Prompting

when you say the prompt you mean the parts your application controls, which is most of what actually got sent.

Prompt

it turned out to be writing a clear specification and testing it against an eval set, not finding a magic phrase.

Prompt engineering

you ask with instructions only, and add examples only after measuring that instructions alone were not enough.

Zero-shot

examples teach format and edge-case handling far more reliably than prose descriptions of the format.

Few-shot / in-context learning

you open with you are a senior database engineer, and the vocabulary and defaults shift, but it does not know anything new.

Role prompting

you finally spelled out the fields, the ordering and what to do when a value is missing, and the model stopped being unreliable.

Output format specification

you wrap the document in tags so the model stops reading its contents as instructions addressed to it.

Delimiters and XML tags

do not add commentary kept producing commentary, and respond only with JSON worked first time.

Negative instructions

you tell the model to work through it step by step, and the answer gets better on anything with arithmetic or several constraints at once.

Chain of thought

you add answer only from the provided documents, and say so if they do not cover it, and the invented details stop.

Grounding via context

a line buried in a retrieved document told the model to ignore your system prompt, and it partly listened.

Instruction hierarchy

you put the important instruction halfway down a long prompt and the model behaved as if it were not there.

Lost in the middle

the prompt lives in version control with named placeholders, not concatenated inline at three different call sites.

Prompt template

you change a word, re-run the eval set, look at what broke, and change it back, instead of reading one good output and declaring victory.

Prompt iteration

you have the model rewrite your prompt, then test the rewrite exactly like any other change rather than trusting it.

Meta-prompting