Prompting Strategy by Task Type: Analysis, Research, Drafting, Brainstorming.

Anthropic's core prompting guidance is task-agnostic at its base: be clear and direct, give Claude context the way you would a brilliant but brand-new employee with no knowledge of your norms, and iterate against a defined success criterion. That base principle does not change from task to task. What changes is how a business user applies it once the type of work is known, and the exam objective (CCAOF-D1-O4) itself names four task types to reason about: analysis, research, drafting, and brainstorming. An analysis task benefits from the response being structured and treated with some skepticism before it's trusted; a research task benefits from claims scoped to sources and hedged where evidence is thin; a drafting task benefits from audience and tone specified up front; a brainstorming task benefits from room to diverge rather than early constraints.

The Associate exam objective (CCAOF-D1-O4) tests whether a candidate recognizes these as genuinely different prompting shapes, not variations that a single generalized template can absorb. This four-shape breakdown is a practical mental model for applying Anthropic's general prompting principles to each named task type, not an official taxonomy Anthropic itself publishes. Anthropic's own "6 Techniques for Effective Prompt Engineering" reference is useful supporting evidence for the underlying idea that technique should match the task: its "break complex tasks into steps" technique happens to be illustrated with an analysis example ("Analyze this quarterly sales data... 1. Identify top performers 2. Compare to prior quarter..."), while its "show examples of good" and "define the AI's role" techniques map naturally onto drafting and brainstorming instead, but Anthropic does not present these as a formal four-category system.

Analysis tasks: structure before conclusions. The instinct to avoid is accepting one confident synthesis as the whole answer. Anthropic's own worked example breaks an analysis prompt into explicit steps, identify, then compare, rather than a single open-ended "tell me what you find." Asking Claude to categorize or structure the evidence first, and only then draw conclusions, surfaces weak or thin evidence that a single-pass synthesis would quietly smooth over.

Research tasks: source-scoped and hedged. Anthropic's own guidance on prompting for business use notes that good prompting reduces, but does not eliminate, hallucination risk. A research prompt should explicitly ask Claude to distinguish claims from the evidence backing them and to flag uncertainty rather than default to a single confident narrative. This is a different failure mode than an analysis task: the risk isn't missed structure, it's an unhedged claim presented with the same confidence as a well-sourced one.

Drafting and brainstorming tasks: audience-first versus constraint-light. These sit at opposite ends of the same spectrum. Drafting benefits from the "show examples of good" and "define Claude's role" techniques, an explicit audience, tone, and a reference example of the target register, per the "be clear and direct" rule: if a colleague with minimal context would misread a vague instruction, so will Claude. Brainstorming needs the opposite treatment: light or no constraints, and an explicit instruction to generate many divergent options, since over-specifying the format or the answer early collapses the breadth the exercise exists to produce.

The failure mode is template reuse, not bad prompting. A well-crafted prompt built for one task type, structured steps for analysis, say, does not transfer cleanly to brainstorming, where that same structure suppresses the divergence the task needs. Recognizing which of the four shapes a request needs is itself the skill this objective tests, applying the underlying "clear and direct" principle is assumed baseline competence.