extract from what it can see, so the first step is putting the material in front of it.

There are usually two ways to share material. The simplest is to paste the text directly into your message. The second is to attach a file, if your Claude app supports it. Many Claude apps let you attach documents, and the types of files and the size limits can differ by app, plan and version. For the current details, the help centre at support.claude.com is the best place to look.

Whichever way you share it, it helps to label the material and say what you want done with it. “Here is the transcript of yesterday’s meeting. Please summarise the key decisions” is much clearer than pasting a wall of text and hoping for the best.

A few small details in your request change the summary a great deal.

flowchart LR
  A["Source material"] --> B["Your instructions: length, audience, focus, format"]
  B --> C["Claude reads and condenses"]
  C --> D["Summary shaped to your need"]
        

There is more than one way to shrink a document, and each suits a different need.

When you want particular facts rather than a general overview, extraction is the right tool.

To extract well, you must tell Claude exactly what to look for. “Pull out the important stuff” is vague. “List every date, deadline and person responsible mentioned in the text” is precise. The clearer the description of what you want, the more reliable the result.

Extraction is particularly useful when the information is scattered across a document. An action item mentioned in paragraph three, a deadline in paragraph nine and an owner’s name in paragraph fourteen can be gathered into one tidy list.

flowchart TD
  A["Long messy document"] --> B["You describe exactly what to look for"]
  B --> C["Claude scans for matching items"]
  C --> D["Action items"]
  C --> E["Dates and deadlines"]
  C --> F["Names and numbers"]
  D --> G["Clean organised result"]
  E --> G
  F --> G
        

The format of an extraction decides how easy it is to use afterwards.

Extracted information can be presented in several ways. A list works well when items are simple and few. A table works well when each item has several attributes, such as a task, an owner and a due date. A labelled list works well when you want groups, such as separating decisions from risks.

If you plan to use the result somewhere else, such as a spreadsheet or a message, tell Claude the layout you need. A table with named columns is often the easiest to copy into other tools.

Here is a simple routine you can follow every time.

The most important quality of a summary or extraction is that it matches what the material really says.

Language models generate text by predicting what fits well. That strength can become a weakness when the source is silent on something. Claude may be tempted to fill a gap with a plausible detail, or to blend what the document says with general knowledge. This is sometimes called hallucination, which means producing information that sounds right but is not actually supported. You can greatly reduce this risk with clear instructions.

flowchart TD
  A["Source says it"] --> B["Include in the summary"]
  C["Source is silent"] --> D{"What does your instruction say?"}
  D -- "Use only the source" --> E["Write not stated"]
  D -- "No instruction" --> F["Risk of a plausible guess"]
        

Long documents bring practical limits, and a few habits help you work around them.

Claude can only take in a limited amount of text at one time. This limit is called the context window, and it works like the size of a working desk. Everything you share, plus the conversation so far, must fit on that desk. Different Claude models and apps have different sizes, and very large documents may exceed what fits.

When material is very long, a useful approach is to work in stages. Summarise each section separately, then combine those section summaries into one overall summary. This is sometimes called a summary of summaries. It keeps each step manageable and lets you check each part along the way.

These tasks appear in almost every kind of work and study.

Documents often contain personal or confidential details, so a little care goes a long way.

Before sharing a document with any AI tool, think about what it contains. Personal data, client information and confidential business details deserve extra thought. If your task does not need real names or identifiers, you can replace them with neutral labels such as “Customer A” before sharing, and still get an equally useful summary.

If you are using Claude through work, follow your organisation’s rules about what may be shared with AI tools. How conversations and files are handled can also differ between products and plans, so it is sensible to read the privacy information that applies to yours.

These errors appear often, and each has an easy fix.

Here is the whole picture in one place.

Summarizing shrinks material into a shorter version that keeps the main ideas. Extracting pulls out specific facts, such as dates, names or action items, and lists them separately.

Paste the text into your message, or attach the file if your Claude app supports it. Supported file types and size limits vary, so check the help centre at support.claude.com for current details.

Whatever suits your purpose. State a target such as one sentence, five bullet points or about a hundred words. If unsure, ask for a layered summary with a headline first and more detail below.

Yes. It can miss a point, over-compress a nuance or occasionally state something the source does not say. Clear instructions reduce this, and spot-checking important facts against the original is a sensible habit.

Tell it to use only the material provided and to write “not stated” when something is missing. You can also ask it to indicate where in the document each key point came from.

A table works well when each item has several details, such as task, owner and deadline. A simple list suits short, similar items, and grouped headings suit items that fall into categories.

Work in sections. Summarise each part with the same instructions, then combine the section summaries into an overall one, and go back to the original for any critical detail.

Yes. Ask for different versions, such as a two-line version for executives and a simple explanation for a new colleague. Just tell Claude who each version is for.

Share only what the task needs, replace real identifiers with neutral labels where possible, follow your organisation’s rules and check the privacy information for your product and plan.

Use it to understand and prepare questions, but not as a replacement for professional advice. For decisions with real consequences, consult a qualified person and read the original.

Summarizing shrinks long material into the main ideas, while extracting pulls out specific facts and presents them in a form you can use. Claude is well suited to both, because it can follow structure, find scattered details and restate them at different lengths for different readers. You get the best results by sharing the source clearly, saying whether you want a summary, an extraction or both, and describing the length, audience, focus and format you need. To keep results faithful, tell Claude to use only the material provided and to mark anything not stated, and spot-check names, numbers and dates against the original. For very long documents, work in sections and combine the pieces. Protect private information by sharing only what is needed, and remember that for serious legal, medical or financial matters, a summary is a starting point and not a substitute for professional advice or the source itself.