Prompt Engineering
A prompt is everything the model reads before it answers: the system prompt, your request, any examples, and any documents you paste in. Prompt engineering is the work of shaping that text on purpose and then checking the results, the way you'd test any other part of a process. The Prompt Report, a 2024 survey of the field, catalogued 58 text prompting techniques, including zero-shot and few-shot prompting, chain-of-thought and decomposition.
Today most of the gains come from clear specification more than clever phrasing. Current models follow plain instructions well, so state the goal, the audience, the constraints and the output format, and give the model the material it needs. Choosing and arranging that material has its own name, context engineering, and Qurtoo's article context engineering is the new prompt engineering covers how the two fit together. Prompt engineering is still worth learning, because every chatbot, agent and automation starts with someone writing down what good output looks like and then measuring whether they got it.
Example
Role: You are an editor for a small accounting firm's client newsletter.
Goal: Rewrite the draft below so a business owner with no finance training understands it.
Constraints:
- Keep it under 200 words.
- Keep every number exactly as written.
- Don't add advice the draft doesn't give.
Format: A headline, then two short paragraphs.
<draft>
{paste the draft here}
</draft>When to use it
- You'll run the same kind of task more than once and want consistent results
- Output quality varies from one run to the next
- You're writing a prompt other people or an automation will reuse
- A model keeps missing a constraint you thought was obvious
When not to use it
- A one-off question where a plain request already works
- The model lacks the facts it needs; fix the context or add retrieval before you rewrite the wording
- Tweaking wording without a test set to tell you whether the change helped
Articles that use this
- XML Tags Are the New Markdown: Why Closed Delimiters Beat Free-Form PromptsTechniques
- Chain-of-Thought Is Sabotaging Your Model's Sense of Direction: Prompt Grids as Symbols, Not SentencesTechniques
- Marc Andreessen Told His AI to 'Never Hallucinate.' Here's the Four-Line Scaffold That Actually WorksTechniques
- Chain-of-Thought Assumes Left-to-Right. Mercury 2 Doesn't: How to Prompt a Diffusion LLM by Handing It a Skeleton, Not a TrailTechniques
- Delete Your Temperature Parameter: Why Self-Consistency Sampling Is Dead on Reasoning Models Like Gemini 3.5 FlashTechniques

