AI Prompting Basics: The Patterns That Don't Expire

Most AI answers are mediocre. Not because the model is bad — because the prompt is.

I see it constantly, with beginners and senior developers alike: you type half a question, get half an answer, and conclude the AI is useless. In reality, you simply never told the model what you actually wanted.

The good news: learning AI prompting does not mean memorizing a hundred tricks. Trick lists expire with every model generation. What lasts is a handful of basic patterns — and they have more to do with clear thinking than with technology.

Why good prompts aren't secret knowledge

A language model doesn't guess out of malice. It completes patterns — and everything you don't tell it, it has to assume. When context is missing, the model fills the gaps with whatever is most probable. The result isn't wrong, it's generic: the average answer to an average question.

So the leverage isn't in the model; it's in what you feed it. A good prompt is a small specification: who's asking, what it's about, what material exists, what the result should look like. Sounds trivial. It's the entire difference.

The four patterns that don't age

1. Context before task

Tell the model who you are and what the result is for before you state the task. "Write an email to my landlord" is a different task once the model knows you've been waiting three months for a repair and have already asked twice, politely.

2. Specify format and boundaries

Say what the output should look like — length, structure, tone — and just as importantly, what you don't want. "Five sentences max, factual, no filler phrases" saves you three rounds of corrections. The model can respect boundaries; it just can't guess them.

3. Show examples instead of describing

If you want a particular style or format, include a sample. One good example beats three paragraphs of description. Models are pattern completers — hand them the pattern.

4. Force clarifying questions

The most underrated move: "Ask me three clarifying questions before you answer." It flips the game — instead of the model filling your gaps with assumptions, you fill them with facts. For anything more complex than a search query, it's almost always worth it.

Before / after

Here's what that looks like in practice:

Before: "Summarize this text."

After: "I'm a team lead and need to pass these meeting notes to colleagues who weren't there. Summarize in five bullet points max: decisions and open action items with owners only, no play-by-play. If anything is ambiguous, ask."

Same text, same model — but the second version produces something you can forward as-is.

For developers: prompting is the gateway drug to specification

If you write code, the pattern will look familiar: context, requirements, boundaries, acceptance criteria — that's a specification in miniature. Exactly the skill that appreciates in value as code gets cheap. If you can prompt precisely, you can usually spec precisely. The practice pays twice.

What you can do this week

Take one prompt you use regularly — the email summary, the draft text, the research question. Rebuild it deliberately along the four patterns: context, format, example, clarifying questions. Compare the answers. The difference is usually stark enough that you won't go back.

Want to go deeper? The ten most useful prompt patterns are available as a free cheatsheet — and the full step-by-step practice guide, no jargon, is in my audiobook "KI-Prompting: Praxiswissen für Anfänger" (German) — on Spotify and Apple Books.