AI for Everyday Work: Writing, Research, Meetings & Data · Writing and summarizing with AI · lesson 9 of 17 · 12 min

Summarizing anything: reports, threads, articles and videos

Summaries that are actually useful

"Summarize this" usually produces a bland paragraph. A good summary is shaped by who needs it and what they'll do with it. The same 30-page report might need a two-line summary for a CEO, a list of actions for a project team, and a risk list for finance.

The summary template

Summarize the text below for [audience] who needs to [decide / act / stay informed].
Give me:
1. The main point in one sentence
2. [3-5] key points
3. Actions or decisions needed (with owners and dates if mentioned)
4. Anything unclear, missing or worth questioning
Keep it under [150] words. Use only the text provided.
[paste text or attach file]

The fourth item is where AI adds surprising value: it often spots gaps or contradictions you'd miss when skimming.

Different summary shapes

TL;DR: "Give me a two-sentence summary."
Decision summary: "What decision is being asked for, what are the options, and what does the author recommend?"
Action list: "List every action, who owns it and the deadline."
Compare: "Summarize the differences between version 1 and version 2."
Explain simply: "Summarize this for someone with no background in finance."

Long documents and multiple files

Many assistants can handle long documents and several files at once; limits vary by tool and plan. For very long material:

Ask for the structure first ("What are the main sections?").
Summarize the sections that matter to you.
Ask targeted questions: "What does it say about pricing? Quote the relevant sentences."

Asking for quotes makes checking easy: you can search the document for the quoted sentence.

A reading-list shortcut

When you have several articles to get through, paste them together (or attach them) and ask: "For each article, give me the main point in one sentence and one thing I should act on. Then tell me which single article is most worth reading in full." You still read the most important one yourself, but you spend your attention where it matters.

Articles, web pages and videos

Some assistants can read web pages or video transcripts directly, depending on the tool and plan. If yours can't, paste the text or the transcript. Always keep the original link so you can refer back.

Worked example

Sana, a marketing executive in Lahore, receives a 25-page industry report the evening before a strategy meeting. She uploads it to her work assistant and asks for a decision-focused summary for her team: key trends, implications for their brand, and questions to raise. The summary highlights three trends and notes that one chart's data source is unclear. In the meeting, she raises that question, and her manager asks her to follow up with the publisher. She looks prepared and thoughtful, and she read the key sections herself after the summary pointed her to them.

Sample output check

A strong summary: stays within the word limit, names specific numbers from the text (not rounded or invented), lists actions with owners where the text states them, and flags uncertainty honestly ("The report doesn't say which markets the survey covered"). If you spot a figure in the summary, search the document to confirm it before repeating it.

Summaries for different people from one source

One document often needs several summaries. After your first summary, try: "Now give me a version for the finance team focusing on costs and risks" and "Now a two-line version for the weekly team update." Each version is quick to produce, but check that numbers stay identical across all of them.

Hands-on: one source, three summaries

Take one long document you actually need (a report, a policy, a long thread) and run all three prompts. Compare them.

SUMMARY A - Decision summary (for a manager)
Summarize the attached for a manager who must decide [what] by [when].
1. The decision being asked for, in one sentence
2. The options and the author's recommendation
3. Costs, risks and deadlines mentioned (quote the numbers exactly)
4. What the document does NOT say that a decision-maker would want to know
Under 150 words. Use only the document. Quote the key sentence for each point.

SUMMARY B - Action list (for the team)
List every action in the attached: action (verb first) | owner | deadline | page/section.
If an owner or deadline isn't stated, write "Not stated". Don't infer.

SUMMARY C - Plain-English explainer (for a newcomer)
Explain the attached to someone new to the topic in 5 bullets, no jargon.
Then list 3 terms they'll need to know, with one-line definitions from the document.

Before: "Summarize this" produces a paragraph that restates the introduction.

After: a decision summary your manager can act on, a checkable action table, and an explainer for the new hire, all consistent with each other. Check that the numbers match across all three versions.

Summarizing across sources

Deep research features and source-grounded notebooks can summarize many files at once. A Gemini Notebook (formerly NotebookLM) answers only from the sources you add and links each answer to the passage; Claude and ChatGPT projects hold reference files you can question together; Microsoft 365 Copilot can summarize files and threads you have access to. For multi-source summaries, always ask "which source says this?" for each key point.

Video and audio

Many assistants can work from a video's transcript or an uploaded recording, depending on tool and plan. Ask for timestamps ("give the time for each key point") so you can jump to the moment and check.

Measure it

A good summary lets the reader act without opening the original, and lets you verify each key point in under a minute. If you can't find where a claim came from, the summary isn't finished.

Pitfalls

Trusting a summary of something important without reading key parts yourself.
Summaries that sound confident but include details not in the original. Ask: "Use only the text provided."
Uploading confidential documents to unapproved tools.
Losing nuance: summaries simplify. For contracts, policies and anything legal, read the original.

Video lecture: Summarizing anything: reports, threads, articles and videos

Lecture coming soon · 11 chapters · about 8 minutes. Read the full transcript below.

