Examples
Worked examples
- Is an instance
A literature-search platform's underlying chatbot being configured with an undisclosed system prompt that instructs it to prioritise certain source types or refuse certain query categories, regardless of what the researcher explicitly asks
- Is an instance
A survey-analysis tool's embedded AI assistant behaving differently from the same underlying model accessed directly, because the tool wraps it in its own system prompt
Counter-examples
Looks similar, but isn't
- Not an instance
Text a researcher deliberately types into a chatbot's input box is a user prompt, refined through prompt engineering, not a system prompt
- Not an instance
General model behaviour that persists across every deployment of that model regardless of platform reflects training and fine-tuning, not a system prompt specific to one product
Editorial commentary
A system prompt sits in front of every user interaction with a given AI product or platform, set once by whoever built or deployed that product, and it shapes every response the model gives within that deployment — often without the end user ever seeing its content, since many commercial products treat their system prompt as proprietary.
Why this matters for a research office
Two consequences follow directly from the system prompt being hidden. First, a reproducibility gap: if a researcher uses a third-party AI product — a chatbot embedded in a survey platform, a literature-discovery tool’s built-in assistant — the outputs are shaped jointly by the visible user prompt and the invisible system prompt, plus the underlying model version and its own sampling settings. Reporting only the user-visible prompt in a Methods section, as current AI-disclosure guidance generally asks for, is necessary but not sufficient for true reproducibility, because another researcher re-running ‘the same prompt’ against a different or updated version of the same product may get a different system prompt applied underneath it, or may run it against the tool after the vendor silently changed that hidden layer. Where feasible, note the product name, version, and access date, not just the prompt text, and flag that the underlying system prompt is not independently controlled.
The other consequence: where guardrails live
System prompts are also frequently where a deployment’s safety constraints, tone, and behavioural guardrails are implemented — meaning that evaluating an AI tool’s behaviour, including its bias characteristics, requires accounting for this hidden configuration layer, even when its exact content cannot be inspected. Two products built on the same underlying model can behave meaningfully differently, including in their bias and refusal patterns, purely because of differing system prompts.
How this differs from prompt engineering
Prompt engineering is the end user’s or researcher’s own, visible, iterative craft of writing their input to elicit a desired result. A system prompt is the platform developer’s separate, prior, usually invisible instruction layer that every user input is processed through — a researcher engineering their own prompt has no visibility into, and generally cannot override, the system prompt sitting in front of it.
Also known as
Developer prompt · Pre-prompt
Machine-readable encodings
Use in your systems
<role vocab="credit"
vocab-identifier="https://casrai.org/dictionary/"
vocab-term="System prompt"
vocab-term-identifier="https://casrai.org/dictionary/term/system-prompt" />{
"@context": "https://schema.org",
"@type": "DefinedTerm",
"@id": "https://casrai.org/dictionary/term/system-prompt",
"name": "System prompt",
"identifier": "https://casrai.org/dictionary/term/system-prompt",
"description": "A instruction set or context injected by an AI platform or application developer before any end-user input, configuring a model's persona, constraints, or behaviour for an entire session or deployment -- typically invisible to and not authored by the end user, as distinct from prompt engineering, the user's own iterative practice of crafting their own input.",
"inDefinedTermSet": "https://casrai.org/dictionary/domain/genai-disclosure#set",
"url": "https://casrai.org/dictionary/term/system-prompt",
"sameAs": [
"Developer prompt",
"Pre-prompt"
],
"license": "https://creativecommons.org/licenses/by/4.0/",
"publisher": {
"@id": "https://casrai.org/#organization"
},
"author": {
"@id": "https://casrai.org/#editorial-team"
},
"datePublished": "2026-05-21T01:54:51",
"dateModified": "2026-08-22T14:52:12",
"inLanguage": "en-GB",
"isAccessibleForFree": true
}
