Units
1. Introduction: Personal Productivity with AI — Roles, Boundaries, Authentication and Privacy 2. Email Management: Draft, Reply, Summary and Tone 3. Calendar, Time Management and Task Planning 4. Note Taking and Personal Information Management 5. Research, Reading and Summarizing 6. Writing and Document Production: Report, Presentation, Proposal 7. Meeting: Preparation, Notes, Summary and Action Follow-up 8. No-Code Automation and Workflows 9. Decision Making, Priority and Focus 10. Privacy, Data Security and Verification Habit 11. Establishing an End-to-End Personal Workflow System
Unit 5 / 11

Research, Reading and Summarizing

Gains:

  • Ability to conduct factual research in source-based mode, which is based on the text given by artificial intelligence rather than its memory
  • Ability to quickly evaluate long texts by spending your time according to priority with layered summarization (single sentence, article, section, question-answer)
  • Ability to detect fake sources that artificial intelligence can fabricate by opening and verifying every citation, statistic and quote at its original source

One of the knowledge worker's biggest time-wasters is reading: reports, articles, documents, long email chains, draft contracts. AI radically speeds up your reading and research workflow — summarizing long texts, simplifying complex topics, comparing multiple sources, finding your answer right from within the text. But the most critical lesson of this unit is this: AI is not a search engine and can fabricate sources. Summarization is reliable; The facts he presents as "general knowledge" are dangerous. Therefore, in research, it is much safer to run the AI ​​on the text you give it than to say "you research and tell me".

Terms. Summarizing is preserving the essence of a text by making it shorter. Distillation is simply extracting the most critical idea, beyond the summary. Grounded operation is when the AI ​​responds based on the text you provide — it's the safest mode. Parametric information is “memory” information from the AI’s training that it cannot source — the riskiest mode. Citation indicates the source on which a claim is based.

Two modes: source-based vs. from memory

This distinction is the backbone of using AI in research.

Source-based mode: You give a text (report, article, document), AI summarizes and answers questions based only on that text. The risk is low because you can compare the output with the source. Prefer this mode.

From memory mode: Like "Tell me about topic X", "What are the stats of Y", you want the AI ​​to answer from training memory. The risk is high — AI may not cite sources, may make up dates/figures, may be out of date. Use this mode only for the "idea/startup", not the fact.

Tip: If you need a fact, figure or quote, ask the AI ​​for the source and open and verify that source yourself. Even if AI says "I did research", do not use it without confirming that the link it shows actually contains that information. AI can make up non-existent articles and links.

Step by step: researching a document with AI

  1. Give the document. Paste the text or upload the file (in a supporting tool).
  2. State the purpose. What are you looking for? General summary or answer to a specific question?
  3. Connect it to the source. Set the restriction "Only rely on this document, do not add external information".
  4. Layered summary. First a 1-sentence summary, then a 5-item summary, then a chapter-based summary.
  5. Inquire. "Does this document answer the question? In which section is the answer?"
  6. Verify. Compare critical outputs to the relevant location in the document.

Layered summarization technique

It is much more useful to go layered rather than a single "summarize" command. First, 1 sentence ("What is this document about?") helps you decide whether you should read it or not. Then 5 items give the outline. If necessary, go deeper, chapter by chapter. This pyramid structure allows you to decide wisely where to spend your time — a 1-sentence summary of “do I need this?” before reading the 40-page report. You answer the question.

layer

length

What does it do?

Title summary

1 sentence

Is it worth reading?

Executive summary

5 items

Grasping the outline

Episode summary

2-3 items per chapter

drill down

Question and answer

Variable

Pull specific information

Four copyable research templates

Layered summary of the document below. Rely ONLY on this document:1. One-sentence summary.2. 5-item executive summary.3. The 3 most important numerical data/findings (as stated in the document, fabricated). Document: """[paste]"""

Answer my question based on the following document:Question: [specific problem].- If the answer is in the document, indicate in which section/paragraph it is.- If it is not in the document, say "this document does not answer this question", do not guess.Document: """[paste]"""

Compare the following two sources: - Common points, contradictions, topics skipped by each. - Present in a table. - State which source is more detailed on which subject. Source A: """[text]"""Source B: """[text]"""

Simplify the following technical text [at high school level / executive level]:- Explain jargon or use plain equivalents.- Do not distort the meaning, do not falsify when simplifying.- Mark where you do not understand/are unclear.Text: """[paste]"""

Weak prompt / Strong prompt

Weak: “Tell me about AI regulations.” (From memory mode — AI may return unsourced, partially made-up text that may be outdated.)

Güçlü: "Based on this regulation text dated 2025 that I pasted to you: what obligations are introduced, who it covers, what is the effective date? Just rely on this text, do not add anything that is not in the text; if it is unclear, specify." The powerful version connects the AI ​​to the source, closing the field of fabrication and making the output verifiable.

three mini cases

Case 1 — Melting the reading pile. One investment analyst was scanning 12 different industry reports every morning, ~2.5 hours. He first summarized each report with 1 sentence + 5 items and started reading only the relevant 3-4 of them in full. Scanning time decreased to 50 minutes, and more time was available for important reports.

Case 2 — Fake welding disaster. A student used 5 academic citations provided by AI in his assignment without verifying them. When the consultant checked, it turned out that 3 of the references did not exist — the AI ​​had made up realistic-looking but fake bylines. The assignment was deemed invalid. Lesson: open and verify every attribution at source.

Case 3 — The power of simplification. One engineer had AI say "simplify at the executive level, but don't distort the legal meaning" of the 15-page contract language sent by the legal team; Then, making sure that he understood the simplified version, he had the critical items confirmed by the lawyer again. Thus, he understood quickly and was protected from misinterpretation — balancing simplification with verification.

Common mistakes

  • Relying on memory-to-mode for fact: Asking for unsourced statistics and history.
  • Not verifying attributions: AI fabricates sources; Open and confirm each tag.
  • Asking for a full summary in one go: Go in layers, spend your time wisely.
  • Not seeing the loss of meaning in simplification: Simplification sometimes makes things wrong; If critical, confirm.
  • Not accepting “not in the document”: making the AI ​​fit the gap; "No" is also a valid answer.
  • Not reviewing the comparison: When comparing two sources, AI may skew one side.
Caution: Just because the AI ​​says "I looked it up on the internet" does not mean the information is correct. Even search tools can present the wrong source as correct. In every critical claim, see for yourself the end of the chain — the real source.

In summary

  • There are two modes of research: source-based (safe, based on your text) and memory (risky, unsourced). Use the first one for the case.
  • Layered summary: 1 sentence → 5 articles → section → question-answer; Decide wisely where to spend your time.
  • Put the constraint on AI to "rely on this document only, otherwise say 'no'"; Don't make up the gap.
  • Open and verify every citation, statistic, and quote at its original source — AI can generate fake bylines.
  • Simplification speeds up but can distort meaning; Confirm in critical texts.

Application task

Take a long document (report, article) in your hand. Make a layered summary with the first template, then ask it a specific question with the second template and verify in the document which section the answer is in. Ask the AI ​​for a sourced statistic as well, and actually open that source and check if the information is there — note the result.

checklist

  • [ ] I did the factual research in source-based mode.
  • [ ] I used layered summarization, spending my time by priority.
  • [ ] I gave AI the constraint "if it's not in the document, make it up, say it doesn't exist".
  • [ ] I have verified every reference and statistic at its original source.
  • [ ] I have verified the simplified critical texts.
  • [ ] I checked the benchmark outputs with both sources.
Previous Next