A prompt is the instruction, question or piece of text a person — or another system — gives a generative AI tool to get it to produce something. Canada’s Federal Court uses the word in exactly this sense in its own notice on AI in court proceedings, describing the AI it regulates as a system “usually based on prompts or information provided to the system.”
The same Federal Court notice ties the word to what can go wrong with the output. Its footnote defining a hallucination describes fabricated content as content “fabricated by AI in response to a prompt or request.” Canada’s Cyber Centre uses the word the same plain way, describing generative AI as “a technology that uses machine learning to construct responses based on a prompt or query”, and noting that large language models are “provided a set of parameters (for example, a query or prompt)” to produce anything at all. No prompt, no output — the model does not act on its own between requests.
A single typed prompt is not the whole of what shapes a reply, though. See system prompt for the separate, standing instruction an operator sets up in advance, and prompt engineering for the practice of deliberately improving a prompt’s wording rather than typing the first version that comes to mind.
Two people ask the same drafting tool for a “short cover letter for this file.” One gets three vague sentences; the other, who added the file type, the recipient and the required disclosure line to their own request, gets something usable in one pass. Both typed a prompt in the sense used above; only one of them treated the wording of that prompt as something worth getting right. That gap between the two is the entire subject of prompt engineering.
See also system prompt, prompt engineering and large language model.
This is one term in a plain-English glossary on how AI actually works and where it fits in a Canadian business.
