Zero-shot and few-shot prompting with Claude

1.Zero-shot: instructions without examples

A zero-shot prompt describes the task and gives Claude no worked examples. Anthropic's documentation does not use the term "zero-shot". Its advice on writing clear instructions is still the advice for this case, because the wording of your instructions is all Claude has to go on. The core rule is that Claude responds well to clear, explicit instructions. If you want more than the minimum, you have to ask for it. Don't count on the model to infer that from a vague request.

The documentation compares two versions of one request. "Create an analytics dashboard" is the weak version. The strong version adds: "Include as many relevant features and interactions as possible. Go beyond the basics to create a fully-featured implementation." The task is the same. The second prompt also says how much work is expected. The docs suggest a test: show your prompt to a colleague who knows little about the task. If they would be confused, Claude will be too. If the order or completeness of the steps matters, write the steps as a numbered or bulleted list.

The documentation prefers the version with a reason: "Your response will be read aloud by a text-to-speech engine, so never use ellipses since the text-to-speech engine will not know how to pronounce them." If you give the motivation behind a constraint, Claude can apply it to related cases the rule never names. A zero-shot prompt carries only the information you write into it, so this reasoning is an important part of it.

Sources1

2.Few-shot: show the output instead of describing it

Some requirements are hard to put into words, such as a house style for summaries, a tone, or a strict layout. For these, Anthropic calls examples one of the most reliable ways to steer Claude's output format, tone and structure. Claude's blog lists when examples are worth adding: the format is easier to show than describe, you need a specific tone or style, the task involves subtle patterns or conventions, or simple instructions haven't produced consistent results.

Here's an example of the summary style I want:

Article: [link to article about AI regulation]
Summary: EU passes comprehensive AI Act targeting high-risk systems. Key provisions include transparency requirements and human oversight mandates. Takes effect 2026.

Now summarize this article in the same style: [link to your new article]

Examples add tokens and can add unintended patterns, so add them one at a time. The blog's advice is to start with one example (one-shot). Add more examples (few-shot) only if the output still does not match what you need. Treat it as a ladder: an instruction first, then one example, then several.

Sources12

3.Picking examples that teach the right pattern

Claude copies what it sees in examples, so a careless example teaches the wrong thing. The blog warns that Claude 4.x and similar models pay very close attention to details in examples. If every example opens the same way or has the same length, Claude may copy that too. The prompting guide gives three properties to aim for. Examples should be relevant, meaning they closely mirror your real use case. They should be diverse, meaning they cover edge cases and vary enough that Claude doesn't pick up unintended patterns. And they should be structured, meaning each is wrapped in <example> tags, with the set inside <examples>, so Claude can tell them apart from instructions.

When outputs go wrong, it is tempting to add another rule for each failure. Anthropic's context-engineering guidance advises against stuffing a laundry list of edge cases into a prompt. It recommends curating a set of diverse, canonical examples that show the expected behaviour. The Sonnet 5 prompting guide makes a related point: positive examples of the behaviour you want tend to work better than negative examples or instructions about what not to do.

Sources123

Exam traps

Each one states something that sounds right. Open it to see what is actually true.

1. 1.The more edge-case rules you write into the prompt, the more reliably Claude will behave.Why is that wrong?

   Anthropic advises against listing every edge case as a rule. It recommends a small, curated set of diverse, canonical examples that show the expected behaviour.

   Covered in Picking examples that teach the right pattern

2. 2.Several near-identical examples are the safest way to lock in a format.Why is that wrong?

   Near-identical examples teach their accidental similarities as well. Examples should vary enough that Claude doesn't pick up patterns you didn't intend.

   Covered in Picking examples that teach the right pattern