Definition
In the sense used on this site, a prompt is the input a person gives an AI language model in order to get a piece of work back. It usually combines an instruction (what to do), material (what to work from) and constraints (how the output should look and what it must avoid). It can be a single line typed into a chat window or a long, structured template with named inputs that a team fills in each time.
The word has other meanings in marketing, such as a request to rate an app, a permission request on a phone or a push notification, and the verb is used loosely for nudging people to act. Those are not what this entry describes. Here, a prompt is always an instruction to a model.
Why a prompt works like a brief
A language model writes the most plausible continuation of what it receives. Given a generic prompt, the plausible continuation is generic copy. Given a prompt that names the audience, states one task, supplies the facts, sets the constraints and format and describes what good looks like, the plausible continuation is much closer to usable work. The guide to prompt patterns for marketing lists those elements and the handful of structures that cover most tasks, and treats better output as a matter of a fuller brief rather than cleverer wording.
Seeing a prompt as a brief has a practical consequence. The material an approved brief already holds, such as the audience, the proof and the voice, can often be pasted straight into the prompt, which keeps AI work tied to the same decisions as the rest of the team's output. And a prompt that tells the model what to do when the material does not cover something turns an invisible invention into a visible gap.
In practice: libraries, testing and data rules
Three habits turn individual prompts into something a team can rely on.
- A shared library. Good prompts tend to live in personal chat histories, so results depend on who is asking. The tutorial on building a team prompt library replaces that with templates for repeated tasks, each with named inputs, shared context blocks, a quality bar, test cases, a version history and an owner.
- Testing before reliance. One impressive answer shows a prompt can work once. The guide to testing prompts and models runs a prompt against a set of real cases with a rubric written in advance, and repeats the test whenever the prompt, model or tool changes.
- Data rules. What goes into a prompt goes to the tool that processes it. The tutorial on writing a marketing AI use policy settles what data may go into a prompt: public and non-personal data into approved tools, personal data only into tools with suitable contractual terms and only when the task needs it, and special-category data not at all.
Review keeps the prompt in the record after publication too: the AI-assisted content review checklist asks for the prompt and the tool used, so a reviewer knows what the model was given and what it could not have known.
A wooden toy maker fixes its product descriptions
Quappithe Wooden Toys used an AI tool to draft descriptions for 40 new products, with a one-line prompt asking for "a fun, engaging product description". The drafts all opened the same way and 6 of them mentioned age ranges the business had never stated. The marketing assistant rewrote the prompt in four parts: the audience (parents and grandparents buying for children aged 1 to 4), the material (each product's dimensions, wood type, finish and the safety standard it had been tested to, pasted from the product sheet), 2 approved descriptions as examples of the voice, and an instruction to write "not in source" rather than guess. On a test set of 10 products, 9 drafts needed only light edits and none invented an age range. The prompt went into the team library with a version number and the 10 test products.
Related terms
A Language ModelAn AI system trained on large amounts of text that writes by predicting which piece of text is likely to come next. is the system that receives the prompt and writes by predicting what text is likely to come next, which is why gaps in the prompt are filled with typical text. GroundingMaking a language model answer from specified material, such as documents retrieved for the question, rather than from what it absorbed in training. makes a model answer from specified material, such as retrieved documents, and in practice that material arrives inside the prompt. Fine-TuningTraining an existing language model further on a set of examples so that a style, format or task pattern is built into the model itself. builds a style or format into the model itself through further training; teams usually try better prompts and examples first. A Creative BriefA short document telling the people making creative work what it must achieve, for whom, with which message and within which limits. sets out the audience, message and proof for campaign work, and is often the best source for the context a prompt needs.
Common confusions
- A prompt is a clever phrase. Wording tricks matter far less than the content of the brief: audience, task, material, constraints, format and the quality bar.
- A good prompt works forever. Models and tools change, so a prompt in regular use is retested when they do.
- Anything can go into a prompt. The AI use policy decides what data may be included; personal data goes in only when the task needs it and the tool's terms allow it, and only the minimum the task requires.
- Every prompt in marketing is an AI prompt. Rating prompts, permission prompts and push prompts are different things; this entry covers only instructions to an AI model.
- Prompting replaces review. A better prompt reduces errors; it does not remove them, so AI drafts are still checked against sources before publication.
