Study notes · 3.5% of the exam

Adapt prompting strategies based on task type (analysis, research, drafting, brainstorming)

Change the prompting strategy to fit the task type: analysis needs data and criteria, research needs sources and dates, drafting needs audience and tone, brainstorming needs quantity first and filtering second.

Key points

  1. 1

    Analysis: supply the data, state the exact questions and criteria ("rank causes by frequency and total downtime; flag any that grew quarter over quarter"), and ask Claude to show its working and state assumptions. Never ask it to fill missing data.

  2. 2

    Research: use web search or Research so facts are current, ask for the source URL and retrieval date for each claim, and ask Claude to separate what the sources say from its own inference. Training knowledge alone is stale and unsourced.

  3. 3

    Drafting: name the audience, tone, length, key facts and call to action, and paste an example of the voice to match. A drafting prompt that reuses analysis wording ("analyse thoroughly") produces analysis, not a letter.

  4. 4

    Brainstorming: ask for many diverse options ("20 ideas, one line each, no evaluation") and filter in a second prompt against agreed criteria. Asking for "the best idea" first narrows the search prematurely.

  5. 5

    Research mode is built for comprehensive information gathering that needs several searches and synthesis with citations; extended thinking is for harder reasoning without external sources; plain web search is for a quick current fact.

  6. 6

    Mixed requests (research, then options, then a recommendation) should be split into a sequence with the right strategy at each stage, in evidence-first order: research, then ideation, then drafting from the verified findings.

  7. 7

    Never ask for citations after the conclusions are written; post-hoc sourcing invites Claude to find or invent references that fit. Gather evidence first.

  8. 8

    Persuasive tone belongs to drafting, not analysis; unfiltered ideation belongs to brainstorming, not research; sources and dates belong to research, not idea generation.

  9. 9

    Signals of a mismatched strategy: an analysis with no numbers or criteria, a research answer with no sources or dates, a letter that reads like a report, a brainstorm with only two options.

  10. 10

    Adapt structure as well as content: tables for comparisons, numbered steps for procedures, short paragraphs for executive readers, bullet lists for option sets.

Test yourself on Adapt prompting strategies based on task type (analysis, research, drafting, brainstorming)

Ten questions, with the answer and explanation after each one.