prompt engineering

basicwidespread

Prompt engineering is the art of phrasing your request to a language model so that you get exactly the answer you need.

In plain terms

A language model is like a very well-read, but completely new colleague: it knows about many things, but knows nothing about you, your company, or what the result is for. If you just tell it "write a letter," you get an average letter. If you explain who it is for, what the goal is, what tone and length, you get something usable.

Prompt engineering is the practice of this conscious guidance. The prompt itself is the text written to the model: request, background, examples, expectations. The word "engineering" indicates that this is not done by intuition but systematically: you try it, check the result, and improve it.

It is not programming because you write in everyday language, not code. And it is not a collection of magic spells: a good prompt resembles a clear work assignment more than a secret formula.

How it works

Define the goal: what you are asking for, who it is for, and how the result will be used.
You provide the context the model cannot know: background, data, constraints, desired tone, and format.
If needed, you show one or two examples of a good response. For more complex tasks, you can ask the model to think through the task step by step; many current models do this on their own.
You read the answer and see where it differed from what you expected; this most often shows what was missing from your request.
You refine the prompt and try again. At companies, proven prompts are saved, tested, and integrated into their own systems or agents.

An everyday example

You want to write a complaint letter to your service provider and first ask the chatbot only: "write a complaint letter". You receive a generic, meaningless text. When you describe what happened, when, what compensation you request, and that you want a firm but polite tone, the letter is almost ready to send.

Who is affected

everyoneFor anyone using a chatbot, whether the answer is generic or genuinely useful depends on how the request is phrased.

employeeMore and more office tasks now require the ability to quickly extract high-quality results from AI tools. → Work & HR

parent, student, teacherIn learning, what matters is whether you ask the model for explanation and practice or just a ready-made solution. → Education & Family

developerThe behavior of AI-based applications and agents is largely determined by the instructions written for them. → Code

A common misconception

MythPrompt engineering is the knowledge of secret magic words and tricky formulas that can "crack" the model.

In realityThe essence is clear, detailed communication: goal, context, expectations. With current models, tricks matter less, while precise situation descriptions matter more.

Limits and risks

No prompt guarantees a flawless answer: the model can still err or hallucinate facts, so verify important claims.
What works for one model may function differently for another or a new version, so prompts sometimes need to be retested.
When providing detailed background information, it is easy to input personal or corporate confidential data; consider what you share with an external service.
A prompt does not replace expertise: if you do not understand the topic, it is hard to judge whether the answer is good.

Similar, but not the same

context engineering – Prompt engineering focuses on how the request is worded, while context engineering focuses on which documents, data, conversation history and tools are put in front of the model as a whole.
fine-tuning – Fine-tuning changes the model itself through further training, while prompt engineering only changes the text given to it.
prompt injection – Prompt injection is an attack where someone feeds an instruction to the model (either typed directly or hidden in a webpage, email, or document) that diverts it from its original task. Prompt engineering, by contrast, is the intended, well-meaning writing of instructions.

Related concepts

Broader

generative AI, large language model, human–machine communication

Narrower

system prompt, few-shot prompting, chain of thought, role assignment

On the edge

prompt, context engineering, context window, fine-tuning, prompt injection

A brief history

The term spread alongside large language models, after GPT-3, released in 2020, showed that model behavior could be guided purely by the text written to it, with just a few examples. The general public encountered it after ChatGPT's launch at the end of 2022.

Where things stand

By the end of 2026, prompt writing has become a common skill taught in courses and workplace training. It is less often mentioned as an independent profession, as professional attention has shifted toward broader context engineering and instructing agents.

Status: 2026 November

In the news

Appears in 5 Agent Age stories; the curve shows weekly occurrence in the news magazine.

Nova Scotia classroom AI rules vary, prompting calls for a provincewide policyOct 19, 2026 · Education & Family
Workday: demand for basic AI skills fell 25%, while demand for AI-building skills rose 51%Oct 5, 2026 · Work & HR
AI detectors are pushing colleges back to handwritten learningSep 22, 2026 · Education & Family
Guidelines needed as mental health practices adopt AI, reports sayAug 31, 2026 · Health
Gravity raises $30.5 million to place ads inside AI chatbotsAug 6, 2026 · Marketing

Also known as

prompt engineering · prompt engineering · Prompting · prompt engineering · prompt writing · prompt design · prompting · prompt engineering

Sources

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.