Modules/Module 4/Lesson 2
Lesson 2 of 6 ~10 min read

Summarising Long Documents

Lesson 4.2 — Summarising Documents, Articles, and Videos with AI

Stack of documents and a laptop on a desk

There are only so many hours in a day, but the volume of information demanding your attention keeps growing. Reports, research papers, long-form articles, meeting transcripts, YouTube explainers, dense PDFs — the reading queue never empties. AI is one of the most powerful tools available for reclaiming that time. Used well, it can turn a 40-page report into a three-paragraph briefing, or pull the key argument from a one-hour video in under a minute.

This lesson covers five distinct summary styles, how to handle documents too long for a single paste, and how to make sure the summary you get is actually accurate.


What AI Summarisation Actually Does

When you paste text and ask an AI to summarise it, the model reads the entire input and generates a compressed version based on what it identifies as most important. It is not just cutting sentences — it is interpreting meaning, identifying themes, and rewriting in a new form.

This is powerful. It is also where the risk lies. The AI decides what is important based on patterns in its training, which may not match your specific priorities. That is why your prompt matters enormously — telling the AI what you care about produces a summary tuned to your needs, not a generic one.

Key takeaway: A generic "summarise this" prompt gives you a generic summary. A specific prompt — "summarise this for a non-technical manager, focusing on the financial implications" — gives you something genuinely useful.


Five Summary Styles with Prompts

Style 1: The Executive Summary

Best for: Reports, proposals, business documents

Prompt:

"Summarise the following document as an executive summary. The audience is a senior manager with limited time. Cover: the main problem or opportunity, key findings or recommendations, and any action required. Maximum 200 words. Use plain English — no jargon."

Output looks like: Three to four short paragraphs. Clear and scannable.


Style 2: The Bullet-Point Brief

Best for: Articles, blog posts, news stories

Prompt:

"Read the following article and extract the five most important points as bullet points. For each point, write one sentence of explanation. Keep each bullet under 25 words."

Output looks like:

  • Point one: brief explanation
  • Point two: brief explanation

This style is ideal for morning news reading or keeping up with industry publications.


Style 3: The Critical Summary

Best for: Academic papers, research, opinion pieces

Prompt:

"Summarise this text, but also briefly note: (1) the main argument, (2) the evidence used to support it, and (3) any significant gaps, counterarguments, or limitations the author acknowledges. Use three clearly labelled sections."

This forces the AI to engage analytically rather than just compress, which is crucial when you need to evaluate whether to trust the source.


Style 4: The Action-Item Extract

Best for: Meeting notes, project documents, email threads

Prompt:

"Read this meeting transcript / document and extract: (1) all decisions made, (2) all action items with owner names if mentioned, and (3) any unresolved questions. Format as three labelled lists."

Why it works: Most meeting notes are narrative. This prompt converts narrative into structure, which is what people actually need.


Style 5: The Layperson Translation

Best for: Technical papers, legal documents, medical reports

Prompt:

"Summarise this document as if explaining it to an intelligent person with no specialist knowledge in this area. Avoid technical terms where possible. If you must use a technical term, briefly define it in brackets. Aim for 150–250 words."

This is particularly valuable for medical results, legal contracts, or financial statements — documents people often receive but struggle to interpret.


Summarising YouTube Videos

YouTube is one of the most underrated knowledge sources available, but watching a full video when you just need the core idea is inefficient. Here is how to extract summaries:

Method 1 — Copy the transcript:

  1. Open the video on YouTube
  2. Click the three dots below the video → "Show transcript"
  3. Copy the transcript text
  4. Paste into your AI tool with a summary prompt

Method 2 — Use a tool with web browsing: If your AI tool supports web browsing (ChatGPT with browsing, Perplexity, Grok, or Claude with web access), paste the YouTube URL and use:

"Summarise this YouTube video. What is the main argument? What are the three most useful takeaways? Who is the intended audience?"

Method 3 — Use a dedicated tool: Services like Summarize.tech, Eightify, or the YouTube Summary Chrome extension are built specifically for this task and often produce cleaner results for very long videos.


Handling Long Documents

Most AI tools have a context limit — they can only process a certain amount of text at once. A 50-page PDF will exceed that limit. Here are strategies for handling long documents:

StrategyHow to use itBest for
ChunkingSplit the document into sections and summarise each separately, then summarise the summariesReports with clear sections
Extract and pasteCopy only the sections most relevant to your questionResearch papers with abstract/conclusion
Use a tool with large contextClaude (200K tokens), Gemini 1.5 Pro (1M tokens)Very long single documents
PDF-native toolsAdobe AI Assistant, ChatPDF, AskYourPDFPDFs specifically
Ask for the key section"Which section of this document is most relevant to X?" before summarisingWhen you're not sure where to look

Key takeaway: For most long documents, you do not need a summary of everything — you need a summary of the parts relevant to your question. Being specific about what you need dramatically reduces the impact of context limits.


Accuracy: The Critical Step Most People Skip

AI summaries can contain errors. The model may confidently misstate a figure, omit a crucial caveat, or paraphrase in a way that subtly changes meaning. This is not malicious — it is a limitation of how language models work.

The five accuracy rules:

  1. Never trust a summary of something you haven't seen at all. If the stakes are high, read at least the original introduction and conclusion yourself.

  2. Check all numbers and statistics. These are the most common error points. Go back to the source for any specific figure the summary mentions.

  3. Watch for missing nuance. A summary of a nuanced argument can make it sound more certain than the original. The phrase "the evidence suggests" often gets dropped.

  4. Ask the AI to cite its sources within the document. Prompt: "For each key point in your summary, tell me which section or paragraph of the original document it comes from." This forces accountability.

  5. Ask for uncertainty flags. Prompt: "If any part of this document was ambiguous or if you were uncertain about its meaning, please note that in your summary."


A Practical Workflow

Here is a real workflow for a professional who receives a 30-page industry report every month:

  1. Paste the executive summary section into AI with Style 1 prompt → 2-minute read
  2. Paste the recommendations section with the action-item extract prompt → clear list of what the report is calling for
  3. Ask a follow-up question: "Does the report say anything specific about [your industry / your company's situation]?"
  4. Flag anything relevant for a deeper read

Total time: 8–10 minutes instead of 90. And you will often get more out of it because you are reading actively with specific questions in mind.


Practice Task

Find a long article or report you have been meaning to read but haven't. Paste it into your AI tool of choice and try at least two different summary styles from this lesson. Notice how different prompts surface different aspects of the same document. Which style was most useful for your purposes?