Back to glossary

Prompt engineering

İstem mühendisliği

D3

Prompt engineering is the iterative design and refinement of instructions and context used to guide a model’s response for a task.

Review status: 2026-11-26

Technical explanation

It can specify a task, output format, examples, constraints, and supplied context, then test and revise those inputs against representative cases.

Conceptual boundaries

Prompt engineering is not fine-tuning because it changes inference-time input rather than training model parameters, and it is not a guarantee of performance.

Provider-neutral example

A team can compare two instructions for extracting invoice fields and retain the version that is clearer on a representative, reviewed sample.

Limitations

Prompt changes can overfit to examples, conflict with other instructions, or fail in edge and adversarial cases, so evaluation must include representative and adversarial cases.

Related concepts

Atomic claims and evidence

  1. 1.1Microsoft Learn, Prompt engineering techniques for Azure OpenAI
    Source
    Microsoft Learn, Prompt engineering techniques for Azure OpenAI
    Source role
    Authoritative source
    Exact locator
    Scenario-specific guidance; Few-shot learning; Best practices
    Supported claim
    Azure prompt engineering changes request-time prompt content, whereas Azure fine-tuning trains on more examples than fit in a prompt and adapts model weights to a task.
    Last verification
    Review due
    Scope limitation
    This compares Azure implementation approaches and does not establish performance for every prompt or fine-tuning job.
    1.2Microsoft Learn, Customize a model with fine-tuning
    Source
    Microsoft Learn, Customize a model with fine-tuning
    Source role
    Authoritative source
    Exact locator
    In this article: fine-tuning trains on more examples than fit in a prompt and adapts model weights to the task
    Supported claim
    Azure prompt engineering changes request-time prompt content, whereas Azure fine-tuning trains on more examples than fit in a prompt and adapts model weights to a task.
    Last verification
    Review due
    Scope limitation
    This compares Azure implementation approaches and does not establish performance for every prompt or fine-tuning job.
  2. 2.1Microsoft Learn, System message design for Azure OpenAI
    Source
    Microsoft Learn, System message design for Azure OpenAI
    Source role
    Authoritative source
    Exact locator
    Design checklist: Test, measure, and iterate; Common pitfalls; Limitations
    Supported claim
    Microsoft’s Azure OpenAI system-message guidance recommends realistic and adversarial testing because instructions can overfit to examples or fail in edge cases, and conflicting instructions can change behavior.
    Last verification
    Review due
    Scope limitation
    This is provider guidance for Azure OpenAI system messages and does not predict every other implementation’s behavior.