Prompt Engineering for Beginners: 8 Techniques
9 min read
Prompt engineering sounds academic, but the useful part is a small set of repeatable moves. Here are the eight that pay off most often.
Prompt engineering is not a secret language. It is the practice of stating your intent clearly enough that a language model can act on it — and then learning the handful of patterns that consistently push outputs from "fine" to "genuinely useful". You can learn the fundamentals in an afternoon. These eight techniques cover the vast majority of what working prompters actually reach for.
1. Role prompting
Assign the model a persona at the top: "You are an experienced pediatric nurse." This anchors tone, vocabulary, and the standards the model holds itself to. The more specific the role, the sharper the effect.
2. Few-shot examples
Show two or three examples of the input and the exact kind of output you want, then give it the real input. Models are pattern-matchers; a couple of examples teach a format or voice better than any amount of description. This is the highest-leverage technique for consistency.
3. Chain-of-thought ("think step by step")
For anything involving reasoning, maths, or multi-step logic, ask the model to work through it step by step before giving the final answer. Reasoning in the open dramatically reduces careless mistakes. If you only want the conclusion, ask it to reason first and then summarise the answer at the end.
4. Delimiters
When your prompt contains distinct chunks — instructions, source text, examples — separate them with clear markers like triple backticks, XML-style tags, or headers such as "### SOURCE TEXT ###". This stops the model from confusing your instructions with the material it is supposed to act on, a surprisingly common failure.
5. Output format specification
Ask for a table, JSON, bullet list, or a fixed template. Explicit structure makes the output usable immediately and forces the model to organise its reasoning. If you need machine-readable output, describe the exact schema and add "return only valid JSON, no prose".
6. Decomposition
Big, fuzzy tasks produce big, fuzzy answers. Break the work into steps and either run them as separate prompts or ask the model to complete them in order. "First outline, then draft section by section" beats "write the whole thing" almost every time.
7. Negative constraints
Tell the model what not to do: "do not invent citations", "avoid marketing clichés", "do not assume a budget". Models default to the average of everything they have read; naming the failure modes you want to avoid is how you steer away from generic output.
8. Iterative refinement
Treat the first answer as a starting point. Give targeted feedback — "cut the intro", "the third example is wrong, here is why", "make it 30% shorter" — instead of rewriting the prompt from scratch. Two or three rounds of correction usually outperforms one perfectly-crafted prompt.
How to actually get good
Reading about techniques only gets you so far. The fastest way to improve is to study prompts that worked — the exact wording, alongside the real output it produced — and then adapt them to your own problems. That is the whole idea behind pairing every prompt with its actual result: you can judge the technique by the outcome before you borrow it.
Study real prompts
See these techniques in prompts that actually worked — each one sealed with the real AI output it produced:
Ready to add your own? Claim your first pixel free and seal your best prompt on the wall.