Skip to contentScholarGate
LibraryBookshelfDeskReview StudioAssistant
Sign in
On this page
IntuitionHow it worksWhen to use itStrengths & limitationsCommon pitfallsApplicationsFrequently asked🔒 Read the full methodSourcesRelated methods
Cite this pageSpotted an issue on this page? Report or suggest a fix →
Home›Text mining›Prompt Engineering — Instruction Design for Large Language Models
Process / pipeline

Prompt Engineering — Instruction Design for Large Language Models

Prompt Engineering (Instruction Design for Large Language Models) · Also known as: instruction design, LLM prompting, Yönerge Mühendisliği (Prompt Engineering)

Prompt engineering is the practice of crafting structured natural-language instructions — prompts — to elicit targeted outputs from large language models (LLMs). Formalised by Brown et al. (2020) in the context of GPT-3 and extended by Wei et al. (2022) with chain-of-thought prompting, it encompasses four main strategies: zero-shot, few-shot, chain-of-thought, and tree-of-thought. Rather than re-training a model, the analyst shapes the model's behaviour entirely through the design of the input text.

ScholarGate
  1. Process / pipeline
  2. v1
  3. 2 Sources
  4. PUBLISHED
Cite this page →
Tools & resources
Download slides
Learn & explore

Read the full method

Members only

Sign in with a free account to read this section.

Sign in

Method map

The neighbourhood of related methods — select a node to explore.

Prompt Engineering
Few-Shot Text Classifica…GPT Fine-TuningLoRA and PEFTNatural Language Generat…Retrieval-Augmented Gene…Text ClassificationZero-Shot Classification

When to use it

Prompt engineering is applicable whenever you have access to a large language model API and a clearly defined text task — classification, extraction, summarisation, generation, reasoning, or structured output production. It is particularly valuable when labelled training data are scarce, when fine-tuning is too costly, or when rapid prototyping is required. The method assumes LLM API access is available and the target task is unambiguous. It is not appropriate when no LLM API is accessible, when the task requires exact numerical computation rather than language understanding, or when domain-specific accuracy requirements exceed what prompting alone can achieve.

Strengths & limitations

Strengths
  • Requires no model training or labelled dataset — a well-designed prompt can deliver strong results from the first query.
  • Highly flexible: the same underlying model can be directed to classify, summarise, extract, translate, or reason simply by changing the prompt.
  • Fast to iterate — changes to prompt design are immediately testable without any training cycle.
  • Chain-of-thought and tree-of-thought strategies substantially improve performance on multi-step reasoning tasks that naive prompting handles poorly.
Limitations
  • Output quality depends heavily on the model's pre-training knowledge; tasks requiring domain-specific or recent knowledge not in the model may fail.
  • LLMs can hallucinate — generate plausible-sounding but factually incorrect outputs — especially when prompted on topics outside their training distribution.
  • Prompt sensitivity: small changes in wording can lead to substantially different outputs, making robust evaluation over multiple prompt variants necessary.
  • API cost and latency scale with the number of tokens; large few-shot prompts or complex chain-of-thought traces can be expensive at scale.

Frequently asked

What is the difference between zero-shot, few-shot, and chain-of-thought prompting?

Zero-shot prompting states the task and asks the model to respond with no examples. Few-shot prompting prepends two to eight worked examples to show the model the desired format and reasoning pattern. Chain-of-thought prompting adds an explicit instruction to reason step by step before giving the final answer, which markedly improves performance on tasks that require multi-step reasoning.

Do I need labelled data to use prompt engineering?

Not necessarily. Zero-shot prompting requires no labelled data at all. Few-shot prompting needs only a small number of hand-crafted or carefully selected examples — typically two to eight — which are embedded directly in the prompt. Evaluation against a labelled reference set is, however, recommended before deploying any prompt in a production or research context.

How do I know if my prompt is good enough?

