Comparison questionMedium~7 minPrompting & Structured Output / 5 of 20

Interview question

When would you use zero-shot versus few-shot prompting?

Uses examples to clarify ambiguous task behavior while accounting for token cost, bias, maintenance, and evaluation evidence.

1Say itAnswer out loud first

2CompareCompare with Core answer

3RefineRefine with Stronger and Advanced

TL;DR

Zero-shot prompting gives the task and constraints without examples. Few-shot prompting adds representative input-output examples. I start zero-shot when the task is simple and well specified, then add a small set of examples when they improve ambiguous classification, tone, edge-case handling, or output conventions in evaluation.

Full answer below

Practice focus

Whether examples are chosen for a concrete behavioral need instead of added as a ritual.

Answer first

Answer path

Core answer first, then checkpoints toward a stronger answer.

Start
Core answer
Start here. This is the direct spoken answer to practice first.
Overview
Few-shot examples are useful when they teach a boundary the instruction alone does not communicate reliably.
Zero-shot prompting gives the task and constraints without examples. Few-shot prompting adds representative input-output examples. I start zero-shot when the task is simple and well specified, then add a small set of examples when they improve ambiguous classification, tone, edge-case handling, or output conventions in evaluation.
Checkpoint
Stronger answer
Optional upgrade with examples, trade-offs, and sharper phrasing.
The examples need to cover meaningful distinctions, not repeat the easiest case. For a ticket router, I might include a request that looks like billing but is actually account access, plus an explicit unknown case. I keep examples consistent with the schema and avoid leaking real customer data. More examples consume context and can bias the model toward their wording or class distribution.
Finish
Advanced answer
Optional deeper angle for production judgment and nuance.
I treat the example set as versioned behavior and test it for coverage, imbalance, stale policy, and unintended copying. If a prompt needs dozens of examples to remain stable, that is a signal to revisit the design. I may narrow the task or compare retrieval, fine-tuning, and a deterministic classifier against the same evaluation set. The decision is based on measured performance, not a belief that few-shot prompting is always more capable.

Review signals

Use these to catch weak phrasing, shallow reasoning, and risky claims.

Common mistakesCandidateCommon mistakesMistakes that weaken the answer or signal shallow understanding.Open details

Details

Adding examples without measuring whether they improve the task.
Using only ideal happy-path examples.
Including sensitive production inputs verbatim.
Assuming more examples always improve generalization.

Red flagsCandidateRed flagsSignals that an interviewer may treat as risky.Open details

Details

Examples silently redefine business policy.
The few-shot set is not versioned with the prompt.
The answer ignores token cost and class bias.

Follow-up prompts

Quick interviewer prompts inside this drill.

Follow-up questionsCandidateFollow-up questionsLikely next questions after the main answer.Open details

Details

Treat these as the next interviewer turn. Answer one out loud, then save the gap in your practice notes.

How would you select representative examples?
What would make you remove an example?
When would you choose a different approach entirely?

Continue the section

Next in Prompting & Structured Output: Decompose a model task

Step 6 of 20

When should you split one large prompt into multiple steps?

Practice route

Read core answer Review checklist Follow-up prompts Track progress

Finished this drill?

Practice freely. Sign in only when you want to save completion, confidence, notes, or bookmarks.

New to Aporeon? Create account

Drill details

Human editorial direction

Aporeon is shaped by Aleksandar Tomovski, a software developer with experience on both sides of technical interviews. Content is reviewed for accuracy, natural spoken delivery, useful depth, and honest trade-offs.

About the creator Editorial standards

Was this drill useful?

Help improve the interview library.

Found this answer useful?

Say thanks with a standalone, one-time $5 contribution. No account required.

No subscription or perks.Choose another amount

Opening drill

Comparison questionMedium~7 minPrompting & Structured Output / 5 of 20

Interview question

When would you use zero-shot versus few-shot prompting?

Uses examples to clarify ambiguous task behavior while accounting for token cost, bias, maintenance, and evaluation evidence.

1Say itAnswer out loud first

2CompareCompare with Core answer

3RefineRefine with Stronger and Advanced

TL;DR

Full answer below

Practice focus

Whether examples are chosen for a concrete behavioral need instead of added as a ritual.

Answer first

Answer path

Core answer first, then checkpoints toward a stronger answer.

Start
Core answer
Start here. This is the direct spoken answer to practice first.
Overview
Few-shot examples are useful when they teach a boundary the instruction alone does not communicate reliably.
Zero-shot prompting gives the task and constraints without examples. Few-shot prompting adds representative input-output examples. I start zero-shot when the task is simple and well specified, then add a small set of examples when they improve ambiguous classification, tone, edge-case handling, or output conventions in evaluation.
Checkpoint
Stronger answer
Optional upgrade with examples, trade-offs, and sharper phrasing.
The examples need to cover meaningful distinctions, not repeat the easiest case. For a ticket router, I might include a request that looks like billing but is actually account access, plus an explicit unknown case. I keep examples consistent with the schema and avoid leaking real customer data. More examples consume context and can bias the model toward their wording or class distribution.
Finish
Advanced answer
Optional deeper angle for production judgment and nuance.
I treat the example set as versioned behavior and test it for coverage, imbalance, stale policy, and unintended copying. If a prompt needs dozens of examples to remain stable, that is a signal to revisit the design. I may narrow the task or compare retrieval, fine-tuning, and a deterministic classifier against the same evaluation set. The decision is based on measured performance, not a belief that few-shot prompting is always more capable.

Review signals

Use these to catch weak phrasing, shallow reasoning, and risky claims.

Common mistakesCandidateCommon mistakesMistakes that weaken the answer or signal shallow understanding.Open details

Details

Adding examples without measuring whether they improve the task.
Using only ideal happy-path examples.
Including sensitive production inputs verbatim.
Assuming more examples always improve generalization.

Red flagsCandidateRed flagsSignals that an interviewer may treat as risky.Open details

Details

Examples silently redefine business policy.
The few-shot set is not versioned with the prompt.
The answer ignores token cost and class bias.

Follow-up prompts

Quick interviewer prompts inside this drill.

Follow-up questionsCandidateFollow-up questionsLikely next questions after the main answer.Open details

Details

Treat these as the next interviewer turn. Answer one out loud, then save the gap in your practice notes.

How would you select representative examples?
What would make you remove an example?
When would you choose a different approach entirely?

Continue the section

Next in Prompting & Structured Output: Decompose a model task

Step 6 of 20

When should you split one large prompt into multiple steps?

When would you use zero-shot versus few-shot prompting?

Practice focus

Answer path

Overview

Review signals

Follow-up prompts

Continue the section

When would you use zero-shot versus few-shot prompting?

Practice focus

Answer path

Overview

Review signals

Follow-up prompts

Continue the section