---
language: en
title: "Prompt engineering"
description: "Prompt engineering formulates and tests instructions for a task. Learn how to use success criteria and examples without mistaking improvement for a guarantee."
canonical: "https://tropk.ai/en/resources/glossary/prompt-engineering"
last_modified: "2026-10-12T16:12:48.856Z"
---

# Prompt engineering

Reviewed on 2026-10-06.

## What it means

[Prompt](/en/resources/glossary/prompt) engineering is the work of formulating, testing, and revising instructions so that a model performs a task according to defined criteria. It can involve context, examples, the expected format, and rules for using supplied materials. The focus is the input sent to the system and its evaluation, not a formula capable of controlling every response in every situation.

## Why it matters for marketing

For marketing, this practice helps make requests for analysis and content creation clearer and more verifiable. Asking for a short response may address length, but it does not automatically resolve missing information. The person commissioning the work needs to distinguish between instruction, source, and system capability issues, evaluating whether a change actually improves the task and the review cost.

## Example in practice

Imagine a team using AI to classify comments about a campaign. The initial instruction accepts only positive or negative; comments that praise the product and criticize the delivery receive inconsistent labels. The analyst adds a mixed category and examples of this case. She compares the versions using comments set aside for testing before adopting the new instruction.

## How to apply or verify

Start with success criteria and an identifiable version of the prompt. Use representative cases, reference results where appropriate, and difficult examples. Record the model, materials, and execution conditions to compare versions. Anthropic recommends establishing criteria and tests before improving instructions. The person responsible for the task should review relevant errors, in addition to looking at the aggregate score.

## Important distinctions

Prompt engineering modifies inputs; [fine-tuning](/en/resources/glossary/fine-tuning) modifies the model through additional training. Providing examples within the instruction is not the same as training its parameters. A change that works for one model or task may fail in another context. Latency, cost, missing information, and incorrect facts may require changes to the model, [retrieval](/en/resources/glossary/information-retrieval), or process rather than a longer instruction.

## Sources and references

- [Prompt engineering overview](<https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/overview>) — Anthropic
- [Define success criteria and build evaluations](<https://platform.claude.com/docs/en/test-and-evaluate/develop-tests>) — Anthropic
- [Machine Learning Glossary: fine-tuning](<https://developers.google.com/machine-learning/glossary#fine-tuning>) — Google for Developers

## Continue a leitura

- [Prompt set](</en/resources/glossary/prompt-set>)
- [Prompt monitoring](</en/resources/glossary/prompt-monitoring>)
- [Generative prompt](</en/resources/glossary/generative-prompt>)

## About the author

[Vinicius Pacheco](</en/team/vinicius-pacheco>)

Responsible for Tropk’s technical content, focusing on crawling, indexing, structured data, and agent access to content.

---

[All pages in Markdown](<https://tropk.ai/sitemap.md>)