Prompt engineering is the deliberate practice of designing, testing and revising the wording given to a generative AI tool so its output is more accurate, more consistent, or better suited to a specific task — treating the prompt itself as something worth getting right, rather than typing the first version that comes to mind.
Canadian legal regulators already treat the activity as distinct, billable work rather than something incidental to using a tool. The Law Society of Alberta’s own survey of guidance across Canada quotes the Law Society of Manitoba on exactly this point: “it is appropriate to charge for the time spent in crafting and refining AI inputs and prompts and in reviewing, confirming, analyzing and editing generative AI tool output.” The distinction that guidance is drawing is between the (billable) work of getting the prompt right and reviewing what came back, and the (non-billable) time the tool merely saved by being fast.
Getting a prompt’s wording right is not only about better answers, either. Canada’s Cyber Centre lists it as a security control in its own top-ten list of AI security actions, instructing organizations to “Implement prompt injection and jailbreak mitigations” and specifically to isolate system prompts and protect prompt history from anything a user or an external document supplies. The Centre’s own case file for why this matters is a 2025 incident in which attackers used a crafted prompt to get GitHub Copilot to run commands nobody had authorized; the specific opening was closed once it was found. Anyone tuning a business tool’s prompt is doing both jobs at once — writing for a better answer today, and making the tool harder to manipulate.
A brokerage tests three different wordings of the same request to a drafting tool before settling on a version of the prompt that reliably includes a required disclosure line every time, rather than most of the time. That work — comparing outputs, keeping the version that behaves consistently, and still reading every draft before it goes out — is prompt engineering. It is not a substitute for a person checking the result; the Law Society guidance above is explicit that reviewing the output is separately billable work of its own, not something the improved prompt makes optional.
See also prompt and system prompt.
This is one term in a plain-English glossary on how AI actually works and where it fits in a Canadian business.
