# How to actually ask: prompting without the magic words
> There are no secret incantations. Good prompting is just clear instructions to a brilliant, literal-minded assistant with no memory. A practical guide that follows straight from how the model works.
Source: https://rfriedmann.de/blog/how-to-prompt/
Published: 2026-06-23 · Track: learn · Level: Beginner
There's a small industry of "prompt secrets", magic phrases, ten-word tricks, lists
of incantations that supposedly unlock the model. Almost all of it is noise. Good
prompting isn't a spell-book; it follows directly from what the model actually is, and
once you've got the mental model, you can work out a good prompt yourself instead of
copying someone's ritual.
The whole thing rests on one picture, and everything practical falls out of it.
## The one mental model
Picture a brilliant, widely-read, eager assistant who is also completely literal, has
no idea what's in your head, and forgets you the moment the conversation ends. That's
the thing you're talking to. It [predicts the most plausible
continuation](/blog/what-is-an-llm/) of whatever you give it, working only from the
words in front of it, with [no memory of anything outside this
conversation](/blog/the-context-window/).
So prompting is not casting a spell. It's **briefing a capable stranger who will take
you at your word.** Vague brief, vague work. Clear brief, good work. Everything below
is just that idea applied.
## What actually moves the needle
In rough order of how much they help:
- **Say what you actually want.** The single biggest lever, and the one people skip.
"Make this better" gives the model nothing to aim at. "Rewrite this email to be two
sentences shorter and more direct, keep the friendly tone" gives it a target. The
model can't read the standard in your head; put it in words.
- **Give it the context it can't have.** It doesn't know your audience, your
constraints, your situation, unless you say so. "Explain this to a non-technical
client" produces a completely different, better answer than the same question bare.
You're filling a window it would otherwise fill with guesses.
- **Show one example.** If you want a particular format or style, show a small sample
of it. "Like this: [example]" is worth a paragraph of description, because matching a
pattern is exactly what the model is built to do. One good example often beats a long
list of rules.
- **Give it a role, when it sharpens the task.** "You're a copy-editor; flag anything
unclear" nudges the model toward the right register and priorities. Useful as
framing, not as a magic costume, it sets the tone, it doesn't grant new powers.
- **Ask for steps on hard problems.** For anything involving reasoning, maths, logic,
multi-step analysis, "work through it step by step" genuinely helps, because the
model reasons *in* the text it writes. Room to think on the page leads to better
answers than demanding the conclusion cold. (Newer "reasoning" models do this
internally, so it matters less with them, but it rarely hurts.)
- **Iterate; don't expect one shot.** The first answer is a draft. "Good, but make it
shorter and cut the jargon" is normal and effective. A conversation beats trying to
cram the perfect instruction into a single opening message.