Prompt Technique Taxonomy for the CCAR-P Exam

Three techniques, in order of weight

Before you can choose a prompt technique for a task, you need to know what the choices are and what each one provides. The Claude Certified Architect - Professional (CCAR-P) exam sets this up at the remember level with three core techniques: zero-shot, few-shot, and chain-of-thought. They are best understood as a progression of weight. Zero-shot is the lightest, chain-of-thought the heaviest, and few-shot sits between them. Each adds something the previous one lacked, and each costs something the previous one did not. Knowing precisely what each provides is the foundation for choosing among them, which is developed in technique selection by task complexity.

Zero-shot: instructions only

Zero-shot prompting gives the model instructions and nothing else, no worked examples. It is the default technique to try first because it is the cheapest in tokens and the simplest to maintain, and for well-specified tasks the model already handles reliably, it is often all that is needed. Starting zero-shot means you only add weight when the task proves it needs it, rather than paying for scaffolding up front.

Few-shot: showing instead of telling

Few-shot prompting adds a handful of input/output examples to the prompt. Its purpose is to demonstrate a desired format or judgment that is easier to show than to describe. When the shape of the output or the kind of decision you want is hard to capture in instructions alone, a few concrete examples can convey it directly. The examples are content, though, which means they carry their own risks: they must be maintained as the task evolves, and, as bias introduced by prompt construction explains, an unbalanced set can quietly teach the model the wrong norm.

Chain-of-thought: reasoning made explicit

Chain-of-thought prompting instructs the model to reason step by step before producing its final answer. It fits tasks where the reasoning path itself matters to getting the answer right, such as multi-step logic or judgments that depend on several interacting conditions. Making the steps explicit both improves those answers and makes the reasoning checkable.

One distinction the exam cares about: chain-of-thought is an explicit step-by-step reasoning instruction you write into the prompt. It is not the same as the model's internal extended-thinking feature, which is a separate capability that adds a reasoning pass at the model level and is controlled differently, as covered in extended thinking as a cost-latency tradeoff. Confusing the two is a classic error.

What the exam trips candidates on

The two traps target sloppy assumptions about the techniques. The first is assuming few-shot always outperforms zero-shot regardless of task complexity. A scenario will imply more examples are always better; the credited reading knows zero-shot is the default and few-shot earns its place only when the task needs demonstration. The second is confusing chain-of-thought, an explicit step-by-step reasoning instruction, with the model's internal extended-thinking feature. A scenario will blur the two; the credited reading keeps the prompt-level instruction and the model-level feature distinct.

Common misreadings to avoid

How this shows up on the exam

Remember-level questions ask you to name or define the three techniques or to catch the chain-of-thought versus extended-thinking confusion. The reliable answer knows zero-shot as instructions-only and the default, few-shot as a handful of demonstrating examples, and chain-of-thought as a prompt-level step-by-step instruction distinct from the extended-thinking feature.

This taxonomy is the basis for technique selection by task complexity and its applied counterpart matching technique to scenario-based tasks, and it connects to bias introduced by prompt construction, since few-shot example sets are a common source of bias.