Test the prompt on a representative sample of your data and compute the metrics appropriate to your task — accuracy and F1 for classification, ROUGE or BERTScore for summarisation, human quality ratings for open-ended generation. Then inspect failures and revise the prompt iteratively. A prompt should be validated on at least a few dozen examples before being trusted for research or operational use.

When should I switch from prompt engineering to fine-tuning?

When prompting consistently fails to meet accuracy requirements on a well-defined task, when the task requires knowledge or formatting conventions not captured by prompting alone, or when inference cost at scale makes large prompts impractical, fine-tuning or parameter-efficient methods such as LoRA become the preferred path.

Sources

  1. Brown, T. et al. (2020). Language Models are Few-Shot Learners. Advances in Neural Information Processing Systems (NeurIPS), 33, 1877-1901. link ↗
  2. Wei, J. et al. (2022). Chain-of-Thought Prompting Elicits Reasoning in Large Language Models. Advances in Neural Information Processing Systems (NeurIPS), 35. link ↗

How to cite this page

ScholarGate. (2026, June 1). Prompt Engineering (Instruction Design for Large Language Models). ScholarGate. https://scholargate.app/en/text-mining/prompt-engineering

Related methods

Few-Shot Text ClassificationGPT Fine-TuningLoRA and PEFTNatural Language GenerationRetrieval-Augmented GenerationText ClassificationZero-Shot Classification

Which method?

Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.

  • Few-Shot Text ClassificationText mining↔ compare
  • GPT Fine-TuningDeep learning↔ compare
  • LoRA and PEFTDeep learning↔ compare
  • Natural Language GenerationText mining↔ compare
  • Retrieval-Augmented GenerationText mining↔ compare
  • Text ClassificationText mining↔ compare
  • Zero-Shot ClassificationText mining↔ compare
Compare side by side →

Similar methods

GPT Fine-TuningFew-Shot Text ClassificationZero-Shot ClassificationFine-Tuned TransformerLoRA and PEFTFine-Tuned Question AnsweringNatural Language GenerationBERT Fine-Tuning

Related reference concepts

Natural Language ProcessingPromptingNatural Language Processing in Clinical DocumentationInformation ExtractionLanguage ModelingQuestion Answering and Dialogue Systems

Spotted an issue on this page? Report or suggest a fix →

ScholarGate — Prompt Engineering (Prompt Engineering (Instruction Design for Large Language Models)). Retrieved 2026-07-21 from https://scholargate.app/en/text-mining/prompt-engineering · Dataset: https://doi.org/10.5281/zenodo.20539026
Quick facts
Originator
Tom Brown et al. (GPT-3 / few-shot framing, 2020); chain-of-thought extended by Jason Wei et al. (2022)
Year
2020 (few-shot prompting); 2022 (chain-of-thought)
Type
NLP pipeline — structured instruction design for large language models
Strategies
Zero-shot / few-shot / chain-of-thought (CoT) / tree-of-thought (ToT)
Input
Natural-language text prompt sent to an LLM API
Output
Model-generated text (classification label, reasoning trace, summary, structured data, etc.)
MinimumSamples
5
Difficulty
2
RequiresNormality
No
Related methods
Few-Shot Text ClassificationGPT Fine-TuningLoRA and PEFTNatural Language GenerationRetrieval-Augmented GenerationText ClassificationZero-Shot Classification
ScholarGate

A content-first reference library for research methods — what each one is, how it works, and where it comes from.

Open data (CC-BY)

Explore

  • Library
  • Search the library…
  • Browse by field
  • Fields
  • Journey
  • Compare
  • Which method?

Reference

  • Subjects
  • Atlas
  • Glossary
  • Methodology
  • Philosophy

Your tools

  • Bookshelf
  • Desk
  • Chat

Company

  • About
  • Pricing
  • Contact
  • Suggest a method

Entries are compiled from published sources for reference. Verifying the accuracy and suitability of any information for your own use remains your responsibility.

© 2026 ScholarGate · A research-method reference library
  • Privacy
  • Cookies
  • Terms
  • Delete account