Examples
Worked examples
- Is an instance
Specifying 'You are a careful systematic reviewer. List inclusion criteria first, then...' to structure an LLM's literature-triage output, and archiving that exact prompt text alongside the resulting output
- Is an instance
Iteratively refining a prompt across several attempts to get a consistent output format, and reporting only the final, working version used to generate the reported result
Counter-examples
Looks similar, but isn't
- Not an instance
Typing a single unstructured word or short query into a chatbot is using an LLM, but is not prompt engineering in the disciplined, methods-relevant sense this entry describes
- Not an instance
The hidden instruction layer a platform configures before any user input is a system prompt, a distinct concept the end user typically does not write or see
Editorial commentary
For research administration purposes, prompt engineering matters less as a general AI-use-case craft and much more as a specific reproducibility gap. Because an LLM’s output can vary meaningfully with prompt wording, few-shot examples, model version, and sampling settings, two researchers — or the same researcher at two different times — can get materially different results from what looks like ‘the same’ AI-assisted step, unless the exact prompt and context are recorded.
The reproducibility angle
Current disclosure guidance from ICMJE and JAMA Network treats AI used as part of a study’s actual methods (not just manuscript writing) as requiring Methods-section reporting sufficient to understand and, where possible, repeat what was done — which in practice means recording the specific prompts used, not just the name of the tool. Nature Portfolio’s Reporting Summary template likewise expects methodology disclosure detailed enough for another researcher to assess and attempt replication. Archiving the exact prompt text alongside any reported AI-assisted output is the minimum practice that makes this kind of disclosure meaningful rather than nominal.
Authorship, not just reproducibility
Prompt engineering is itself a genuine human intellectual contribution, and can merit acknowledgement when it was central to producing a reported result — but per ICMJE and COPE‘s consistent position, it does not by itself confer authorship of AI-generated text, since the AI tool — not the prompt engineer — produced the actual words, and an AI tool cannot be an author regardless of how carefully it was prompted.
How this differs from system prompt
Prompt engineering is the researcher’s or user’s own iterative practice of crafting their input (the ‘user prompt’ layer) to a system they are actively querying. A system prompt is a separate, usually hidden, instruction layer set by the platform or product developer before any user input arrives, configuring the model’s behaviour for an entire session or deployment — something the end user typically neither writes nor sees, and which complicates reproducibility in its own, distinct way (see that entry).
Also known as
Prompt design · Prompting
Machine-readable encodings
Use in your systems
<role vocab="credit"
vocab-identifier="https://casrai.org/dictionary/"
vocab-term="Prompt engineering"
vocab-term-identifier="https://casrai.org/dictionary/term/prompt-engineering" />{
"@context": "https://schema.org",
"@type": "DefinedTerm",
"@id": "https://casrai.org/dictionary/term/prompt-engineering",
"name": "Prompt engineering",
"identifier": "https://casrai.org/dictionary/term/prompt-engineering",
"description": "The practice of designing, refining, and structuring the input text (prompt) given to a generative AI system to elicit a specific desired output -- including techniques such as role assignment, few-shot examples, chain-of-thought scaffolding, and output-format specification -- considered here specifically for its reproducibility implications in research work, as distinct from system prompt, the separate, developer-set instruction layer a deploying application configures independently of the end user.",
"inDefinedTermSet": "https://casrai.org/dictionary/domain/genai-disclosure#set",
"url": "https://casrai.org/dictionary/term/prompt-engineering",
"sameAs": [
"Prompt design",
"Prompting"
],
"license": "https://creativecommons.org/licenses/by/4.0/",
"publisher": {
"@id": "https://casrai.org/#organization"
},
"author": {
"@id": "https://casrai.org/#editorial-team"
},
"datePublished": "2026-05-21T01:54:44",
"dateModified": "2026-08-22T14:52:11",
"inLanguage": "en-GB",
"isAccessibleForFree": true
}
