You're prompting GPT-5.5 for English-to-Spanish translation. The zero-shot version works fine. When would adding few-shot examples reliably outperform the zero-shot prompt?
Zero-shot prompting just describes the task: translate this sentence from English to Spanish. Few-shot prompting adds a small number of input-output example pairs before the actual request. The intuition used to be that more examples are always better, but that is no longer true with modern frontier models.
For common tasks like translation, summarization, or sentiment classification, models like Claude Opus 4.7 and GPT-5.5 already know how to do the task. They were trained on enormous amounts of multilingual text. Showing them three more translations does not teach them anything new about Spanish.
Where few-shot still wins is when you need something specific that the model cannot guess. A particular output format like JSON with named fields, a niche style like legal phrasing, or a domain pattern like a company-specific summary structure. The examples lock in the shape, not the knowledge.
The trap is adding examples by reflex. They cost tokens, add latency, and if the examples are noisy, they can actively make the output worse.
Few-shot examples teach format and style, not knowledge. That is the right frame for choosing between zero-shot and few-shot in 2026.
For common tasks on capable models, zero-shot is already strong. Translation, summarization, classification, basic reasoning: Claude Opus 4.7 and GPT-5.5 hit ceiling-near accuracy from a clean instruction alone. Adding examples may help by a point or two, or may hurt if the examples are inconsistent. The cost is more input tokens and slightly higher latency.
Few-shot reliably wins in three cases. First, format adherence: you need a specific output schema, layout, or structured response that the zero-shot prompt occasionally violates. Second, niche tasks: a domain-specific style, an unusual labeling scheme, a company-internal terminology the model has not seen. Third, structured outputs that have to match a downstream parser exactly.
The trap is treating few-shot as a generic accuracy upgrade. It is not. Bad or noisy examples actively mislead the model, and good examples cost tokens. The senior move is to A/B test zero-shot vs few-shot on a golden set rather than assume the few-shot version wins.
Quick reference
TL;DR
Few-shot wins when you need a specific format, style, or niche pattern the zero-shot prompt does not reliably produce, not for raw accuracy on common tasks.
Imagine you ask a fluent bilingual friend to translate a sentence. They will do it correctly without any examples because they already know both languages. But if you want every translation written in a very specific layout, like always putting the original sentence above the translation, you have to show them what you mean once or twice. The examples are not teaching them Spanish. They are teaching them YOUR format. Few-shot examples in prompts work the same way. With strong modern models, the language knowledge is already there. Examples are most useful when you need to lock in a structure or a niche style the model would not guess on its own.
Everything you need to truly understand this topic: intuition, mechanics, step by step explanation, code, formulas, and worked example. Click to expand.
Everything you need to truly understand this topic: intuition, mechanics, step by step explanation, code, formulas, and worked example.
When the task is common and the model already handles it in zero-shot Examples add tokens and latency without moving accuracy on tasks the model already knows.
When you do not have clean, consistent examples to show Noisy or inconsistent few-shot examples actively mislead the model and can flip metrics negative.
When context window is tight and the task input is already large Few-shot examples consume tokens; if the budget is already strained, the cost may exceed the lift.
When a single structured output schema directive can do what examples would Modern APIs support JSON schemas natively; the directive is cheaper and more reliable than demonstrating format via examples.
Quick scan · 2–4 min
Everything important, quickly.
3 min: format vs knowledge framing + when zero-shot wins on frontier models + three cases where few-shot still wins + token-cost and noisy-example traps.
Side-by-side
Concern
Zero-shot
Few-shot
Token cost
Lower
Higher; examples eat input budget
Best for
Common tasks, strong models, fast iteration
Niche tasks, strict format, hard-to-describe style
Failure mode
Format drift on edge cases
Noisy examples mislead the model
Maintenance
Just an instruction
Curated examples to keep current
2026 default
Yes on Claude Opus 4.7, GPT-5.5, Gemini 3.1 Pro
Reserve for format and niche style
Real-world examples
Real products, models, and research that use this idea.
1OpenAI's structured outputs feature uses a single schema directive instead of few-shot for JSON adherence, because GPT-5.5 follows the schema reliably zero-shot.
2Anthropic Workbench defaults to zero-shot for most prompt drafts, with few-shot reserved for style-mimicking tasks like brand voice or legal phrasing.
3Cursor uses few-shot examples in its code-edit prompts because the desired diff format is highly specific and not what a model guesses by default.
Interviewer follow-ups
What an interviewer would ask next. Try answering before peeking at the approach.
QHow would you pick few-shot examples that actually help?
Sign in or go Pro to see the approach for this follow-up.
QWhen can few-shot examples actively hurt accuracy?
Sign in or go Pro to see the approach for this follow-up.
QDoes chain-of-thought count as few-shot or zero-shot?
Sign in or go Pro to see the approach for this follow-up.
Don't say this
Red flags and common mistakes that signal junior thinking. Click to expand.
Most common mistake
Adding few-shot examples by reflex on a task the model already does well, which adds latency and tokens without measurably moving accuracy.