Summaries people can act on
Why 'summarize this' disappoints
The map analogy
The four-part summary
Example 1: the reading list
Example 2: meeting prep in Lahore
Watch me do it
Many sources, video and audio
Summaries for different people
Common mistakes
Recap and try this now

Lecture transcript

Summaries people can act on

Here's a familiar moment. It's nine at night, the meeting is at nine tomorrow, and there's a thirty page report you were supposed to read. You paste it into an AI and type summarize this. You get a tidy paragraph. And tomorrow, someone asks about the numbers on page eighteen, and you have no idea. In this lecture you'll learn to get summaries that are shaped for the person who needs them and easy to check. You'll see a simple example, a realistic meeting prep scenario, watch me produce three summaries from one document, and leave with templates you can reuse tonight.

Why 'summarize this' disappoints

Why does summarize this disappoint? Because a summary isn't a neutral thing. The same report needs a two line version for a CEO, an action list for the project team, and a risk list for finance. When you don't say who it's for, the AI picks the safest option, which is usually a restatement of the introduction. And the details you'll actually be asked about, the numbers, the deadlines, the caveats, often don't make the cut. The fix is simple. Tell the AI who needs the summary, what they'll do with it, and how you'll check it.

The map analogy

Think of summaries like maps. One city, many maps. A tourist wants landmarks. A delivery driver wants routes and one way streets. A city planner wants zones and pipes. Nobody would say one map is the correct summary of the city. It depends on the journey. So when you ask for a summary, name the journey. For a manager who must decide, you want the decision, the options, the costs and risks, and what's missing. For a team, you want actions, owners and deadlines. For a newcomer, you want plain language and the key terms. Same document, different maps.

The four-part summary

Here's a template that works for most documents. Ask for four things. The main point in one sentence. Three to five key points. Actions or decisions needed, with owners and dates if mentioned. And anything unclear, missing or worth questioning. Then add two rules: use only the text provided, and quote the key sentence for each point. That fourth item is where AI adds surprising value. It will often spot a chart with no data source, a deadline that contradicts another page, or a recommendation with no cost attached. That's exactly the kind of thing that makes you look sharp in a meeting.

Example 1: the reading list

A simple example first. You've got five industry articles saved and no time. Paste or attach them together and ask: for each article, give me the main point in one sentence and one thing I should act on. Then tell me which single article is most worth reading in full. You get a quick overview, and more importantly, a recommendation for where to spend your attention. You still read that one article properly. The AI didn't replace your reading. It helped you choose it.

Example 2: meeting prep in Lahore

Now a realistic scenario. Sana is a marketing executive in Lahore. The evening before a strategy meeting, she gets a twenty five page industry report. She uploads it to her work assistant and asks for a decision focused summary for her team: the key trends, what they mean for their brand, and questions to raise. The summary highlights three trends, and in the unclear or missing section it notes that one important chart has no stated data source. In the meeting, Sana raises that question. Her manager asks her to follow up with the publisher. She looks prepared and thoughtful. And she did read the key sections herself, because the summary told her exactly which ones mattered.

Watch me do it

Let me show you one source, three summaries. I've attached a proposal for a new customer support system. Summary A, the decision summary. The decision being asked for in one sentence, the options, the costs and risks with exact numbers, and what the document doesn't say. Under one hundred and fifty words, quote the key sentence for each point. It says the proposal doesn't mention training time. Useful. Summary B, the action list. Action, owner, deadline, section, and not stated where it isn't. Two actions have no owner. Also useful. Summary C, a plain English explainer for a new hire, with three key terms. Now the final check. I compare the numbers across all three. They match. If they didn't, I'd go back to the document before sharing anything.

Many sources, video and audio

What about many sources at once? Tools built for this, like Gemini Notebook, which used to be called NotebookLM, answer only from the sources you add and link each answer to the passage it came from. Projects in Claude and ChatGPT can hold several reference files you question together, and Microsoft 365 Copilot can summarize files and threads you already have access to. The rule for multi source summaries is: for each key point, ask which source says this. And for videos or recordings, ask for timestamps, so you can jump straight to the moment and check the exact words.

Summaries for different people

One more technique that saves a lot of time. Once you've got a good first summary, you can produce versions for different people in seconds. Try: now give me a version for the finance team, focusing on costs and risks. Or: now a two line version for the weekly team update. Each one is quick. But here's the discipline. Check that every number stays identical across every version. If the budget is two hundred and forty thousand in one, it can't be about a quarter of a million in another, because someone will put them side by side. And keep the original link or file name with every version, so anyone can go back to the source.

Common mistakes

Common mistakes. Trusting a summary of something important without reading the key parts yourself. For contracts, policies and anything legal, read the original. The summary is a map, not the territory. Summaries that include details not in the original. That's why use only the text provided matters. Numbers that drift. A figure that's twelve point four in one version becomes about twelve in another. Always check numbers are identical across versions. And uploading confidential documents into unapproved tools, which is still the most common data mistake people make with summaries.

Recap and try this now

Let's recap. Shape every summary around who needs it and what they'll do with it. Use the four part template: main point, key points, actions, and what's unclear or missing. Ask for quotes, page numbers or timestamps so you can check quickly. And for multiple sources, ask which source says each thing. Here's your try this now. Take one long document you genuinely need this week and run the three summary prompts from the lesson text: decision, actions, plain English. Then pick one key number and find it in the original. If that takes you under a minute, your summary is doing its job.

Key takeaways

Shape every summary around who needs it and what they'll do with it.
Ask for quotes, page numbers or timestamps so each key point can be checked quickly.
Ask what the document does not say; gaps and contradictions are where AI adds surprising value.
For multi-source summaries, ask which source supports each point; for contracts and policies, read the original.

Try it

Summarize one long document from your work with the template. Verify two points by searching for the quoted text in the original.